Partitioning of rare earth elements, Y, Th, U, and Pb between pargasite, kaersutite, and basanite to trachyte melts: Implications for percolated and veined mantle



[1] A new set of partitioning data for rare earth elements (REE: La, Ce, Nd, Sm, Eu, Gd, Dy, Er, and Yb), Y, Th, U, and Pb has been obtained for 25 calcic amphiboles (pargasites and kaersutites) crystallized from alkali-basaltic and basanitic bulk rock compositions at image pressure P = 1.4 GPa, and temperature T between 950° and 1075°C. The variations of amphibole/liquid partition coefficients and of their ratios relevant to petrogenetic studies are discussed with reference to the major element composition of the amphiboles and of the coexisting melt, and to the crystal chemical mechanisms for trace element incorporation. Our results support the conclusions that REE and actinides are incorporated into the M4 cavity in calcic amphiboles and distributed between the two available sites within that cavity and that Pb is incorporated into the A site. In our sample population, REE patterns are systematically enriched in heavy REE (HREE), as expected from the presence of significant cummingtonite component. No significant fractionation is observed between Th and U. The major factor controlling the amount of trace element incorporation is the SiO2 content of the melt. The major implication of this study is that HREE can become compatible in amphibole in systems with SiO2 content greater than ∼50 wt %, whereas LREE always remain incompatible. We use the new DREEamph/l values to calculate the effects of amphibole crystallization during melt migration in the upper mantle by reactive porous flow as well as fractional crystallization of amphibole during melt migration in veined systems. We show that both processes will lead to residual liquids and solids with extremely variable LaN/YbN ratios.

1. Introduction

[2] Rare earth elements (REE), Y, Th, U, and Pb are among the trace elements most used to interpret petrogenetic processes and to define isotope systematics. REE fractionation between melts and mineral phases is commonly used to constrain the mineralogical composition of the magma source, the degree of partial melting, or processes that may have changed the original melt composition (e.g., crustal assimilation). However, correct use of this method is dependent on a thorough understanding of REE partitioning between the melt and the mineral phases (and between any two mineral phases) as a function of both composition and the intensive parameters of the system.

[3] Similarly, detailed knowledge of U, Th, and Pb partitioning between mantle minerals and melts is needed to understand both their isotope systematics and their role during petrogenetic processes (e.g., Pb enrichment is considered a marker for a sediment component in arc magmas [Johnson and Plank, 1999]). The relative fractionation of U, Th, and Pb is also relevant to petrogenetic studies. U/Th, U/Pb, and Th/Pb ratios are particularly important to decipher disequilibria within the U series occurring during mantle melting [Lundstrom et al., 1994; Elliott, 1997; Wood et al., 1999] and also the evolution of radiogenic and unradiogenic trends in 208Pb/204Pb, 207Pb/204Pb, and 206Pb/204Pb ratios following crystallization of a mineral phase [Schmidt et al., 1999].

[4] The behavior of REE, Y, U, Th, and Pb in mantle processes has been so far systematically investigated only for the most abundant mantle minerals suitable for their incorporation, namely, clinopyroxene [LaTourrette and Burnett, 1992; Beattie, 1993a; Hack et al., 1994; Lundstrom et al., 1998; Wood et al., 1999] and garnet [Beattie, 1993b; LaTourrette et al., 1993; Hauri et al., 1994; Salters and Longhi, 1999; van Westrenen et al., 1999]. The potential importance of amphibole in the genesis of subduction-related and intraplate magmatism has been only recently recognized [Ionov and Hofmann, 1995, and references therein]. Studies of trace element partitioning in amphibole at different pressure P, temperature T, and composition X conditions are still sporadic and do not cover the full range relevant to natural rocks. Data are available for amphiboles in equilibrium either with low-SiO2 melts (39–45 wt % [LaTourrette et al., 1995; Adam and Green, 1994; Dalpé and Baker, 1994]) or with high-SiO2 melts (>55 wt % [Sisson, 1994; Klein et al., 1997]). Only Brenan et al. [1995] provide one data point in the range 45–55 wt % SiO2, although this is the optimal compositional range for the onset of amphibole crystallization in hydrous natural systems such as andesites. Moreover, all these works focus only on REE, so that only two measurements are available for U, Th, and Pb [Brenan et al., 1995; LaTourrette et al., 1995].

[5] In this work we provide a set of experimentally determined Damph/l for REE, Y, Th, U, and Pb obtained from pargasite/kaersutite crystallized in equilibrium with alkaline melts (SiO2 = 41.5–54.6 wt %) at different T and X conditions. The distinct site preferences ruling REE incorporation in different amphibole compositions (pargasite/kaersutite and richterite) were detected and discussed by Bottazzi et al. [1999] by combining information obtained from complete chemical characterization by electron and ion microprobe and crystal chemical analysis by single-crystal structure refinement; they will be discussed in section 3.1. In this paper we discuss the various factors affecting Damph/l in pargasite and kaersutite, with the aim of putting constraints on the role of amphibole in mantle processes such as fractional crystallization and mantle metasomatism

2. Experimental Work

2.1. Synthesis and Analytical Procedures

[6] Two distinct starting compositions were used for this experimental work: an alkali basalt from Hessen, Germany [Wedepohl, 1983, sample 4722-13a], and a basanite from Victoria Land, Antarctica [Wörner et al., 1989, sample WR13-141]. They were reproduced by thoroughly mixing oxides and carbonates of Si, Ti, Al, Mg, Fe, Ca, Na, and K. Carbonate components were eliminated by sintering the mixture for 1 hour at 930°C. To avoid Fe loss, Fe was added to the oxide-carbonate mixture after the sintering procedure as Fe0 + Fe2O3. Starting materials were modified by slightly varying the ratios MgO/(MgO + FeO) (Mg #), Na2O/(Na2O + K2O) (Na #), and TiO2/(TiO2 + SiO2) in order to crystallize amphiboles with different crystal chemical features. Starting material compositions are reported in Table A1.

