Geochimicaet CosmochimicaActa, Vol. 60, No. 14, pp. 2709-2720, 1996 Copyright© 1996 ElsevierScienceLtd Printed in the USA. All rights reserved 0016-7037/96 $15.00 + .00
Pergamon
P I I S0016-7037(96) 00117-2
Compressibility of molten high-Ti mare glass: Evidence for crystal-liquid density inversions in the lunar mantle S. CIRCONE and C. B. AGEE Department of Earth and Planetary Sciences, Harvard University, Cambridge, MA 02138, USA (Received August 15, 1995; accepted in revised form March 26, 1996)
A b s t r a c t - - S t a t i c compression density measurements of molten Apollo 14 black glass have been performed at liquidus conditions in the pressure range 1.0-11.5 GPa. A range of values for the BirchMuruaghan isothermal bulk modulus (K2173 K : 13-23 GPa) and the pressure derivative ( d K / d P = 7 3, respectively) fit the data within the uncertainties; K2173 K : 15.5 GPa and d K / d P = 5.2 provides the best fit to the four measured pressure-density pairs corrected to isothermal conditions. The ultramafic high-Ti mare glass is significantly more compressible than other ultrabasic and basic silicate liquids previously studied. The compressibility appears to be related to TiO2 content. The Apollo 14 black glass sample has the highest TiO2 content of the pristine mare glasses, which are thought to be primary melts from a differentiated and strongly heterogeneous lunar mantle. Our density measurements confirm the prediction that molten black glass will be denser than equilibrium olivine and orthopyroxene at pressures greater than ~2.0 and ~0.5 GPa, respectively. These density inversions correspond to depths greater than approximately 410 and 100 km in the lunar mantle. Since orthopyroxene replaces olivine at - 1 . 5 GPa (300 kin) on the liquidus, our results constrain the maximum depth of buoyancy for high-Ti magmas at <400 km in the lunar mantle. 1. INTRODUCTION
cases TiO2. The predicted conditions for density crossovers (Delano, 1990) are expected to occur at depths relevant to the origin of picritic lunar glasses. The pristine glasses sampled from the lunar surface have been classified based on a number of characteristics (Delano, 1986). The Mg#'s [ M g / ( M g + Fe)] of the pristine glasses are high (0.45-0.65), relative to those of crystalline lunar basalts (typically <0.50), suggesting that they are derived from primary melt fractions in the lunar mantle and have not undergone fractional crystallization prior to eruption. They exhibit a wide range of TiO2 contents (0.3-16.4 wt%), FeO contents are high ( 1 6 - 2 5 wt%), oxygen fugacities are near the iron-wtistite buffer, and only minor amounts of alkalis (totaling less than 0.6 wt%) are present. They are compositionally distinct from terrestrial magmas. Ultramafic komatiites have higher Mg#'s and < 1 wt% TiO2. Terrestrial hasalts have higher SiO2, A1203, CaO, and alkali contents. Models for the petrogenesis of the pristine lunar glasses must reconcile the wide range in TiO2 content with the fairly uniform, unfractionated Mg#'s. The generally accepted model for the petrogenesis of the suite of pristine lunar glasses involves the partial melting of heterogeneous (i.e., low- and high-Ti) source regions at depth (e.g., Ringwood and Kesson, 1976). It is postulated that these regions were created by sinking diapirs of dense FeO, TiO2-rich residual cumulates left over from the crystallization of a magma lunar ocean. Phase equilibria experiments place constraints on the depth from which the lunar picritic glasses were derived. Since olivine and orthopyroxene are thought to be the primary phases remaining after partial melting of the lunar mantle, the depth of origin is identified as the pressure at which multiple saturation of these phases on the liquidus of the picritic glass is observed. These experiments put the
Magma compressibility studies have shown that crystal-liquid density inversions may have been important in the early magmatic differentiation of the Earth (Agee and Walker, 1988a, 1993; Miller et al., 1991; Ohtani et al., 1995). Specifically, liquidus olivine is neutrally buoyant in ultramafic magmas such as komatiite and peridotite in the range 7 - 1 2 GPa, or an equivalent PREM (Dziewonski and Anderson, 1981) depth of about 215-360 km in the Earth's mantle. In fact, it has been argued that the geochemical signature of primordial olivine flotation is reflected in the nonchondritic Mg/Si values of upper mantle peridotites (Agee and Walker, 1988b). This differentiation process is relevant to a magma ocean setting, in which equilibrium crystals would coalesce due to buoyancy forces at some depth in the molten outer layer of the early Earth. The role of crystal-liquid density crossovers in the mobility of terrestrial magmas produced by partial melting of a solidified mantle is less clear. These magmas, with the possible exception of Archean komatiites, should be buoyant at the shallow depths at which the solidus is intersected. The Earth's Moon is smaller and attains pressures of only ~4.7 GPa at its center (--1738 kin). Although an early magma ocean has also been hypothesized for the Moon, the olivine/ultrabasic liquid density inversions at 7 - 1 2 GPa are not relevant. The shallow lunar pressure gradient (0.1 GPa/ 20 krn near the surface, ~ 1/6 that of the Earth's) suggests that density inversions in a partially melted lunar mantle would be even more unlikely. However, crystal/liquid density inversions may have played an important role in the petrogenesis of ultramafic lunar magmas because they are significantly denser than their past and present terrestrial analogues due to high concentrations of FeO and in some 2709
2710
S. Circone and C. B. Agee
depth of origin for the pristine lunar glasses at a range of depths, typically 4 0 0 - 5 4 0 km ( 2 - 2 . 5 GPa) (see Delano, 1986). Phase equilibrium experiments on the Apollo 14 black glass with 16.4 wt% TiO2 (Wagner and Grove, 1995) place the multisaturation point of olivine + orthopyroxene + chrome spinel at approximately 1.5 GPa, corresponding to a depth of 300 km in the lunar mantle. An alternative model (Hubbard and Minear, 1975; Wagner and Grove, 1993) suggests that a low-TiO2 picritic glass composition was generated at depth and assimilated varying amounts of the Fe,Ti-rich cumulates at shallow depths ( ~ 1 0 0 kin) to produce the observed range of glass compositions. This model requires that large-scale overturn and mixing of the lunar mantle did not occur prior to magma formation and emplacement, although Hess and Parmentier (1995) have argued that this scenario played an important role in mare volcanism. Another model invokes polybaric fractional melting from 4 - ~ 2 GPa (Longhi, 1992), followed by melt segregation at the depth of olivine + orthopyroxene multisaturation. However, based on calculations by Delano (1990), Longhi (1992) concludes that this model cannot produce high-TiO2 picritic glasses because they are negatively buoyant at these pressures. These calculations, based on 105 Pa ultrasonic (Rivers and Carmichael, 1987) and density (Lange and Carmichael, 1987) measurements on silicate liquids, suggest that the glass with the highest observed TiO2 content becomes negatively buoyant relative to equilibrium olivine at 2 - 2 . 5 GPa. Crystal/liquid density inversions for the less dense, low-TiO2 glasses would occur at much higher pressures ( ~ 4 GPa, or > 1000 km depth) and not impact on magma mobility. A variation on the polybaric melting model (Hess, 1991) involves the adiabatic rise of diapirs originating from a thermal boundary layer at depths in excess of 700 km. The ascent rate of the diapir must be sufficient to compensate the tendency of the TiO2-rich partial melt to segregate and remain in the mantle, again due to negative buoyancy forces. Thus, the density of these ultramafic liquids at lunar mantle pressures places important constraints on their petrogenesis. In the lunar system, crystal/liquid density crossovers may occur at the relatively modest pressures of the lunar mantle because the discrepancy between the dense Fe,Tirich liquid and residual mantle density is much smaller than that in terrestrial systems. We have measured the high pressure density of molten high-TiO2 lunar glass to determine the conditions for crystal/liquid density crossovers. Specifically, we have studied a synthetic analogue of the Apollo 14 black glass, which contains the highest (16.4 wt%) TiO2 content of all pristine glasses sampled. We have undertaken the following study to address three questions. First, what is the compressibility of a silicate liquid with high TiO2 content? None have been measured to date, although a 105 Pa ultrasonic study of simple titanosilicate liquids (Webb and Dingwell, 1994) suggests that compressibility is highly composition-dependent at low pressure. Second, under what conditions would liquid density crossovers with mantle residue phases occur? This has yet to be determined experimentally. And finally, what constraints does this information place on the petrogenesis of the ultramafic high-Ti lunar glasses?
