Trace element partitioning between pyrochlore, microlite, fersmite and silicate melts

We present experimentally determined trace element partition coefficients (D) between pyrochlore-group minerals (Ca2(Nb,Ta)2O6(O,F)), Ca fersmite (CaNb2O6), and silicate melts. Our data indicate that pyrochlores and fersmite are able to strongly fractionate trace elements during the evolution of SiO2-undersaturated magmas. Pyrochlore efficiently fractionates Zr and Hf from Nb and Ta, with DZr and DHf below or equal to unity, and DNb and DTa significantly above unity. We find that DTa pyrochlore-group mineral/silicate melt is always higher than DNb, which agrees with the HFSE partitioning of all other Ti–rich minerals such as perovskite, rutile, ilmenite or Fe-Ti spinel. Our experimental partition coefficients also show that, under oxidizing conditions, DTh is higher than corresponding DU and this implies that pyrochlore-group minerals may fractionate U and Th in silicate magmas. The rare earth element (REE) partition coefficients are around unity, only the light REE are compatible in pyrochlore-group minerals, which explains the high rare earth element concentrations in naturally occurring magmatic pyrochlores.


Introduction
To understand the behavior of trace elements in igneous rocks, trace element partition coefficients between minerals and melts are needed. Over the last decades, numerous experimental studies focused on the trace element partitioning between major rock forming minerals and melts [1], but few experimental partition coefficients are available for accessory phases such as rutile, ilmenite, spinel or apatite in basaltic compositions [2][3][4][5][6][7][8][9][10][11][12] and even less data are available for accessory mineral phases such as perovskite or pyrochlore in alkaline rock compositions [13]. As the aforementioned accessory mineral phases commonly occur in alkaline igneous rocks, they may exert a strong control on the trace element evolution of alkaline magmas.
The pyrochlore-group of minerals in silicate rocks and carbonatites encompasses a very large und very complex group of minerals [28]. The general formula of minerals of the pyrochlore supergroup is A 2 B 2 O 7 , where the A-site is often occupied by monovalent or divalent cations, and the B-site is mainly occupied by pentavalent cations such as Nb or Ta. Pyrochlore senso stricto is a mineral in which the B-site is occupied by Nb, and microlites are minerals in which the B-site is occupied by Ta [28]. We do not aim to describe the entire compositional variability of pyrochlores and related minerals in alkaline rocks, and we would like to refer the interested reader to excellent papers by Mitchell, Chakhmouradian, and others [18,21,22,26,[29][30][31][32][33]. However, common accessory mineral phases in carbonatites and alkaline silicate rocks are Ca-Na pyrochlores that can (and often do) incorporate almost all geochemical indicator elements, including the rare earth elements (REE), the large ion lithophile elements (e.g., K and Ba), and, as they are Nb-and Ta-rich minerals, also the high field strength elements Nb, Ta, Zr, and Hf (e.g., [21,22,26,30,31,33,34]. As there are, to our knowledge, no previous experimental studies on the partitioning of trace elements between pyrochlore-group minerals and melts, we set out to determine trace element partition coefficients between pyrochlore (Ca 2 Nb 2 O 7 ), microlite (Ca 2 Ta 2 O 7 ), fersmite (CaNb 2 O 6 ), and silicate melts. The long-term objective of this work is to determine the effects of temperature, pressure, and perhaps most importantly, chemical composition of minerals on trace element partition coefficients.
Hence future experiments will be extended towards systems with carbonate melts. Our experiments presented here were done in simplified chemical compositions, and must hence be considered as a first step towards a better understanding of trace element partitioning in complex natural systems where pyrochlore-group minerals occur.

Starting materials
To ensure the nucleation and growth of pyrochloregroup minerals and fersmite to a reasonable size and, thus, enabling in-situ analysis by laser ablation inductively-coupled plasma mass spectrometry (LA-ICP-MS), we conducted a few exploratory experiments to establish chemical compositions which precipitated pyrochlore-group minerals and fersmite. In order to avoid possible experimental problems related to Fe-loss, all starting materials did not contain Fe. Starting material SM_Pyro1 is a composition in the system CaO-Na 2 O-Al 2 O 3 -SiO 2 -Nb 2 O 5 -F, and HWM in the system CaO-Na 2 O-Al 2 O 3 -SiO 2 -Ta 2 O 5 . The choice of this rather simple chemical composition was made after our reconnaissance experiments in more complex systems yielded only small Nb-mineral crystals (< 15 µm), which were impossible to analyze with our LA-ICP-MS set-up. Both starting materials were prepared from analytical grade oxides, hydroxides, and carbonates (Table 1), which were ground in an agate mortar under ethanol. The starting material mixtures were heated in air at 1000 °C for 3 h in order to ensure complete decarbonation. Subsequently, the mixtures were heated to temperatures well above 1600 °C (10 min), i.e. above the liquidus in these systems, quenched, and the resulting glasses were re-ground to fine powders. All starting materials (except in run SE_ Pyro2, see below) were doped with a trace element mixture (SKM-TE1) containing the following elements: La, Ce, Nd, Sm, Gd, Dy, Yb, Lu, Y, Li, Sr, Ba, Rb, Sc, Mn, Ni, Co, Mo, Te, Ga, Ge, In, Cr, Zr, Nb, Hf, Ta, Pb. U and Th were added separately using 1000 µg/g ICP-AES standard (Alfa Aesar, in diluted nitric acid) solutions. The starting material compositions are given in Table 1. Note that run SE_Pyro2 was an exploratory run to which no trace elements were added. The doped starting materials were subsequently dried and denitrified in a laboratory-type drying cabinet at 110 °C (12 h).