[7] All starting materials, with the exception of samples 1 and 8 (natural powders), were doped with a mixture of 33 trace elements of geological relevance, including REE and U, Th, and Pb. REE (La, Ce, Nd, Sm, Eu, Gd, Dy, Er, and Yb) were added as REE2O3 powders with concentrations increasing from La (55 ppm) to Yb (550 ppm) in order to minimize the REE-oxygen spectral interferences during secondary ion mass spectrometry (SIMS) analyses and to optimize counting statistics to better than 2%. Y was added as Y2O3 with a concentration of 70 ppm. U, Th, and Pb were added as U2O3, ThF2OHx and PbO, respectively. Because of the very low Damph/l reported from previous studies [Brenan et al., 1995; LaTourrette et al., 1995] these elements were added at relatively high concentrations (3000 ppm U and Th and 5000 ppm Pb) to achieve counting statistics <2% in SIMS analyses. The doping procedure consisted of preparing the trace element mixture with elements in the desired proportions and then in mixing it with the major element mixture. In order to reduce errors due to mixing small quantities, greater amounts of starting material (∼10 g) than those required for a single experimental charge were prepared.

[8] In order to promote amphibole crystallization, 20 wt % of H2O was added with a microsyringe to all starting materials. This large amount was required in order to offset H2O loss during capsule sealing and the experimental run. SIMS analyses of H2O content of the residual glasses indicate concentrations always <6 wt %.

[9] Syntheses were done with a 22 mm piston cylinder apparatus of the type described by Johannes [1973] using a polycrystalline CaF2 assembly. In order to reduce the loss of hydrogen during the experiments the capsules were surrounded by BN [Truckenbrodt et al., 1997]. Temperature was measured and controlled with Pt/Pt90Rh10 thermocouples and a Eurotherm 812 controller. All experiments were performed at the fixed pressure of 1.4 GPa. Experiments were run in piston-out mode; first the system was pressurized to 0.1 GPa above running conditions after which charges were brought to superliquidus conditions (1245°/1255°C) for 1 hour. Pressure was then lowered to run pressure, and samples were cooled down to the equilibrium temperature for 36 hours using temperature ramps between 0.1° and 1°C/min. This procedure optimized the growth of small numbers of large unzoned crystals. Run temperatures ranged from 950° to 1075°C as a function of the composition of the starting material (see Table 1 for details). Initial quench rates of ∼80°C/s were attained by cutting off the power. Graphite-lined Pt capsules were used in all experiments, giving image conditions around 2 log units below three-dimensional fayalite magnetite quartz (FMQ). No reversal experiments were performed.

Table 1. Experimental Details and Mineral Assemblages of Run Productsa
SampleCodeTsl, °CTeq, °CMineral Assemblage
  • a

    The code is a combination of (1) the composition of the starting material (A, olivine alkali-basalt 472213a; B, basanite WR13-141), (2) the vector along which the composition was varied (N, natural composition; K, K2O/(K2O+Na2O); M, MgO/(MgO+FeO*); T, SiO2/(SiO2+TiO2)), and (3) the value of the A/(A + B) ratio between the two oxides that were varied. Melt denotes rock powder, and asterisk denotes not doped

1A-N-melt*12451015Gl + Amph + Cpx
2A-N-melt12451015Gl + Amph + Cpx + Phl
3A-N-synth12451015Gl + Amph + Cpx
4A-K-1.0012451015Gl + Amph + Cpx
5A-K-0.8112451015Gl + Amph + Cpx
6A-K-0.7112451015Gl + Amph
7A-M-0.451245950Gl + Amph + Cpx
8B-N-melt*12451015Gl + Amph + Cpx + Ol
9B-T-0.891245975Gl + Amph + Cpx +Ilm
10B-T-0.8912451015Gl + Amph + Cpx + Ilm
11B-T-0.8912451035Gl + Amph + Cpx + Ilm
12B-T-0.8912451055Gl + Amph + Cpx
13B-T-0.9412451015Gl + Amph + Cpx
14B-T-0.9412451035Gl + Amph + Cpx
15B-T-0.9412451055Gl + Amph + Cpx
16B-T-0.9412451075Gl + Amph + Cpx + Phl
17B-T-0.971245975Gl + Amph + Cpx + Ol
18B-T-0.9712451015Gl + Amph + Cpx + Ol
19B-M-0.4512451045Gl + Amph + Cpx
20B-M-0.7512451050Gl + Amph + Ol
21B-M-0.9012451050Gl + Amph + Ol
22B-M-1.0012551070Gl + Amph + Ol
23B-K-1.0012451030Gl + Amph + Cpx
24B-K-0.5012451030Gl + Amph + Cpx
25B-K-0.8112451030Gl + Amph + Cpx + Phl