2. EXPERIMENTAL METHOD 2.1. Experimental Procedure
The sinking and floating spheres technique (Agee and Walker, 1988a, 1993) was used to measure the high-pressure density of molten Apollo 14 black glass. In these experiments, the density of the liquid relative to spheres of known density is determined by the sinking (more dense) or floating (less dense) of spheres in the liquid under equilibrium static compression conditions. Experiments were performed in a Boyd and England (1960) piston-cylinder (up to 2.5 GPa) and Walker-style (Walker et al., 1990) multi-anvil apparatus (3.0-11.5 GPa). Experimental charges consist of two spheres (diameter range 200-550 #m) packed in a powdered starting material, which is a synthetic lunar basalt analogue with the composition of the Apollo 14 black glass (Table 1 ). The oxide mixture starting material was conditioned at 105 Pa, 1075°C for 24 h under Fe-FeO conditions. The crystalline spheres were abraded from fragments of single gem-quality crystals of olivine (OL1-Fo89.7Fa~o.3and OL2Fos4.3Fa15.7), garnet (GT1-PY69.vAllv.0GrI3.3,GT2-PY63.0AI28.yGr8.3, GT3-Py61.4A1359GrzT),and synthetic corundum (see Table 1 for analyzed compositions). Note: Gr~00spheres were also tried; however, they recrystallized into multiphase aggregates in experiments at 2.54.0 GPa, producing uninterpretable results. For both types of experiments, one sphere each was located at the top and bottom of a Mo capsule. These capsules provide several advantages: (1) they maintain a low fo~, (2) the large internal dimensions (2 mm high by 1.5 mm wide) minimize capsule wall drag effects on moving spheres, and (3) thick metal capsules act as a heat sink and their high thermal conductivity promotes a smoother temperature distribution in the multi-anvil experiments. The sinking and floating spheres technique requires only a short experiment duration (-<2 min) for the buoyancy force to move a sphere up or down in the liquid sample, minimizing the contamination of the liquid with Mo and the alteration of the sphere density through the formation of a reaction rim. Special care was taken to pack the Apollo 14 black glass starting material around the spheres so that they would not adhere to the capsule surface. Experiments were compressed at room temperature, then heated to temperature and held under constant pressure and temperature conditions for 120 s (piston-cylinder) or 40-90 s (multi-anvil). The length of time was sufficient to allow the charge to become molten and the spheres to move to the top or bottom of the capsule. A rapid quench was achieved in a few seconds by shutting off power to the heater. In successful sink/float experiments, the temperature must be sufficient to completely melt the charge in a short time duration, thus the appropriate P, T conditions slightly overshoot the liquidus. At lower temperatures, a solid/liquid mush forms that impedes sphere movement and the liquid does not correspond to the bulk composition of the starting material. If the temperature is too high, the spheres dissolve rapidly and alter the liquid composition. They often disappear completely. The temperature range that will yield interpretable results is typically within a few tens of degrees Kelvin of the liquidus. The phase equilibria data of Wagner and Grove ( 1995 ) provided the P, T conditions in the piston-cylinder experiments. In the multi-anvil experiments, the near liquidus conditions were determined by trial and error. In the piston-cylinder experiments, the pressure of the t/2" talc + Pyrex assembly was calibrated by DTA (differential thermal analysis) against the melting point of Au at 0.8, 1.4, 1.8, and 2.7 GPa (Akella and Kennedy, 1971). Experiments and calibrations were performed using the cold piston in technique. Temperature was measured using a W97R3/W75 Re2s thermocouple positioned at the capsule top ( 1.6 mm from charge center). Estimated uncertainties in experiment conditions are _+0.1 GPa and _+10°C. The multi-anvil experiments used an octahedral pressure medium (8 mm truncations, 12 mm octahedron) plus gasket (2.0 mm wide by 3.3 mm deep) assembly cast as one piece with Ceramacast 584. The Re foil heater (coiled two times) was positioned inside the 3.3 mm diameter drilled bore hole with a W97R3/W75Re25 thermocouple located on the outer surface of the heater and centered along axis. The capsule with the sample chamber centered in the hotspot of the heater was situated between two crushable alumina rods and insu-
Compressibility of lunar glass
2711
Table 1. Major element chemistry of Apollo 14 black glass starting material and crystalline spheres
SiO2
glass
OLI(I 1)
34.00
39.93
OL2(23) GTl(28) 39.60
GT2(44)
GT3(26)
41.85
41.24
40.81
0.08
0.02
TiO2
16.40
0.22
A1203
4.60
22.42
23.31
23.53
Cr203
0.92
1.85
0.28
0.02
8.64
14.34
17.86
0.39
0.43
0.41
FeO
24.50
MnO
0.31
NiO
9.88
15.06
0.32
0.07
MgO
13.30
48.50
45.40
19.93
17.69
17.16
CaO
6.90
0.09
0.21
5.29
3.26
1.04
Total
100.93
98.82
100.34
100.59
100.62
100.86
Fo89.7
Fo84.3
PY69.7
PY63.0
PY61.4
Fal0.3
FalS.7
All7.0
A128.7
A135.9
Gr13.3
Gr8.3
Gr2.7
Formulas:
The Apollo 14 black glass starting material also contains 0.16 wt% K20 and 0.23 wt% Na20. Sphere compositions based on microprobe analysis of the spheres used in the experiments (number of analyses shown in parentheses). Garnet compositions are cast into normalized percentages of pyrope, almandine, and grossular.
lated by a ceramic alumina sleeve. The pressure calibrations for Hertel KO5 or Hertel KF-1 tungsten carbide (1 inch cubes with 8 mm truncations) and the Ceramacast 584 octahedra were determined using the following phase transformations: quartz/coesite (1273 K, 3.0 GPa), fayalite a / 7 (1473 K, 5.8 GPa), and coesite/stishovite (1473 K, 9.3 GPa). The uncertainty in pressure is estimated at _+0.2 GPa in experiments below 9.5 GPa. We estimate that the uncertainties are on the order of _+0.5 GPa above 9.5 GPa because the pressure calibration curve is extrapolated. Temperature uncertainties are on the order of _+40°C (Circone and Agee, 1995). Experiments were sectioned by grinding on a diamond-embedded wheel until the positions of both spheres were determined. Sections were then polished and analyzed by electron microprobe. Data collection conditions were 15 kV and 15 nA. Natural and synthetic oxides and silicates were used for standards and the raw WDS data were converted to weight percent oxide compositions using a BenceAlbee correction routine. Compositional data on Na20, K20, and MnO contents were not obtained because they are relatively minor components of the liquid (Table 1) and do not contribute significantly to the liquid density. Since the liquids quench inhomogeneously, liquid compositions were measured along transects of the charges from top to bottom and side to side to confirm that the liquid composition did not vary systematically with position in the charge. Each analysis was obtained by rastering the electron beam over 8 # by 8 # areas to obtain an average composition for the inhomogeneously quenched liquid. The sphere compositions were also checked in each experiment.