Experimental techniques
All experiments were run in conventional 1-atmosphere vertical furnaces in air [11], using the Pt-wire loop technique (e.g., [11,13,[35][36][37]. To prepare the loops, we mixed about 20 mg of starting material powder into a viscous slurry with a synthetic glue, and we loaded this mixture onto a 0.1 mm diameter Pt-wire loop. The samples were then introduced into the hotspot of a vertical alumina tube furnace (Gero GmbH, Germany). Temperature was controlled with a thermocouple external to the alumina tube by a Eurotherm controller,  Table 2 for details). After quenching of the experimental charges, the samples were embedded in epoxy resin and polished using several different diamond pastes, and carbon coated for qualitative and quantitative analysis.

Analytical techniques
The experimental run products were examined with a JEOL6610LV scanning electron microscope with an EDX system and the major element concentrations of all phases were determined with a 5-spectrometer JEOL JXA 8530F electron microprobe analyzer (EMPA) at the Institut für Mineralogie at the University of Münster. Mineral phases were analyzed using beam spot sizes between 2-5 µm at a current of 15 nA. Counting times were 20 s on the peak and 10 s on the background. Glasses and quenched melts were analyzed with a 10 µm defocused beam and counting times were 5 s on the peak and 3 s on the background to minimize loss of volatile elements. Matrix-matched minerals were used as reference materials. Additionally, a set of secondary standards was measured together with the unknowns to monitor external precision and accuracy [38]. Trace elements were analyzed using a LA-ICP-MS system at the University of Münster that consisted of an Element 2 (ThermoFisher Scientific) SF-ICP-MS connected to a 193 nm ArF excimer laser ablation system (Teledyne Photon Machines Analyte G2). Laser repetition rate was 5 Hz using an energy density of about 4 J/cm 2 . Prior to sample analyses, the system was tuned with the NIST SRM 612 standard for high sensitivity, stability, and low oxide rates ( 232 Th 16 O/ 232 Th < 0.1%). Ablation time was 40 s and the background was measured for 20 s prior to sample ablation. Spot size was 12-25 µm for pyrochlores and fersmite crystals and melts, and 35 µm for the reference materials. 43 Ca was used for internal calibration and NIST SRM 612 was chosen as the external standard using concentration values given in the GeoRem database, version 26/2019 [39]. Groups of 8-17 spot analyses were bracketed by three standard analyses to monitor instrumental drift. All concentrations were calculated using the Glitter software (version 4.4.4 [40]. Standard reference glasses BCR2-G, BIR1-G, and BHVO2-G were analyzed as monitor for precision and accuracy for silicate phases during the course of this study. Obtained results match the published range of concentrations given in the Geo-ReM database [39].

Experimental results
The experiments (Table 2) yielded euhedral-to-subhedral pyrochlore and fersmite crystals from < 15 µm up to about 150 µm across. In some runs, pyrochlore or fersmite crystal were thin and acicular so that they could not be analyzed with LA-ICPMS (c.f., Table 2) No other crystals were observed. The melt quenched to a dark and inclusion-free homogeneous glass. Electron microprobe and LA-ICP-MS analyses of pyrochlore and fersmite crystals and quenched melts indicate major-and trace element homogeneity, which is taken as evidence for the attainment of equilibrium between crystals and melts in our runs. Representative run products are shown in Fig. 1.
We would like to note that the exploratory experiment SE_Pyro2 was run with a starting material that did not contain added trace elements. Consequently, the trace element concentrations in both minerals and quenched melts in this run are substantially lower than in all other experiments (Table 2). However, the partition coefficients of this run are in excellent agreement with D's from the other runs (see discussion below, Table 4), which

Table 2 Experimental conditions and run products
Fer=Ca-Nb-fersmite; Nb-pyr=Ca-Nb-pyrochlore; melt=quenched melt; Ta-pyr=Ca-Ta-microlite Note that in run HW3 no cooling rate was used. Note also that fersmite in runs SE_Pyro3, and pyrochlores in SE_Pyro2 were acicular and too small for EMPA analysis. All experiments were run in air and at atmospheric pressure. T/°C start: T at which the samples were inserted into the furnace, T/°C end: The temperature of the run after the cooling rate; cooling rate: The samples were cooled from T start to T end , and the runs were held at T end for at least  confirms the quality of our trace element analyses at very low concentrations, and also that Henry's law has been attained [41].