[10] Major elements were analyzed on polished mounts of the experimental charges with a JEOL JXA-840A electron microprobe (EMP). Trace element compositions of the amphiboles and coexisting glasses were determined by a Cameca IMS 4f ion probe on spots of 20–25 μm diameter and using 30Si as internal standard and different well-characterized amphiboles [Ottolini et al., 1993] as external standards. Because REE display similar ion yields relative to Si in amphibole and glasses [Ottolini et al., 1993], the same sensitivity factors were used for both phases. More details of the analytical procedures for the quantification of REE and Y are reported by Bottazzi et al. [1994]. For quantification of heavy elements the method developed by Schmidt et al. [1999] was adopted, and ionic currents were measured for 232Th, 238U, and 206Pb. The accuracy of U, Th, and Pb analyses is difficult to estimate because of the scarcity of suitable standards. However, any systematic error in the calibration of ion currents versus concentrations is cancelled when calculating ratios between concentrations. Matrix effects on partition coefficients are therefore very small because of the small compositional changes between amphibole and glass, and hence accuracy of ratios between concentrations is estimated to be of the order of a few percent. Combination of chemical analyses and single-crystal structure refinements (SREF) allowed us to calculate accurate crystal chemical formulae for all the crystals, and these were used to decipher the crystal chemical mechanisms for trace element incorporation (REE from Bottazzi et al. [1999] and high field strength elements (HFSE) from Oberti et al. [2000] and Tiepolo et al. [2000a, 2000b]) and are used in this work to unravel the effects of the composition of melt and mineral structure on the incorporation of REE, U, Th, and Pb in amphiboles.

2.2. Experimental Products

[11] Experimental run products are amphibole, glass, and subordinate high-temperature phases which vary as a function of the composition of the starting material (Table 1). In Mg-rich systems, olivine is the dominant high-temperature crystalline phase, whereas clinopyroxene is the dominant phase in Fe-rich systems. The presence of both olivine and clinopyroxene was observed in intermediate compositions. Ilmenite is present as an accessory mineral in Ti-rich systems (T-0.89; see Table 1) at temperatures below 1055°C, and in samples 2, 16, and 25, accessory mica was found. As shown by textural relations in Figure 1, amphibole crystallization follows that of olivine and clinopyroxene. Textural relations testify to simultaneous crystallization of ilmenite, mica, and amphibole. Glass is clear and no quench crystals were found. The degree of crystallinity is <50% in all experiments.

Figure 1.

Backscattered image of textural relations between mineral phases and glass in run products.

[12] Amphiboles are titanian pargasites and kaersutites with the following compositional variations: dehydrogenation (O3O2−) from 0.61 to 1.11 atoms per formula unit (apfu); [4]Al from 1.92 to 2.19 apfu; M4Na from 0.02 to 0.18 apfu; and cummingtonite component, expressed as M4(Fe+Mg), from 0.12 to 0.47 apfu. Sample 16 was crystallized by a powder mixture similar to sample 15 but enriched in fluorine and is used to monitor the effect of fluorine incorporation on Damph/l; it has a higher F content (0.75 apfu), lower [4]Al content (1.77 apfu), and lower dehydrogenation (0.24 apfu). Within the sampling, Mg # ranges from 0.36 to 1.00, and Na # ranges from 0.67 to 0.99 and always closely resembles that of the starting materials. Glasses coexisting with amphibole range from picrobasalt through trachyandesite to trachyte. Major element compositions for amphiboles and glasses are given in Tables A2 and A3 [from Tiepolo et al., 2000a]. The REE, Y, U, Th, and Pb contents in amphiboles and glasses, which are the subject of the present work, are reported in Tables A4 and A5, respectively. The concentrations of each REE in the glass are <864 ppm, whereas those of U, Th, and Pb are <6360 ppm.

2.3. Henry's Law and Equilibrium Conditions

[13] Because no reversal experiments were performed and run time was not varied, the approach to equilibrium was assessed on the basis of major and trace element zoning of crystals and glasses and on the basis of crystal morphology. Major element homogeneity was checked by electron microprobe traverses within amphibole grains and by comparing random spots in the glass. Results indicate a lack of zonation in both amphibole and glass. Intragrain trace element homogeneity was checked on samples in which crystal size allowed multiple spot analyses. Glass homogeneity was checked by comparing trace element composition of random spots and of spots close to and far from amphibole grains. Results for sample 1 are reported in Figure 2a. The similarity in trace element composition in both cases suggests homogeneous distribution of trace elements in both amphibole and melt. Major and trace element homogeneity of both amphibole and glass and the euhedral habitus of amphibole crystals (Figure 1) demonstrate conditions close to equilibrium.

Figure 2.

Demonstration of homogeneity and equilibrium in experimental phases. (a) Comparison between rare earth elements (REE) concentrations (ppm) in sample 1 at core and rim of amphibole grain (circles) and in glass (squares) close to and far from the amphibole grain. (b) Comparison of amphibole/glass concentration ratios of REE between doped (3) and undoped (1) samples.

[14] In order to verify whether the doping of trace elements led to deviation from Henry's law behavior, the amphibole/melt partition coefficients for REE of undoped (sample 1) and doped (sample 3) samples synthesized from the same starting material at the same experimental conditions are compared in Figure 2b. For most REE, doped and undoped samples overlap. The higher Gd value of the undoped sample is related to the molecular interference of 141Pr+16O on 157Gd, which can be neglected for the doped sample. Discrepancies in Eu values are probably related to slightly different experimental conditions which may affect fO2 (e.g., H2O content). Because U, Th, and Pb are below detection limits in amphiboles from the undoped sample, it was not possible to directly verify their Henry's law behavior. Nevertheless, deviation from Henry's law is not expected because although the doping levels for U, Th, and Pb are higher on a weight basis, they are not so extreme on a molar basis. Henry's law behavior has been shown previously to apply for elements doped at levels well above 1 wt % [Beattie, 1993c], which is well in excess of the doping levels used in our study. Because both equilibrium and Henry's law requirements seem to have been satisfied, ratios of trace element concentrations between amphibole and melt can be considered as equilibrium solid/liquid partition coefficients. Amphibole-liquid partition coefficients Damph/l were calculated for all samples by averaging repeated analyses on adjacent spots, and resulting values are reported in Table 2.

Table 2. Amphibole/Melt Partition Coefficients for REE, Th, U, and Pb and La/Yb, La/Sm, and U/Th Ratiosa
  • a

    Vaules are averaged within all experiments.