2.2. Calculation of Sphere Density The sinking and floating spheres technique does not yield a direct measurement of the density of the liquid at experimental conditions.
Instead, each experiment provides an open-ended bracket on the density of the liquid. Sinking spheres indicate that the spheres are denser than the liquid and floating spheres indicate that the spheres are less dense than the liquid. The sink/float technique uses the fact that silicate liquids are more compressible than the crystalline spheres used in the experiments. Using a sphere that has a density slightly greater than the liquid at given P, T conditions, experiments performed at increasingly higher pressure will eventually result in the liquid density exceeding that of the sphere. The liquid density can be bracketed over several pressure ranges if spheres with appropriate density, stability at experimental P, T conditions, and a well-known equation of state are available. In this study, we were able to observe conditions under which the spheres were neutrally buoyant in the liquid, providing a direct measurement of liquid density. However, neutral buoyancy results should be viewed cautiously. Additional experiments with clear sink and float results at bracketing pressures are highly desirable, and sphere position in the sectioned experiments should clearly indicate movement from the starting positions at the top and bottom of the capsule. Since the interpretation of results to obtain liquid density depends solely on the known density of the spheres at experimental P and T, a well-defined equation of state is needed for all of the spheres employed. In this study, sphere densities have been calculated using the Birch-Murnagham equation of state:
3 KT[ P = 2
L\Pr,o/
7'3 -
] \Or,o~
1-~(4
J
\\P~0/
in which Kr is the bulk modulus, K' is the pressure derivative, and
2712
S. Circone and C. B. Agee Table 2. Equation of state parameters for calculating sphere density at experimental P, T.
K298
dK/dT
K'
Or0
~1
o~2
(Mg,Fe)2SiO4
128.54A -0.02176A 5.3B 3.1131E-05 6.5693E-09 -5.8733E-01C
(Ca,Mg,Fe)3Al2Si3Ol2
168.37D -0.02420D 4.5E 2.3030E-05 6.0867E-09 -4.3472E-01F
AI203G (Mg,Fe)SiO3
254.34 -0.02530 4.23 2.7062E-05 1.3695E-09 - 1.1965E+00 100.9H
-0.00534H 5.01 4.8470E-05J
Note: The isothermal bulk modulus is defined as KT = K298 + dK/dT (T-298), where
KT is in GPa and T is in Kelvin. The thermal expansion is defined as ~(T) -- 04) + Otl*T + t~2/T2 in K-1. The density P298,0 (g/em3) was calculated using the following equations: 13298.0 (olivine) = 3.229 XFo + 4.417 XFa (Hazen, 1977) 10298,0 (game0 = 3.559 Xpy + 4.319 XAI + 3.593 XGr F P298,0 (corundum) = 3.982 G 0298,0 (pyroxene) = 3.206 XEn + 4.066 XFs(Krupka et al., 1985a) ASumino et al. (1977) BKumazawa and Anderson (1969) CSuzuki (1975) DSuzuki and Anderson (1983) ELevien et al (1979) FSkinner (1956) Gsee Goto et al., (1989) for all parameters HKT and dKddTcalculated from the relation KT = (Ks-1 + T0t2/pCp)"l, where KS is from Kumazawa (1969) and FrisiUo and Barsch (1972); Cp from Krupka et al., 1985b; values for t~ and p in table. IEstimated JFrisillo and Buljan (1972)
Pr.o and Pr.e are the densities of the sphere at temperature and at 105 Pa or high pressure. The density at temperature and 105 Pa is defined as: PT,O =
P298,0
(:;)-' 1
+
a(T)dT
(2)
98
and the parameters for Eqns. 1 and 2 for olivine, garnet, and corundum are listed in Table 2. The uncertainty in the calculated sphere density is estimated at ___0.03 g/cm 3 (-0.75%). The short experiment duration (under 2 min) resulted in no significant growth on the sphere surfaces (see below). Thus, no additional correction to the sphere densities is required.
2.3. Effect of Composition on Liquid Density The reference liquid density at 105 Pa, which is necessary for determining the liquid compressibility from the sink/float experiments, has been calculated using the data of Lange and Carmichael (1987). This dataset provides partial molar volumes and their temperature dependence for all of the major components of the Apollo 14 black glass starting material. Iron and Ti are expected to be divalent and tetravalent, respectively. Reduced species should be
limited to at most a few percent, based on the study of Schreiber et al. (1982), the Fe-FeO conditioning of the starting material, and the use of Mo capsules in the experiments. The measured liquid compositions were not significantly altered from that of the starting material (Table 1 ) in the experiments, although some Mo dissolved in the liquid and the Fe and Ti contents were slightly lower. These slight changes in liquid composition must be considered before interpreting the results of the sink/float experiments. The analyzed Mo content of the liquids, calculated as MoOa, ranged from 0.9 to 7.2 wt% MOO3. Some trends were observed in the liquid composition analyses: the multi-anvil experiments had higher Mo contamination (in all likelihood due to the higher experiment temperatures), the weight percent totals were slightly high, and comparison of a measured oxide component to the wt% total of each analysis suggested a weak correlation (25% of the observed variation) for MOO3, whereas the other oxides showed essentially no variation. The latter observations suggest that the assumed valence state of Mo 6` may be too high. Previous 105 Pa metal/silicate partitioning studies under controlled oxygen fugacity conditions have inferred that the valence state of Mo is less than 6+ under reducing conditions (Schmitt et al., 1989; Hillgren, 1991; Holzheid et al., 1994). The experiments of Holzheid et al. (1994), which were performed on Fe-free silicate liquids, suggest that Mo shifts
Compressibility of lunar glass from predominantly6+ to predominantly4+ below log fo2 = - 10.6. In our experiments, the starting material is conditioned at Fe-FeO, and the Mo capsules maintain low oxygen fugacities and Fe203 at <1 wt% (suggesting log fo2 < -10; Schreiber et al., 1982). Thus, assuming a lower valence state for Mo is reasonable and recalculation of the Mo content as M o 4+ 0 2 reduces the correlation with wt% total to <10%. The effect of dissolved Mo on silicate liquid density has not been determined experimentally.However, based on systematicsobserved in the dataset of Lange and Carmichael (1987), the partial molar volume of MoO2 in silicate liquid and its dependence on temperature can be estimated. At 105 Pa, oxide constituents of silicate liquids show similar partial molar volumes on a per oxygen basis (Stebbins et al., 1984; Lange and Carmichael, 1987) and similar compressibilities (Rivers and Carmichael, 1987), suggesting that silicate liquid structure is based on the packing of oxygen atoms with cations in the interstices (Rigden et al., 1989). The alkalis, with their larger ionic radii and higher coordination numbers, appear to be the exception to this rule and are excluded from the following discussion (based on Table 9, Lange and Carmichael, 1987). The ratios of the molar volumes of the crystalline oxides at 1673 and 298 K exhibit a narrow range (1.04-1.06; comparisons for SiO2 use the data for cristobalite). The ratios of the partial molar volume of the oxide in silicate liquid at 1673 K to the molar volume of the oxide at 298 K exhibit a much wider range (0.99-1.45), in part due to the large partial molar volumes of A1203 and Fe203, which appear to be in a lower coordination state in the liquid. We have used these ratios to make an estimate of the partial molar volume of dissolved MOO2, which has the rutile structure in the crystalline state at 298 K. Beginning with a 105 Pa, 298 K molar volume of 19.58 cm3/mol (Robie et al., 1979), the partial molar volume of dissolved MoO2 at 1673 K is expected to be between 19.4-28.4 cm3/mol. The temperature dependence of the partial molar volume ranges from 0 to 0.009 cm3/ mol. K. The combined uncertainties for the partial molar volume and thermal expansion of dissolved MoO2 result in an uncertainty of +_0.02 g/cm3 in the estimated densities of the Mo-contaminated liquids. This uncertainty estimate is conservative because it allows for the unlikely decrease in M o 4+ coordination in the silicate liquid. The uncertainty is less than _+0.0l g/cm 3 if the A1203 and Fe203 data are not considered. Since the solution of MoO2 in the liquid is counterbalanced by the loss of FeO and TiO2, in most cases the changes in liquid composition have a negligible effect on the interpretation of the experimental results. The exceptions are discussed below. 3. RESULTS The experimental conditions, analyzed liquid compositions, and calculated sphere and 105 Pa liquid densities for the sinking and floating spheres experiments performed on molten Apollo 14 black glass are summarized in Table 3. Backscattered electron images of three sectioned experiments are reproduced in Fig. 1. These experiments illustrate the progressive change in buoyancy of GT1 spheres in molten Apollo 14 black glass with increasing pressure from 5.0 to 6.0 GPa. Figure lb demonstrates the conditions under which p~q ~ PGTI. Figure 2 summarizes the results of the sinking and floating spheres experiments. The experiments bracket the liquid density under several P, T conditions. Neutral buoyancy of the OL2 spheres at 1.5 GPa, 1738 K yields a liquid density of 3.30 g/cm 3. The GT1 spheres were neutrally buoyant at 5.5 GPa, 2073 K and 6.0 GPa, 2108 K, bracketing the liquid density at 3.65 and 3.66 g/cm 3, respectively. Two pressures for the density crossover between liquid and GT1 spheres were obtained because of slight differences in the liquid compositions. The series of experiments that bracket the liquid density at lower pressure
2713
(308A8-sink at 5.0 GPa, 321A8-neutral at 5.5 GPa, and 310A8 float at 6.0 GPa) have higher calculated 105 Pa liquid densities caused primarily by higher Mo contents (Table 3). The experimental and ideal 105 Pa liquid densities are the same for the neutral buoyancy result at 6.0 GPa (401A8). At higher pressure, the GT2 spheres become neutrally buoyant between 10.0 and 11.5 GPa, corresponding to liquid densities between 3.82-3.85 g/cm 3. Experiments yielding only floating or sinking spheres provide additional constraints on liquid density at low and high pressure. The buoyancy of OL1 (San Carlos olivine) spheres defines a minimum density bracket of pl~q > 3.22 g/cm 3 at 1.0 GPa, 1688 K. At high pressure, the GT3 and corundum spheres were denser than the liquid at all pressures investigated, placing a maximum density bracket of Pl~q< 3.88 g/ cm 3 at 10.0 GPa (2333 K) that is consistent with the results obtained using the GT2 spheres. The slope of the pressure vs. density curve flattens considerably at higher pressure and temperatures. Above 10 GPa, the density difference between the GT2 spheres and molten Apollo 14 black glass appears to be small, and their compression curve slopes are similar. However, the solid and liquid do not have the same composition; therefore, this result does not imply that the dT/dP slope of the Apollo 14 black glass liquidus is near zero at these conditions. Due to the flattening of the pressure vs. density curve at high pressure, we were not able to achieve the flotation of GT2 spheres in molten Apollo 14 black glass, because the maximum pressure for the experimental setup was reached. Since we are primarily concerned with the density of the liquid at pressures relevant to the petrogenesis of high-Ti lunar picritic magmas (up to 4.0 GPa), we did not pursue higher pressure experiments using cubes with smaller truncations.
3.1. Compressibility of Molten Apollo 14 Black Glass The results summarized in Fig. 2 indicate that the liquid density increases steeply with pressure. The initial slope is on the order of 0.13 g/cm3/GPa. The slope becomes shallower as pressure increases, resulting in a concave downward curve. The curve in Fig. 2 was obtained at near liquidus conditions, reflecting the change in liquid density as a function of pressure and temperature. In order to look at the effect of pressure alone, we have calculated the liquid density under isothermal conditions. These isothermal curves (Fig. 3) suggest that the compression slope (>0.10 g/cm3/GPa) is steep over a broad pressure range (up to - 6 GPa), although the flattening of the compression curve at higher pressures is preserved. This slope is significantly higher than those measured for other ultrabasic liquids over the same pressure range and at comparable temperatures. Agee and Walker (1988a, 1993) obtained slopes of ~0.075 g/cm3/ GPa and 0.065 g/cm3/GPa for komatiitic and peridotitic liquids, respectively. The isothermal curves can be fit to Eqn. 1 to yield values for the isothermal bulk modulus and the pressure derivative. A range of values for Kr and K' satisfy the isothermal curves (Fig. 4) within the uncertainties for pressure and density, but solutions to Eqn. 1 in which both Kr and K' are either
2714
S. Circone and C. B. Agee Table 3. Summary of sink/float experimental conditions, results, and analyzed liquid compositions.
Run ID
67PC
64PC
63PC
62PC
88PC
84PC
301A8
P (GPa)
1.0
1.5
2.0
2.5
1.5
2.0
4.0
1688
1708
1738
1758
1707
1738
1983
spheres
T (K)
OL1
OL1
OLI
OLI
OL2
OL2
GTI
result
float
float
float
float
neutral
float
sink
12
8
10
10
# of analyses
13
34.26(0.77) 33.69(1.02)
34.22(0.49) 34.29(0.56)
14
15
34.32(0.84)
34.64(1.23)
SiO2
34.65(1.44)
TiO2
15.03(0.59) 15.63(0.46) 15.36(0.29) 15.74(0.43) 15.23(0.41) 15.40(0.67) 15.10(1.35)
AI20 3
4.71(0.13)
4.74(0.15)
4.73(0.08)
4.77(0.10)
4.74(0.12)
4.74(0.12)
3.81(1.09)
Cr20 3
0.67(0.40)
0.92(0.32)
0.92(0.23)
1.00(0.10) 0.75(0.14)
0.81(0.26)
0.69(0.22)
FeO
22.79(1.01)
22.86(0.70) 23.19(0.67)
22.01(0.61) 22.73(0.27)
22.64(0.66)
23.03(1.47)
MgO CaO MoO 3
14.00(0.67) 14.38(0.76) 13.88(0.61) 14.84(0.54) 14.19(0.67) 14.25(0.65) 14.29(0.75) 6.67(0.31)
6.49(0.18)
6.35(0.18)
6.56(0.25)
6.60(0.36)
6.75(0.32)
1.88(0.84) 1.55(0.64) 2.18(1.36)
6.41(0.21)
2.35(0.44)
1.84(0.59) 2.00(0.46)
3.17(1.04)
Total 100.41(0.69) 100.75(0.45) 100.43(0.46) 101.27(0.23) 100.33(0.65) 100.75(0.51) 101.47(0.37) Pliq
3.09
3.09
3.09
3.07
3.09
3.08
3.01
3.09
3.08
3.07
3.06
3.08
3.07
2.98
Psphere 3.22
3.23
3.25
3.26
3.30
3.31
3.62
Pideal liq
low or high are not consistent with the data. The "best fits" to the calculated isothermal densities (Fig. 3) yield K2173 K = 15.5 GPa, K ' = 5.2 and KITI3K = 14.9 GPa, K ' = 4.7. Isothermal bulk moduli for molten Apollo 14 black glass can also be calculated from 105 Pa ultrasonic velocity data for multicomponent silicate liquids (Rivers and Carmichael, 1987). Values of K2173K = 14.4 GPa and K1713K = 18.8 GPa are calculated (Lange and Carmichael, 1987); however, these experiments do not provide information on the pressure derivative, and the dataset includes only one TiO2-rich liquid. If the ultrasonic data is in agreement with the static compression data in this study, these KT values would be consistent with K ' = 5 - 6.5 (2173 K) and K ' = 3.5 (1713 K). The results for 2173 K can be compared to the K2173 K and K ' of other silicate liquids (Fig. 4). Molten Apollo 14 black glass is, with the possible exception of anorthite liquid, more compressible than any other silicate liquid measured under static or dynamic compression conditions thus far. There are three main compositional differences between the ultramafic black glass, komatiite, and peridotite (Agee and Walker, 1988a; Miller et al., 1991). The black glass has less SiO2 (34 vs. 4 4 - 4 7 wt% SIO2), a lower Mg# (0.49 vs. 0.820.90), and a factor of 40 times more TiO2! The higher compressibility is not related to either SiO2 or FeO content.