3. Amphibole/Liquid Partition Coefficients: Discussion

[15] The major factors affecting solid/liquid partition coefficients are expected to be the crystal chemistry of the solid phase [Blundy and Wood, 1994], the melt composition [Watson, 1976], and pressure and temperature conditions [Adam and Green, 1994]. Because pressure is kept constant in all experiments, its effect on Ds/l will be neglected in the following discussion. Moreover, given the close relation between amphibole composition and equilibrium temperature, the narrow range of temperatures studied here does not allow discrimination of the individual contributions of temperature and composition for samples crystallized from the same starting material. Thus the following discussion is focused only on the effects that crystal chemistry and melt composition may have on Damph/l.

3.1. REE and Y

[16] All amphiboles studied show a marked increase in compatibility from light REE (LREE) to heavy REE (HREE) (Figure 3). Bottazzi et al. [1999] compared the DREEamph/l patterns in pargasite, kaersutite, and richterite and showed that the REE site preference is mainly a function of the major element composition of the B group sites (Na and Ca at M4 and Mg and Fe2+ at M4′). LREE preferentially enter the [8]M4 site, whereas HREE enter smaller sites with lower coordination numbers, namely, the [6+2]M4′ site in pargasite and kaersutite and the [6]M2 site in richterite, which is larger than the [6]M2 site in pargasite. The availability of the M4′ site depends on the presence of the cummingtonite component. The higher compatibility for HREE (centered on Dy) in all the samples of this work is in agreement with their incorporation at the M4' site: Bottazzi et al. [1999] showed that the structure-refined size of the M4′ site (1.03 Å) corresponds to the observed peak in the partitioning pattern at Dy-Tb ([8]r=1.027–1.040 Å), whereas the peak should be at Nd ([8]r=1.109 Å) if the REE preferentially occupy M4 (structure-refined size of 1.10–1.11 Å). The behavior of Y is analogous to that of HREE.

Figure 3.

REE patterns of partition coefficients from the amphibole/liquid pairs of our sample set.

[17] Damph/l for the individual REE vary by up to 1 order of magnitude. LREE are always incompatible (e.g., DLaamph/l ranges from 0.057 to 0.50), whereas HREE are generally higher and may even become compatible (e.g., DYbamph/l ranges from 0.25 to 2.01). Figure 4a shows the dependence of Damph/l for selected elements on melt composition. Good correlations can be found between the individual Damph/l and Xnf/X, which is a measure of melt polymerization defined on a molar basis as the sum of network-forming cations (Si + fraction of Al balanced by alkalies [Nielsen, 1985]) over the total amount of cations. The regression equations describing these trends are reported in Table 3.

Figure 4.

Variation of Damph/l as a function of melt composition (expressed by Xnf/X): (a) selected REE, (b) actinides, and (c) Pb. Regression equations in Table 3.

Table 3. Regression Equations Showing the Variation of the Damph/l as a Function of Melt Composition (Expressed as Xnf/X)a
  1. a

    Standard deviations are reported in parentheses.

LaLn(DLaamph/l) = −7.8 (0.6) + 10 (1) Xnf/X
SmLn(DSmamph/l) = −7.0 (0.5) + 12 (1) Xnf/X
YbLn(DYbamph/l) = −5.8 (0.6) + 9.5 (1.0) Xnf/X
PbLn(DPbamph/l) = −7.6 (0.7) + 8.4 (1.2) Xnf/X
ULn(DUamph/l) = −11 (1) + 11 (2) Xnf/X
ThLn(DThamph/l) = −11 (1) + 11 (2) Xnf/X

[18] As the crystal chemical mechanism ruling REE incorporation into M4′ in pargasite and kaersutite is the local charge balance for the presence of Al in the tetrahedral sites, some dependence of DREEamph/l on the [4]Al content is expected. However, its small range of variation within this experimental set precludes a quantitative prediction. No significant correlation between the DREEamph/l with Mg # and with Na # could be found. Thus the silica content in the melt is the main compositional factor ruling the variation of Damph/l for REE in this data set. The effect of the cummingtonite component on HREE incorporation can be appreciated only when dividing the data into subsets with nearly constant silica content in the melt: each subset shows positive correlation between HREE and the abundance of the cummingtonite component in the amphibole.

[19] The amount of DREEamph/l fractionation is slightly variable: DLaamph/l/DYbamph/l and DGdamph/l/DYbamph/l range between 0.125 and 0.257 and 1.04 and 1.84, respectively. A quantitative model of these variations is not straightforward. Tiepolo et al. [2000a] proposed a model to predict DNbamph/l/DTaamph/l fractionation from the dimensions of the M1 site in the frame of the elastic-strain theory of Blundy and Wood [1994]. They also showed that the effect of elastic properties of the site (i.e., Young's modulus) may be assumed to be negligible when considering a homogeneous group of amphiboles such as pargasites. When applying the same approach to REE, however, no significant correlation is observed between DLaamph/l/DYbamph/l (and DLaamph/l/DSmamph/l) and the overall M4 site dimension. Most probably, the incorporation of the REE into two distinct sites within the same structural cavity and the lack of information about the actual geometry of the single M4′ sites in which HREE occur preclude a quantitative prediction of REE fractionation. A reliable model of DLREEamph/l/DHREEamph/l might, however, be obtained by considering partitioning data for largely different chemical systems.