No correlation between silicate liquid compressibility and SiO2 content was discernible in the 105 Pa ultrasonic study of thirty-two multicomponent silicate liquids with 31-83 wt% SiO2 (Rivers and Carmichael, 1987). The high-pressure density measurements of Agee and Walker (1988a) on komatiite-fayalite mixtures showed that the compression slope is independent of FeO content. Furthermore, the black glass contains almost no highly compressible large alkali cations. Thus, the high compressibility of molten Apollo 14 black glass relative to other ultrabasic liquids must be related to the high TiO2 content of the liquid. Values of 10-15 GPa have been derived for the bulk modulus of the TiO2 component in silicate liquid (Rivers and Carmichael, 1987; Webb and Dingwell, 1994). Although values can be calculated from the results in this study (KI673K of TiO2 is 4.8-17.2 GPa from the "best fit" values at 1713 and 2173 K, respectively), the uncertainty is even greater due to the range of values for K and K' that fit the experimental data, the large temperature corrections to liquid density and compressibility, and the limitations of the ultrasonic data discussed above. Additional study of Ti-bearing silicate liquids at high pressure is needed to obtain a well-constrained value. Regardless, the derived bulk modulus of the TiO2 component from 105 Pa studies is smaller than those of other silicate
Compressibility of lunar glass
2715
Table 3 continued
Run ID
308A8
321A8
310A8
401A8
402A8
405A8
410A8
P (GPa)
5.0
5.5
6.0
6.0
8.2
8.5
9.0
T (K)
2048
2073
2077
2108
2238
2258
2283
spheres
GT1
GT1
GT1
GTI
GT2
GT2
GT2
result
sink
neutral
float
neutral
sink
sink
sink
13
14
12
13
6
12
13
# of analyses SiO2
34.44(1.98) 33.47(1.36) 34.53(1.84) 34.85(1.18) 32.72(2.22) 33.39(2.14) 34.09(3.98)
TiO2
14.72(2.10) 15.09(1.31) 14.61(1.89) 14.45(1.04) 15.95(1.77) 15.38(1.90) 14.97(3.39)
A1203
4.43(1.76)
4.91(1.56)
3.89(1.39) 5.13(0.93) 3.78(1.33)
4.49(1.63)
4.85(2.70)
Cr20 3
0.83(0.22)
0.88(0.17)
0.79(0.15) 0.87(0.09) 0.76(0.27)
0.83(0.23)
0.91(0.40)
FeO
22.44(2.23) 22.30(1.23) 22.04(2.02) 22.12(1.09) 23.65(2.61) 22.69(2.60) 21.67(4.50)
MgO
14.32(1.15) 13.99(0.63) 14.31(1.10) 14.43(0.54) 13.11(1.14) 13.44(1.34) 13.90(2.21)
CaO
6.64(0.22)
6.53(0.32)
6.79(0.20) 6.70(0.18) 6.54(0.68)
6.56(0.35)
6.75(0.73)
MoO2
3.60(1.00)
3.84(0.83)
4.30(0.84) 2.75(0.62) 3.81(0.98)
3.74(0.99)
3.77(1.79)
Total 101.42(0.22) 101.00(0.53) 101.27(0.73) 101.30(0.46) 100.31(0.51) 100.52(0.49) 100.91(0.71) Pliq
2.99
2.99
2.98
2.95
2.96
2.93
2.91
Pideal liq
2.96
2.95
2.95
2.94
2.90
2.89
2.89
3.65
3.66
3.66
3.79
3.80
3.81
Psphere 3.64
melt components, with the exception of the large alkalis. The reason for the high compressibility of the TiO2 component is unknown. At high pressure, increases in Ti coordination number with pressure may enhance liquid compressibility. Increased Ti coordination from fourfold and/or fivefold to sixfold have been demonstrated in alkali-bearing silicate liquids quenched at 0 - 3 . 0 GPa (Paris et al., 1994). However, this mechanism is unlikely to play a role in the compressibility of the Apollo 14 black glass. The presence of CaO and the absence of alkalis suggest that Ti should be in sixfold coordination at 105 Pa (Lange and Carmichael, 1987; Dingwell, 1992).
3.2. Density Crossovers in the Lunar Mantle The primary focus of this study was to test the hypothesis put forth by Delano (1990) that density crossovers between picritic magmas and lunar mantle phases played a potentially important role in the petrogenesis of the TiO2-rich glasses. Early petrogenesis models for the pristine lunar glasses involved static partial melting of a chemically heterogeneous lunar mantle at depth to produce the array of sampled glasses (e.g., Ringwood and Kesson, 1976). These models, which assume that a olivine + orthopyroxene mantle residuum was left in the source region and that no fractionation occurred prior to eruption, equate the depth of melt segregation with
the two-phase saturation point on the liquidus. Lateral chemical heterogeneity in the source region (via ilmenite-enriched zones) produced the wide range (0.4-16.4 wt%) in TiO2 content of the sampled picritic glasses. The calculations of Delano (1990) predicted a density crossover between equilibrium olivine and molten Apollo 14 black glass at ~ 2 2.5 GPa, or 400-500 km depth, which coincides with the pressure of the olivine + orthopyroxene multisaturation point on the liquidi of several picritic lunar glasses (see Delano, 1986). Olivine is the stable liquidus phase at lower pressures. At pressures above the cotectic, olivine is replaced by less dense orthopyroxene on the liquidus. Thus, molten Apollo 14 black glass would be significantly denser than the orthopyroxene-rich mantle residue over a potentially large pressure interval. Garnet would probably be consumed during partial melting, given the low A1203 contents of the picritic lunar glasses and the lunar mantle. The low pressures in the lunar interior are not sufficient to stabilize the highpressure olivine polymorphs or majoritic garnets. Since the densities of olivine and molten Apollo 14 black glass are approximately equal near the suggested cotectic pressure of ~ 2 - 2 . 5 GPa, Delano (1990) suggested that melts with even higher TiO2 contents would be precluded from erupting because they were negatively buoyant at these depths. Recent dynamic petrogenesis models have hypothesized that the picritic glasses formed by pressure-release melting of ther-
2716
S. Circone and C. B. Agee Table 3 continued
Run ID
426A8
455A8
397A8
399A8
412A8
390A8
411A8
P (GPa)
10.0
11.5
8.8
9.2
10.0
9.4
10.0
T (K)
2328
2353
2268
2283
2333
2293
2348
spheres
GT2
GT2
GT3
GT3
GT3
AI20 3
A1203
neutral
neutral
sink
sink
sink
sink
sink
14
13
10
13
14
13
13
result
# of analyses
SiO 2 34.27(1.71) 35.50(1.96) 33.93(0.96) 34.05(1.69) 33.57(1.69) 33.83(2.58) 35.06(1.31) TiO2
14.15(1.53) 17.20(2.56) 14.92(0.67) 14.84(1.48) 15.08(1.31) 15.59(2.28) 15.56(1.38)
AI20 3
5.95(1.28) 4.55(1.72) 5.08(0.45) 5.65(1.15) 4.96(1.04)
5.20(1.62)
6.30(0.90)
Cr20 3
0.93(0.16) 0.87(0.17) 0.85(0.06)
0.87(0.23)
0.93(0.11)
0.84(0.15) 0.91(0.13)
FeO
21.54(2.04) 17.71(1.93) 21.96(1.01) 20.87(1.90) 22.00(1.69) 20.72(2.84) 17.85(1.06)
MgO
14.05(0.88) 14.32(1.40) 13.80(0.43) 14.06(0.97) 13.78(0.81) 13.78(1.49) 14.28(0.89)
CaO MoO2
6.49(0.35) 7.05(0.33) 6.60(0.20) 6.44(0.37) 6.58(0.42)
6.51(0.54) 6.86(0.29)
3.50(0.99) 3.40(0.84) 3.61(0.65)
4.13(1.21)
3.85(1.01) 3.88(0.88)
3.53(0.72)
Total 100.88(0.54) 100.60(0.42) 100.74(0.37) 100.60(0.29) 100.74(0.32) 100.63(0.53) 100.37(0.68) Pliq
2.88
2.83
2.91
2.90
2.90
2.90
2.83
Pideal liq
2.87
2.86
2.89
2.89
2.87
2.88
2.86
Psphere
3.83
3.86
3.85
3.86
3.88
3.95
3.95
mally buoyant mantle plumes originating from depths of ->700 km (Hess, 1991; Longhi, 1992), where high-Ti liquids would be denser than the mantle residue. The results from this study suggest that liquid density inversions place even tighter constraints on the buoyancy of in the lunar mantle. Compositions for olivine and orthopyroxene in equilibrium with molten Apollo 14 black glass can be calculated from crystal-silicate liquid Fe-Mg distributions, using the relationship:
Ko = (X~e)(Xhg) (X sMg)(XFe) L '
(3)
where KD is has been determined experimentally. A range of 0.24-0.28 for the olivine/liquid Ko have been considered. The lower limit considers the effect of TiO2 content on Ko (Eqn. 4; Delano, 1990), and the upper limit assumes that Ko may increase with pressure (Delano, 1980). We allowed a larger range for orthopyroxene, 0.22-0.28, which has not
Table 3 continued Note: All P(i) in g/cm3. The calculated liquid densities neglect the minor contributions
of Cr203 and MnO, which were not determined in the Lange and Carmichael (1987) study. Inclusion of Na20 and K20 in the calculation of the observed liquid compositions decreases the calculated densities by -~0.01 g/cm3. Pliq is the calculated liquid density at 105 Pa and experimental temperature for the experimentally determined liquid compositions shown. Density has been calculated using an estimate of the partial molar volume of dissolved MoO2 in silicate liquid (see text). Uncertainties in this estimate result in density uncertainties of+ 0.02 g/cm 3. Pideal liq is the calculated density at 105 Pa and experimental temperature of a liquid with the ideal Apollo 14 black glass composition (Table 1).
Compressibility of lunar glass
2717
4.2
4.0 E
3.8
~
3.6
G2 G
u)
(a)
"-
3.4
n
.
OLt
3.2I
3.0
P (GPa) FIG. 2. Summary of sinking and floating spheres experiments on molten Apollo 14 black glass. All density-pressure relations are calculated for experimental P and T. The crystalline olivine, garnet, and corundum sphere densities (calculated using Eqns. 1 and 2, the compositions in Table 1, and the parameters in Table 2) are shown as thin curves. Experiments in which the crystalline spheres floated (Pnq > Psphere) are represented by upward-pointing triangles, and experiments in which the crystalline spheres sank (p~q < psph~e) are represented by inverted triangles. Experiments exhibiting neutrally buoyant spheres are shown as circles. The l0 s Pa molten Apollo 14 black glass liquid density (liquidus T ~ 1640 K, Wagner and Grove, 1995) was calculated using the data in Lange and Carrnichael (1987). The heavy curve shows the extrapolated high-pressure liquid density consistent with all of the experimental data at pressures up to 11.5 GPa and liquidus temperatures. Superfluous sinking spheres results obtained at 7.9 GPa, 2218 K and 8.4 GPa, 2248 K with GT3 spheres are not shown (liquids were not analyzed). Neutral buoyancy results obtained at 10.2 GPa, 2342 K and 10.6 GPa, 2348 K with GT2 spheres were not included because liquid compositions and calculated 105 Pa densities deviated significantly from the ideal.
been studied in detail experimentally. The proposed ranges for Ko are consistent with phase equilibria data for Apollo 14 black glass (Wagner and Grove, 1995). These Ko ranges yield equilibrium olivines of FO78 tO Fos0 and equilibrium pyroxenes of En78 to En82. Densities at P and T were calculated using the data in Table 2. The high-pressure densities of molten Apollo 14 black glass and equilibrium crystalline phases are compared in Fig. 5. Equilibrium olivine becomes neutrally buoyant at 2 . 0 -
(c)
[
_1
FIG. 1. Backscattered electron images (65× magnification) of a series of experiments that demonstrate the determination of liquid density. Initially, in all experiments, one sphere was positioned at the top and one at the bottom of the Mo capsule. (a) through (c) illustrate the effect of increasing pressure on GT1 sphere buoyancy in molten Apollo 14 black glass: (a) sink-p~iq < PGXl at 5.0 GPa
(experiment 308A8), (b) neutral-pl~q = PGTI at 5.5 GPa (experiment 321A8), and (c) float-pliq ~- PGT] at 6.0 GPa (experiment 310A8, the dotted oval indicates the position of the second sphere, which was ground away during sectioning). Figure orientation represents that of the charges during the experiments. Although the liquids quench inhomogeneously, there are no apparent large-scale compositional variations or evidence of thermal boundaries (i.e., equilibrium crystal growth) in the charges. We have concluded, based on number of observations, that the ~ 10/zm growth rims on the spheres formed during quench: the sharp compositional gradient at the sphere boundary, the erose outer edges, the trapped liquid inclusions, and the short experiment duration at liquidus conditions. The equilibrium growth rims observed by Agee and Walker (1988a) formed during experiments lasting >10 rain and exhibited smooth diffusion profiles.