[20] This new set of DREEamph/l values is consistent with those previously reported in the literature [LaTourrette et al., 1995; Brenan et al., 1995; Klein et al., 1997; Adam and Green, 1994; Sisson, 1994] for calcic amphiboles. Consideration of the Damph/l for some REE reported by these authors allowed us to extend the correlation with the SiO2 content of the melt up to 80 wt %; the behavior of DYbamph/l is shown in Figure 5. Compatibility for Yb (and other HREE) in calcic amphiboles is reached at SiO2 content close to 50 wt %; the slightly gentler slope shown by amphiboles crystallized at SiO2 content >60.0 wt % may be ascribed to a change in their major element composition from pargasite and kaersutite to hornblende. Changes in the A site composition from full occupancy to vacancy should increase REE incorporation, whereas lowering [4]Al contents from 2 to 1 apfu should decrease it. The trend in Figure 5 shows the latter is the driving crystal chemical mechanism.

Figure 5.

Variation of DYbamph/l values as a function of the SiO2 content of the melt for the samples of this work (red circles) and for previously published data (blue symbols).

3.2. U and Th

[21] U and Th are highly incompatible in amphibole. Damph/l for U and Th vary by ∼1 order of magnitude, ranging from 0.004 to 0.034 for U and from 0.003 to 0.033 for Th. The two data points reported by Brenan et al. [1995] and LaTourrette et al. [1995] are in agreement with our data. Similar to the REE, the variation of Damph/l for U and Th correlate with Xnf/X (Figure 4b), whereas no significant correlation with the amphibole composition exists. The regression equations are reported in Table 3.

[22] No direct information is available so far about Th and U site preference in amphibole. However, on the basis of its ionic radius, Th4+ should enter the M4 cavity ([8]rTh 1.05 Å and [6]rTh = 0.94 Å [Shannon, 1976] compared to structure-refined M4 site size of 1.10–1.11 Å). U may assume different valences (4+, 5+, and 6+), and hence different ionic radii, depending on the oxygen fugacity. Although the valence of U in our set of experiments cannot be directly inferred, U4+ should be the dominant species at oxygen fugacity around 2 log units below FMQ [Wood et al., 1999]. This conclusion is also supported by the strong similarity between DThamph/l and DUamph/l values found for all the available data (Figure 6). The range of DUamph/l/DThamph/l values is generally rather small (0.67–0.96), and only three samples have values greater than unity. In analogy to what is observed for clinopyroxene [Lundstrom et al., 1994], a change in the oxidation state of U would imply changes in DU/Thamph/l around 1 order of magnitude. The ionic radius of U4+ ([8]rU4+ = 1.00 Å and [6]rU4+ = 0.89 Å) is more similar to those of HREE, and so it should occur mainly at the [6+2]M4′ site, whereas Th should partition between the two distinct sites within the M4 cavity with a preference for [6+2]M4′, similar to the behavior of the MREE [Bottazzi et al., 1999]. Because of their differing distributions between M4 and M4′ within the same polyhedron, the averaged site dimension cannot show any easily recognizable relationship with DUamph/l/DThamph/l. Similar to the HREE, the relatively small size of the M2 octahedral site in pargasite and kaersutite should preclude incorporation of Th and U into the M2 site.

Figure 6.

DThamph/l and DUamph/l for the samples of this work (red circles) and for previously published data (blue circle and triangle). Note slope shows DUamph/l/DThamph/l < 1.

3.3. Pb

[23] The higher compatibility of Pb relative to U and Th is due to its incorporation in the amphibole structure according to a different crystal chemical mechanism. Pb enters the A site in amphiboles and can even be a major constituent in joesmithite, ideally PbCa2Mg3Fe23+Si6Be2O22(OH)2. Pb is also incorporated into the A site at the trace element level, as indicated by abundances in amphibole 1 order of magnitude higher than in clinopyroxene, in which a twelvefold coordinated site analogous to the A site in amphibole is lacking. High [4]Al contents will also promote Pb incorporation in amphibole owing to the effects of [4]Al on charge balancing, and thus far lower DPbamph/l values are expected in richterites because of their much lower Al contents.

[24] DPbamph/l in the pargasites and kaersutites of this work range between 0.032 and 0.173, consistent with those reported by Brenan et al. [1995] and by LaTourrette et al. [1995]. The variation in DPbamph/l is a function of melt composition (Figure 4c and Table 3). The slope reported for DPbamph/l is smaller than those of REE, U, and Th. If the slopes in Table 3 are plotted against charge/ionic radius (Z/r) ratios, a positive correlation is observed (Figure 7), in agreement with the prediction that the higher the Z/r ratio of an element the higher its incompatibility in a polymerized melt [Watson, 1976].

Figure 7.

The relationship between the rate of increasing compatibility with melt composition (slopes in Fig. 2) and the Z/r ratio of the element. The coordination numbers taken into account are those compatible with the results of the present work.

4. Petrogenetic Implications

[25] The partitioning data presented in this work are particularly relevant for petrogenetic interpretation of upper mantle processes for three main reasons: (1) they provide systematic partitioning data for a large number of trace elements over a compositional range (43–56 wt % SiO2) that was not represented by previously available data (Figure 5), and give DREEamph/l for all REE of relevance in a range of compositions representative of natural glasses; (2) they show that HREE may be compatible in amphibole at melts containing around 50 wt % SiO2; (3) they show that DUamph/l/DThamph/l are generally lower than unity.

[26] The analogy in the partitioning behavior observed for HREE, U, and Th in the amphiboles of this work and in clinopyroxenes crystallized under conditions close to near-solidus melting of spinel lherzolite [Blundy et al., 1998; Wood et al., 1999] is not surprising. REE, U, and Th are incorporated in the two minerals in analogous sites (M4amph and M2cpx, respectively) according to the same crystal chemical mechanism. Nevertheless, the greater flexibility of the amphibole structure may account for DREEamph/cpx up to 4 [Vannucci et al., 1995; Tiepolo et al., 2000a]. Thus in SiO2-saturated melts the crystallization of amphibole has a stronger control than that of the clinopyroxene on the REE signature of the equilibrium liquid.