2718
S. Circone and C. B. Agee
4.64.23.8-
.~,
"~3.4 2173 K
121 3.0 ¸ 2.6
o
4
6
8
I'o
12
P (GPa) FIG. 3. The density of molten Apollo 14 black glass at pressure and temperature. The solid circles at 0, 1.5, 5.5, 6.0, 10.0, and 11.5 GPa and the heavy curve are identical to those in Figure 2. The open circles at 5.5 and 11.5 GPa represent small density corrections (-0.04 and +0.03 g/cm3, respectively) to the neutral buoyancy results due to small but significant deviations in liquid composition (Table 3). The error bars represent the uncertainties in pressure and density (see text): 105 Pa liquid density +_0.02 g/cm3, sphere density +__0.03g/cm3, and pressure _0.1 for P < 2.5 GPa, ___0.2for 3.0 < P < 9.5 GPa, and ___0.5GPa for P > 9.5 GPa. Isothermal curves were calculated for 1713 K (approximate liquidus temperature of olivine + orthopyroxene multisaturation point; Wagner and Grove, 1995) and 2173 K, assuming a thermal expansion of a = 11.8 × 10-5/K for the liquid (Lange and Carmichael, 1987) and dtz/dP 0. The steeper slope of the 1713 K isothermal curve would appear to suggest that the liquid is more compressible at low temperature; however, this is probably an artifact of the large thermal expansion corrections to density ( + 11% ) for the high pressure experiments.
greater depths as its P - T path is deflected to a steeper slope when it approaches the solidus (i.e., the Verhoogen effect). Delano (1990) has also suggested that the dense, trapped melts were rich in incompatible elements such as K, U, and Th. Thus, they may have supplied their own radioactive heat source to help maintain supersolidus temperatures during sinking. The predictions of neutral buoyancy are for liquidus conditions and in the presence of a two-phase mantle residue. Other factors may influence magma buoyancy. The presence of minor amounts of higher density phases, such as ilmenite in the bulk mantle or chrome spinel in the residue (the latter coexists with orthopyroxene at >1.5 GPa; Wagner and Grove, 1995), would increase the pressure needed for neutral buoyancy. If the magma is generated in a thermally buoyant mantle plume (Hess, 1991), then magma buoyancy is predicted with respect to the partially melted residuum ascending within the diapir. Since the thermal expansion of silicate liquids is greater than that of the solids, the magma could be buoyant with respect to the colder surrounding bulk mantle at greater depths than predicted above. Nonetheless, it is clear that the P, T conditions necessary for crystal/ liquid density inversions are easily attainable at moderate
7.5. O Di
2173 K
6.5..
An36Di64
5,5" ' : : ~ i ~ ' " ~ , . . , ~ , K°matiite
4.5. 2.3 GPa (410-520 km), and equilibrium orthopyroxene is neutrally buoyant at 0.5-0.8 GPa (100-160 km). At the cotectic pressure of 1.5 GPa (300 krn), our compressibility data predict that the density of molten Apollo 14 black glass is 3.30 g/cm 3. By inspection of the liquid mineral compression curves in Fig. 4, we calculate that a mantle residue of 58% olivine and 42% orthopyroxene will be neutrally buoyant relative to molten black glass at this pressure. Our results confirm the prediction of Delano (1990), and we have placed the maximum depth for buoyant rise of highTi lunar basalt at ~400 km. High-Ti magmas generated at depths greater than 400 km would be negatively buoyant and should sink deeper into the lunar interior. Delano (1990) has even speculated that the seismic attenuating zone at 1000-1400 km depth may contain some high-Ti melts that have reach this level by downward migration. P. Hess (pers. commun.) has pointed out that sinking melts may not survive as readily as their rising counterparts. Mantle adiabats tend to have shallower P - T slopes than the liquidi and solidi of most mafic and ultramafic compositions. Therefore, a dense liquid descending along a lunar adiabat may solidify after traveling only a modest distance. On the other hand, the heat of crystallization will tend to keep the magma molten to
3.52.5
0
2'0
3'0
40
K (GPa) FIG. 4. Range of calculated values for the Birch-Murnaghanisothermal bulk modulus (Kr) and pressure derivative (K') of molten Apollo 14 black glass at 2173 K (solid envelope). The combinations of Kr and K' in the envelope satisfy Eqn. 1 and the following input data: the calculated 105 Pa liquid density (Lange and Carmichael, 1987), the pressure-densityrelations of the isothermal curves in Fig. 3, and the uncertainties shown as error bars in Fig. 3. The KITI3K, K' envelope ranges from 13.5 GPa, 5.9 to 21.2 GPa, 2.9 (not shown). The "'best fit" pair of K2173 K and K' obtained by least squares analysis of the data is represented by the square. Measured values for the K2173 K and K' of other silicate liquids are shown for comparison (modified Fig. 6b of Agee and Walker, 1993): komatiite (dotted envelope from Agee and Walker, 1993; point from Miller et al., 1991); basalt analogue (Rigden et al., 1984, 1988); diopside and anorthite (Rigden et al., 1989). Peridotite (Agee and Walker, 1993) has a higher Kr than komatiite for a large range of K' (K' = 2-12, not shown).
Compressibility of lunar glass
Acknowledgments--This research was supported by NASA Grant #NAGW-3996. S. Circone was supported by the Science Scholars Fellowship Program at the Bunting Institute of Radcliffe College. We would like to thank T. P. Wagner of MIT for providing the conditioned starting material. At Harvard University, we would like to thank Jie Li for performing some of the sink/float experiments and Jennifer R. Smith for microprobe analyses of several experiments. R. Lange, J. W. Delano, and D. B. Dingwell provided helpful reviews of the manuscript.
3.63,5-
~ " "
2719
3.43.3-
Editorial handling: M. S. Ghiorso
E3
3.2REFERENCES
3.1 crossover
3.0
crossover
II 0
I, I i
2 P (GPa)
3
4
FIG. 5. Density of molten Apollo 14 black glass (heavy curve, from Figure 2), olivine, and orthopyroxene (thin curves ) at pressure and liquidus temperatures. Density relationships demonstrate the pressures (located by open arrows) at which liquid/mantle residue density crossovers were possible in the lunar mantle. The filled arrow indicates the pressure of the olivine + orthopyroxene multisaturation point (Wagner and Grove, 1995); olivine is the liquidus phase at lower pressures, and orthopyroxene becomes stable above the multisaturation point.
depths in the lunar mantle and may have played an important role in controlling the redistribution of elements within the Moon following the global differentiation event (Delano, 1990; Hess and Parmentier, 1995). The Apollo 14 black glass forms one of the densest liquids so far observed on the Moon due to the high concentration of the dense, compressible TiO2 component. Our results lend strong support to the prediction of Delano (1990) that liquids with even higher Ti content may have formed but were too dense to ever erupt, but they do not necessarily place constraints on the depth of origin of the glasses with lower TiO2 contents. Experiments are currently underway on the Apollo 15 green glass (the most TiO2-poor pristine glass) and an intermediate composition to determine the location of crystal/liquid density inversions for this composition. 4. CONCLUSIONS The density of molten Apollo 14 black glass has been determined up to 11.5 GPa under static compression conditions. The pressure-density relations have been determined at 1.5, 6.0, and 10-11.5 GPa using the sinking and floating spheres technique. Molten Apollo 14 black glass is more compressible than other ultrabasic silicate liquids, apparently due to the high TiO2 content of the liquid. Our experiments provide the first measurements of its kind on Ti-rich silicate liquids. The density inversions in the lunar mantle that were predicted by Delano (1990) are supported by our experimental results, which indicate a maximum depth for buoyant rise of high-Ti lunar basalt at ~ 4 0 0 km.