[27] Whereas melt flow along fractures, either as magma conduits or smaller scale vein systems, is known to be an important transport mechanism in the upper mantle [Spera, 1987; Nicolas, 1986, 1989; Spence and Turcotte, 1990], pervasive porous flow, often coupled with mineralogical reactions and porosity variations, has also been recognized as an important factor in the modification of the composition of both mantle melts and host peridotites [Bodinier et al., 1991; Kelemen et al., 1992; Takazawa et al., 1992; Bedini et al., 1997]. The present data suggest that amphibole crystallization during melt uplift could produce significant LREE enrichment and HREE depletion in the equilibrium liquid when SiO2 content exceeds 50 wt %. This latter value can be attained by amphibole and olivine crystallization, by high-pressure crystallization of orthopyroxene and spinel [Liu and Presnall, 2000], or by reaction of the infiltrating alkaline melt with the host peridotite in orthopyroxene-consuming reactions [Kelemen et al., 1990; Zinngrebe and Foley, 1995; Wulff-Pedersen et al., 1999].

4.1. Pervasive Porous Flow

[28] The plate model of Vernières et al. [1997] has been applied in an attempt to simulate the trace element transfer in the upper mantle during reactive porous flow involving amphibole crystallization. In this model, spatial-temporal constraints are released, and distinct reactions along a single mantle column are considered to simulate reaction zones at decreasing depth.

[29] The simulated case study is reactive porous flow occurring at the base of the lithospheric mantle just above the lower limit of the spinel facies due to melt infiltration from a plume, similar to that described by Bedini et al. [1997]. We have modeled percolation of both SiO2-undersaturated and SiO2-saturated alkaline melts in order to illustrate the effects of the different amphibole compositions on the REE signature under these conditions. The difference between the calculations for these two types of melt is due entirely to the effect of amphibole and melt composition on the trace element partition coefficients as determined in this study. Amphibole and orthopyroxene are the subsolidus products of the reaction between a SiO2-saturated melt and an originally anhydrous spinel-lherzolite [Sen and Dunn, 1994]. The same reaction was extended to SiO2-undersaturated systems by arbitrarily assuming that modal proportions of the reacting phases remain constant in order to appreciate the different role exerted by amphibole as a function of system composition. This model holds only at relatively high pressure (>1.5 GPa) but not at lower pressure where orthopyroxene is not stable and is replaced by Amph + Cpx + Ol. Spinel was neglected in the numerical simulation owing to its low modal fraction and its low capacity for incorporation of REE [Nagasawa et al., 1980; Horn et al., 1994]. Reactants and products have been calculated from the coefficients of (3) of Sen and Dunn [1994].

[30] The plate model was run with 50 increments and considering a reaction domain with a unique reaction zone divided into 50 cells. An anhydrous spinel-bearing lherzolite with 62% olivine, 21% orthopyroxene, 14% clinopyroxene and 3% spinel (similar to the average given by McDonough and Rudnick [1998]) and the trace element composition typical of the depleted mantle (Geochemical Earth Reference Model (GERM) website, available at; see Figure 10) were used as representative of unmetasomatized mantle composition. The trace element composition of the infiltrating three-dimensional oceanic island basalt (OIB) melt (taken from Sun and McDonough [1989]) was kept constant in both simulations, and Damph/l partition coefficients of samples 16 and 9 were considered as examples of SiO2-undersaturated and SiO2-saturated melts, respectively. Olivine/liquid and clinopyroxene/liquid partition coefficients are from Tiepolo et al. [1999] and Tiepolo et al. [2000a], respectively; orthopyroxene/liquid partition coefficients are from Kennedy et al. [1993].

Figure 10.

Variations in REE composition (normalized to C1 chondrite [Anders and Ebihara, 1982]) of the peridotite matrix after the fiftieth increment during numerical simulation of porous melt flow by the plate model of Vernières et al. [1997]. Results are shown separately for (a) SiO2-undersaturated melt and (b) SiO2-saturated melt. REE patterns are reported for intervals of five cells from the unreacted matrix (red dashed line) and varying positions in the column by colors as shown on the scale, in which cell numbers cross reference to Figure 8. Depleted mantle values (red dashed line) are taken from the GERM website (available on the World Wide Web at; values in ppm are La, 0.080352; Ce, 0.5376; Nd, 0.7375; Sm, 0.3045; Eu, 0.11858; Gd, 0.42976; Dy, 0.55942; Y, 3.655; Er, 0.38106; Yb, 0.39249.

[31] Variations of the modal composition of the host peridotite after the fiftieth increment in the different cells of the reaction column are reported in Figure 8. The lowest part of the column contains the most reacted peridotite; clinopyroxene is completely resorbed and replaced by amphibole (up to 19 wt %) and orthopyroxene (up to 35 wt %). The originally anhydrous Sp-bearing lherzolite column, now represented only by cell 50, is progressively transformed into an amphibole-bearing Sp-lherzolite in the upper part (cells 30–49) and into an amphibole-bearing harzburgite in the lower part of the column (cells 1–18). The REE patterns of the interstitial melts from the base (cell 1) to top (cell 50) of the column are reported in Figures 9a (low SiO2 content) and 9b (high SiO2 content). In contrast to the host peridotite, interstitial melts in the higher cells are those mostly reacted with the host peridotite because they have interacted with the entire mantle column. In Figure 9a (the low-SiO2 system) the REE patterns do not differ significantly from that of the starting OIB except for the uppermost cells; in particular, HREE increase and three-dimensional medium REE (MREE) tend to decrease toward the top of the column. Thus the REE patterns of the more reacted interstitial melts are flatter than that of the OIB in the MREE to HREE region. A marked decrease of the LREE contents is observed only in the highest cells, where the La/Yb ratio lowers to 4 (from 11 in the OIB). In Figure 9b (the high-SiO2 system) a completely different evolution of the interstitial melt is observed. In the lowermost sections of the column, HREE are strongly depleted due to their high compatibility in amphibole, and the patterns are convex upward. In the uppermost sections, interstitial melts show a nearly flat HREE pattern. Owing to the steady depletion in LREE the La/Yb ratio decreases continuously from the bottom to the top of the column down to values ∼4 times lower than that of the starting oceanic island basalt (OIB) (from 11 to 3).