Agee C. B. and Walker D. (1988a) Static compression and olivine flotation in ultrabasic silicate liquid. J. Geophys. Res. 93, 34373449. Agee C. B. and Walker D. (1988b) Mass balance and phase density constraints on early differentiation of chondritic mantle. Earth Planet. Sci. Lett. 90, 144-156. Agee C. B. and Walker D. (1993) Olivine flotation in mantle melt. Earth Planet. Sci. Lett. 114, 315-324. Akella J. and Kennedy G. C. (1971) Melting of gold, silver, and copper--proposal for a new high-pressure calibration scale. J. Geophys. Res. 76, 4969-4977. Boyd F. R. and England J. L. (1960) Apparatus for phase equilibrium measurements at pressures to 50 kilobars and temperatures up to 1750°C. J. Geophys. Res. 65, 741-748. Circone S. and Agee C. B. (1995) Effect of pressure on cation partitioning between immiscible liquids in the system TiO2-SiO2. Geochim. Cosmochim. Acta 59, 895-907. Delano J. W. (1980) Chemistry and liquidus phase relations of Apollo 15 red glass: Implications for the deep lunar interior. Proc. 11th Lunar Planet. Sci. Conf., 251-288. Delano J. W. (1986) Pristine lunar glasses: criteria, data, and implications. J. Geophys. Res. 91, 201-213. Delano J. W. (1990) Buoyancy-driven melt segregation in the earth's moon, 1. Numerical Results. Proc. 20th Lunar Planet. Sci. Conf., 3-12. Dingwell D. B. (1992) Density of some titanium-bearing silicate liquids and the compositional dependence of the partial molar volume of TiOz. Geochim. Cosmochim. Acta 56, 3403-3407. Dziewonski A. M. and Anderson D. L. ( 1981 ) Preliminary reference Earth model. Phys. Earth Planet. Int. 25, 297-356. Goto T., Anderson O. L., Ohno I., and Yamamoto S. (1989) Elastic constants of corundum up to 1825 K. J. Geophys. Res. 94, 75887602. Frisillo A. L. and Barsch G. R. (1972) Measurement of singlecrystal elastic constants of bronzite as a function of pressure and temperature. J. Geophys. Res. 77, 6360-6384. Frisillo A. L. and Buljan S. T. (1972) Linear thermal expansion coefficients of orthopyroxene to 1000°C. J. Geophys. Res. 77, 7115-7117. Hazen R. M. (1977) Effects of temperature and pressure on the crystal structure of ferromagnesian olivine. Amer. Mineral. 62, 286-295. Hess P. C. (1991) Diapirism and the origin of high TiO2 mare glasses. Geophys. Res. Lett. 18, 2069-2072. Hess P. C. and Parmentier E. M. (1995) A model for the thermal and chemical evolution of the Moon's interior: implications for the onset of mare volcanism. Earth Planet. Sci. Lett. 134, 501514. Hillgren V. J. (1991) Partitioning behavior of Ni, Co, Mo, and W between basaltic liquid and Ni-rich metal: implications for the origin of the moon and lunar core formation. Geophys. Res. Lett. 18, 2077-2080. Holzheid A., Borisov A., and Palme H. (1994) The effect of oxygen fugacity and temperature on solubilities of nickel, cobalt, and molybdenum in silicate melts. Geochim. Cosmoehim. Acta 58, 1975-1981.
2720
S. Circone and C. B. Agee
Hubbard N. J. and Minear J. W. (1975) A chemical and physical model for the genesis of lunar rocks; Part II, Mare basalts. Lunar Sci. VI, 405-407. Krupka K. M., Robie R. A., Hemingway B. S., Kerrick D. M., and Ito J. (1985a) Low-temperature heat capacities and derived thermodynamic properties of anthophyllite, diopside, enstatite, bronzite and wollastonite. Amer. Mineral. 70, 249-260. Krupka K. M., Hemingway B. S., Robie R. A., and Kerrick D. M. (1985b) High-temperature heat capacities and derived thermodynamic properties of anthophyllite, diopside, dolomite, enstatite, bronzite, talc, tremolite and wollastonite. Amer. Mineral. 70, 261 271. Kumazawa M. (1969) The elastic constants of single-crystal orthopyroxene. J. Geophys. Res. 74, 5973-5980. Kumazawa M. and Anderson O. L. (1969) Elastic moduli, pressure derivatives, and temperature derivatives of single-crystal olivine and single-crystal forsterite. J. Geophys. Res. 74, 5961-5972. Lange R. A. and Carmichael I. S. E. (1987) Densities of Na20K20-CaO-MgO-FeO-FezO3-A1203-TiO2-SiO2 liquids: new measurements and derived partial molar properties. Geochim. Cosmochim. Acta 51, 2931-2946. Levien L., Prewitt C. T., and Weidner D. J. (1979) Compression of pyrope. Amer. Mineral. 64, 805-808. Longhi J. (1992) Origin of picritic green glass magmas by polybaric fractional fusion. Proc. 22nd Lunar Planet. Sci. Conf., 343-353. Miller G. H., Stolper E. M., and Ahrens T. J. (1991) The equation of state of a molten komatiite 1. Shock wave compression to 36 GPa. J. Geophys. Res. 96, 11831 - 11848. Ohtani E., Nagata Y., Suzuki A., and Kato T. (1995) Melting relations of peridotite and the density crossover in planetary mantles. Chem. GeoL 120, 207-221. Paris E., Dingwell D. B., Seifert F. A., Mottana A., and Romano C. (1994) Pressure-induced coordination change of Ti in silicate glass: a XANES study. Phys. Chem. Minerals 21, 510-515. Rigden S. M., Ahrens T. J., and Stolper E. M. (1984) Densities of liquid silicates at high pressures. Science 226, 1071-1074. Rigden S. M., Ahrens T. J., and Stolper E. M. (1988) Shock compression of molten silicate: results for a model basaltic composition. J. Geophys. Res. 93, 367-382. Rigden S. M., Ahrens T. J., and Stolper E. M. (1989) High-pressure
equation of state of molten anorthite and diopside. J. Geophys. Res. 94, 9508-9522. Ringwood A. E. and Kesson S. E. (1976) A dynamic model for mare basalt petrogenesis. Proc. 7th Lunar Planet. Sci. Conf., 1697-1722. Rivers M. L. and Carmichael I. S. E. (1987) Ultrasonic studies of silicate melts. J. Geophys. Res. 92, 9247-9270. Robie R. A., Hemingway B. S., and Fisher J. R. (1979) Thermodynamic properties of minerals and related substances at 298.15 K and 1 bar (105 pascals) pressure and at higher temperatures. U.S.G.S. Bull. 1452, 1-456. Schmitt W., Palme H., and Wanke H. (1989) Experimental determination of metal/silicate partition coefficients for P, Co, Ni, Cu, Ga, Ge, Mo, and W and some implications for the early evolution of the Earth. Geochim. Cosmochim. Acta 53, 173-185. Schreiber H. D., Balazs G. B., Shaffer A. P., and Jamison P. L. (1982) Iron metal production in silicate melts through the direct reduction of Fe(II) by Ti(III), Cr(II), and Eu(II). Geochim. Cosmochim. Acta 46, 1891-1901. Skinner B. J. (1956) Physical properties of end-members of the garnet group. Amer. Mineral. 41, 428-436. Stebbins J. F., Carmichael I. S. E., and Moret L. K. (1984) Heat capacities and entropies of silicate liquids and glasses. Contrib. Mineral Petrol 86, 131-148. Sumino Y., Nishizawa O., Goto T., Ohno I., and Ozima M. (1977) Temperature variation of elastic constants of single-crystal forsterite between - 1 9 0 ° and 400°C. J. Phys. Earth 25, 377-392. Suzuki I. (1975) Thermal expansion of periclase and olivine, and their anharmonic properties. J. Phys. Earth 23, 145-159. Suzuki I. and Anderson O. L. (1983) Elasticity and thermal expansion of a natural garnet up to 1000K. J. Phys. Earth 31, 125138. Wagner T. P. and Grove T. L. (1993) Origin of high-Ti lunar ultramafic glasses. Proc. 24th Lunar Planet. Sci. Conf., 1475-1476 (abstr.). Wagner T. P. and Grove T. L. ( 1995 ) Origin of high-Ti lunar magmas (in prep). Walker D., Carpenter M. A., and Hitch C. M. (1990) Some simplifications to multi-anvil devices for high pressure experiments. Amer. Mineral. 75, 1020-1028. Webb S. L. and Dingwell D. B. (1994) Compressibility of titanosilicate melts. Contrib. Mineral. Petrol. 118, 157-168.