Figure 8.

Numerical simulation of porous melt flow by the plate model of Vernières et al. [1997]. Variation in the modal proportion (wt %) of coexisting minerals are shown as a consequence of the infiltration of melt with the trace element pattern of an OIB through an anhydrous spinel-lherzolite mantle column. Calculation was run for 50 steps and 2% porosity. Reaction progresses from anhydrous lherzolite (top) through amphibole-bearing lherzolite to amphibole-harzburgite (bottom).

Figure 9.

Variations in REE composition of interstitial liquids during numerical simulation of porous melt flow by the plate model of Vernières et al. [1997]. (a) Percolation of SiO2-undersaturated melt; (b) percolation of SiO2-saturated melts. REE patterns are reported for intervals of five cells from the beginning of reaction by colors as shown on the scale, in which cell numbers cross reference to Fig. 8. Normalized to C1 chondrite using values of Anders and Ebihara [1982].

[32] The evolution of the REE pattern of the host peridotite is reported in Figures 10a and 10b for the cases of SiO2-undersaturated and SiO2-saturated percolating melts, respectively. In Figure 10a the host peridotite is progressively enriched in LREE, whereas HREE are only slightly affected by the pervasive porous flow. The harzburgites at the bottom of the column have La/Yb ratios ∼5 times higher than the depleted mantle, whereas the Amph-bearing spinel lherzolites at the top of the column have La/Yb ratios similar to the starting ones. In Figure 10b, harzburgites are characterized by a strong LREE enrichment over HREE (La/Yb is ∼1 order of magnitude higher than in the depleted mantle) and by higher REE contents. In the less reacted cells (from 5 to 50), MREE and HREE tend to the concentrations of the depleted mantle, whereas going from cell 1 to 50 LREE decrease more slowly, leading to a constant decrease of the La/Yb ratio, which in the last cell (50) is close to that of the depleted mantle.

[33] It should be noted that the calculations presented here oversimplify natural conditions in that they completely neglect possibly significant complicating effects such as the variation of partition coefficients with changing mineral composition and changing reaction stoichiometry between cells of the reacting column. Nevertheless, the results are a useful first-order approximation to show that amphibole crystallization during reactive porous flow may lead to radically different REE compositions of both interstitial melts and host peridotite, depending on the SiO2 content of the percolating melt. In particular, amphibole crystallization during percolation of SiO2-saturated melts through mantle sectors may explain the extreme variability of the La/Yb ratio and of the overall REE concentrations which characterize interstitial melts and metasomatized mantle xenoliths [Wulff-Pedersen et al., 1996]. Moreover, our modeling allows reproduction of the REE pattern observed in metasomatized spinel-lherzolites [Bodinier et al., 1988; Fabriès et al., 1989] and explanation of the HREE enrichment during metasomatic events.

[34] The effect of amphibole crystallization on the REE budget is less marked in the case of reactive porous flow of SiO2-undersaturated melts. With the exception of the most reacted ones, they do not significantly change the La/Yb ratio from that of the starting OIB. However, a LREE enrichment is appreciable in host peridotite along the whole mantle column, similar to the case of reactive porous flow involving the crystallization of clinopyroxene.

4.2. Magma Conduits or Vein Systems

[35] A second mechanism of melt migration in the upper mantle is along magma conduits or vein systems, with no or limited exchange with the host peridotite when uplift is faster than the reaction rate. The evolution of a melt due to amphibole crystallization may be numerically modeled by means of a simple Rayleigh fractional crystallization. The following numerical approach can be extrapolated also to magmatic sequences in which amphibole is the main crystallizing phase (e.g., high Al fractionation suites in which the first products are hornblendite and amphibole-gabbros [Arth et al., 1978; Drummond and Defant, 1990]).

[36] The REE pattern of residual liquids at different degrees of amphibole crystallization (from 10 to 90%) is reported in Figure 11 for high (>50 wt %) and low (<50 wt %) SiO2 contents of the initial melt. For illustrative purposes only, a melt with a constant composition equivalent to C1 chondrite for all REE is used as starting composition, and the Damph/l were kept constant for the whole processes. When SiO2 in the system exceeds 50%, the more rapid SiO2 enrichment of the residual melt would result in more rapid increase of the Damph/l and, in turn, a greater REE fractionation than in the simplified model used here. The difference between the low- and high-SiO2 series is straightforward. In low-SiO2 systems, residual liquids become progressively enriched in all REE owing to their overall incompatibility and slightly more enriched in LREE owing to their higher incompatibility with respect to HREE (Figure 11a). Nevertheless, as shown in the inset of Figure 11, this enrichment is quite limited due to the small difference between the Damph/l of LREE and HREE (at F, the fraction of melt remaining, equal to 0.1 the La/Yb and La/Sm ratios are ∼1.5). In high-SiO2 systems, REE fractionation after amphibole crystallization is strong because of the opposite behavior of LREE (incompatible) and HREE (compatible) toward incorporation into amphibole (Figure 11b). The more compatible behavior of MREE with respect to HREE leads to a convex downward pattern for the HREE, which is appreciable even at high F values. The curve describing the increase of the La/Yb of the primary melt as a function of the fraction of melt remaining (inset of Figure 11) is significantly steeper in high-SiO2 than in low-SiO2 melts, where residual solid fraction is 50%, and the La/Yb is ∼3.6 times higher than the starting values. At 90% fractional crystallization the LREE/HREE fractionation may even exceed 1 order of magnitude (La/Yb = 76.6).

Figure 11.

Evolution of C1 chondrite-normalized REE of melts produced by fractional crystallization of amphibole for initial melts with SiO2 (a) lower and (b) higher than 50 wt %. In both cases, melt evolution from a fictive initial melt composition equivalent to C1 is used to illustrate the very different evolution of REE patterns (progressively lighter colored lines) toward residual melts. F values indicate proportion of liquid remaining. The inset shows evolution of La/Yb for low- and high-SiO2 melts from initially equivalent values.

4.3. U, Th, and Pb

[37] The major implication of the new data for the partitioning of Pb and the actinides is that amphibole has only a minor influence on Pb isotope systematics and on the generation of U series disequilibria. Because of the low values of Damph/l for U and Th (∼1 order of magnitude lower than those of LREE) even at high SiO2 content, amphibole may play only a minor role in buffering actinides in the upper mantle. The DUamph/l/DThamph/l ratios lower than unity in most experiments preclude excess in 230Th activity relative to 238U during fractional crystallization or partial melting processes involving amphibole. Further investigations involving different kinds of amphiboles with significantly different M4 site composition (e.g., richterites) are in progress to unravel whether DUamph/l/DThamph/l ratios greater than unity and the inversion of the U/Th ratio observed in clinopyroxene [Wood et al., 1999] may occur also in amphibole. Melt generation by partial melting leaving amphibole in the solid residue should result in U and Th enrichment in the liquid relative to Pb and REE. Moreover, the higher Pb compatibility in amphibole with respect to U and Th may cause unradiogenic trends in 208Pb/204Pb, 207Pb/204Pb, and 206Pb/204Pb, as recently observed for phlogopite [Schmidt et al., 1999].

5. Concluding Remarks

[38] By means of new experimentally determined Damph/l for REE and numerical models we have shown that amphibole can significantly alter the original REE signature of a melt migrating in the upper mantle either by reactive porous flow or through magma conduits and vein systems. In contrast, amphibole is not capable of significantly altering the U-Th-Pb isotopic signature of the mantle. This new set of Damph/l values can also open new perspectives in the interpretation of trace element fractionation during slab melting under high thermal regimes, such as those occurring during Archean times [Martin, 1999]. Some first-order constraints on the role of amphibole in the trace element signatures of adakites were provided by LaTourrette et al. [1995] using partition coefficients derived from amphiboles crystallized from basanitic compositions. Despite uncertainties about the source composition and the melting model they were able to reproduce Sr/Y, Rb/Sr, K/Rb and LaN/LuN values within or very near the bounds observed in adakites but failed to simulate their typical concave upward HREE patterns. This work shows that in siliceous melts amphibole results in a decrease in the Gd/Yb ratio of the melt. This result coupled with the ability of amphibole to fractionate elements within the HFSE group [Tiepolo et al., 2000aTiepolo et al., 2000b], opens new perspectives in the interpretation of slab-derived melts.

Appendix: Appendix A

Table A1. Starting Material Composition
Alkali BasaltN-meltN-synthK-1.00K-0.81K-0.71M-0.45     
FeO tot9.7410.0310.0310.0310.0312.52
FeO tot12.6612.9712.9712.9715.558.463.750.0012.9712.9712.97
Table A2. Amphibole Composition, Unit Formulas Calculated on the Basis of 24 (O, OH, F, Cl) and Site Partitioning for Tia
Σ T8.
Fe3+ + Cr0.460.330.
Σ ((1,2,3)
Fe2+ + Mn0.
Σ (M4)
Σ (A)
Σ X2.
Mg #0.750.520.740.620.630.570.360.550.490.530.560.600.550.520.580.870.430.500.500.760.891.000.570.590.65
Na #0.750.710.760.990.880.780.700.790.790.770.770.740.780.790.750.720.780.760.730.750.810.810.990.840.67
Table A3. Average Compositions of Glasses in the Experimental Charges Obtained by EMP Analyses and SIMS (H2O)a
146.26±1.522.48±0.0515.96±0.529.37±0.314.39±0.586.37±0.241.37±0.271.93±0.116.15 97.86
Table A4. Average REE, Th, U, and Pb Contents (in ppm) in Amphiboles Obtained From SIMS Analysesa
  1. a

    Contents are in ppm. Standard deviations are calculated considering at least two different points within each experimental charge.

Table A5. Average REE, Th, U, and Pb Contents (in ppm) in Glasses Obtained From SIMS Analysesa
  1. a

    Contents are in ppm. Standard deviations are calculated considering at least two different points within each experiments charge. (SiO2 + the fraction of Al2O3 balanced by alkali)/(sum of all cations on a molar basis).

Mg #0.460.260.480.290.300.


[39] J.-L. Bodinier and J. Vernières are gratefully acknowledged for kindly providing the plate model software and advice on its usage. This paper greatly benefited from constructive comments by John C. Ayers and Markus Klein and the considerable editing efforts of Vincent Salters. Funding for this work was provided by the Consiglio Nazionale delle Ricerche to the CSCC and by the Ministero della Università e della Ricerca Scientifica e Tecnologica (project “Transformations in subducted materials and mass transfer to the mantle wedge”) to Riccardo Vannucci and by the Deutsche Forschungsgemeinschaft (grant Fo 181/9) to Steve Foley.