Analytical results
Major, minor and trace elements of all phases were analyzed using EPMA and LA-ICPMS techniques, and the analytical results are given in Tables 3 and 4. Note that some of the crystals (Table 2) were too small to be analyzed.

Partition coefficients
The trace element concentrations of pyrochlore, microlite, and fersmite crystals and quenched melts were used to calculate trace element partition coefficients (D i ) using the following expression: where D i is the partition coefficient for a trace element i, c i, min is the measured concentration of the trace element i in the crystal, and c i, melt is the concentration of the trace element i in the melt. The calculated trace element partition coefficients together with the propagated uncertainties, are given in Table 4, and are depicted in Fig. 2.
The light element Li is incompatible, with partition coefficients slightly less than unity. The divalent elements Sr and Mn partition evenly between pyrochlore, fersmite and silicate melt, with partition coefficients slightly less than unity, the D's for Ni and Co are slightly lower (Fig. 2, Table 4). The Ge partition coefficients are all slightly below or close to unity (Fig. 2, Table 4), indicating that Ge follows the geochemical behavior of Si. Sc and Ba partition coefficients cluster around 0.1, with no apparent systematics (Fig. 2).
The partition coefficients of the rare earth elements (REE and Y) are depicted in Fig. 3. Our results show that D REE pyrochlore/melt increases with decreasing ionic radius of the REE. The light REE (LREE) La, Ce, Pr, Sm and Gd are compatible in pyrochlore in all runs, regardless of temperature or bulk composition. The D REE pyrochlore/melt of the heavier REE (Yb, Lu, and also Y) are slightly below 1, in run HW3 all D REE are above 1. Our data shows that the D REE of Ta-rich pyrochlores (i.e. the HW1-HW3 runs, black filled symbols in Fig. 3) are systematically higher than the D REE from experiments with Nb-rich pyrochlores (runs SE3 and SE6, open triangles in Fig. 3). As the REE are incorporated on the Ca-site in pyrochlores, the different partitioning may be explained by the different nature of the Ca-site in Ta-rich (i.e. microlite) and Nb-rich pyrochlores, respectively. Crystallographic data show that the size of the Ca-site in Nbrich pyrochlore is slightly smaller (average Ca-O bond length in pyrochlore of 2.58 Å [42]) than the Ca-site in Ta-pyrochlore (microlite), with an average Ca-O bond length of 2.73 Å [43].
If we only consider the D REE of the Ta-rich microlites (i.e. the HW runs), we find that D REE of HW3, which was run at 1400 °C, is substantially higher than D's of runs HW2 and HW1 which were run at a final run temperature of 1300° and 1200 °C. However, the D's of HW1 and HW2 are undistinguishable within the errors and hence the temperature effect on D's cannot be fully confirmed.
The D REE fersmite/melt show a much flatter and only slightly downwards concave pattern, with D La / D Lu very close to unity, and D REE of the middle REE (Nd, Sm, Gd, Dy) slightly higher than the D LREE or D HREE , similar to D REE between apatite and melt (e.g., [44]. This clearly indicates that the nature of the Casite in fersmite is quite different to that of microlite or pyrochlore, and crystallographic data confirm that the  [42]). Furthermore, the Ca-site in pyrochlore is eightfold coordinated [42], but the Ca-site in fersmite is smaller and sixfold coordinated [45]. We find that D REE fersmite/melt are highest in run SE_Pyro2, which was run at the highest final run temperature of 1300 °C, whereas run SE_Pyro6, which was run at 1240 °C, resulted in the lowest D REE .
Our data clearly shows that Ce is much more compatible than neighboring REE in runs with the HW starting material, which does not contain F. We can explain the deviant behavior of Ce in these runs, as Ce is a multivalent trace element and under oxidizing conditions Ce

Table 3 Maj or element compositions of minerals and melts
Major element composition of minerals and quenched melts (Fer = Ca-Nb-fersmite, Nb-pyr = Ca-Nb-pyrochlore, melt = quenched melt, Ta-pyr = Ca-Ta-microlite) analyzed with EMPA. n number of analyses, n.d. not detected, n.a. not analyzed. Note that fersmite occurred in all runs with the SE_Pyro1 starting materials (Table 1) but in most runs these crystals were too small too be analyzed, and we only present data from runs SE_Pyro5 and SE_Pyro6. n.a n.a n.a n.a n.a      However, the D's (pyrochlore/melt) in runs with the SE-pyro starting material, which contains F, show no positive Ce-anomaly (Fig. 3), and D (fersmite/melt) with SE-pyro also show only a very weak positive Ce anomaly. We speculate that this behavior may be caused by F in the melt, which may form stable complexes with Ce 4+ , i.e. similar to CeF 8 4− or CeF 6 2− type compounds that have been reported in experiments with aqueous solutions [47]. These complexes in the melt would lead to an increased partitioning of Ce 4+ into the melt, which consequently lowers D Ce mineral/melt in F-bearing runs (Fig. 3).
As to the high field strength elements (HFSE), we find that the HFSE partition coefficients for Nb and Ta are above unity (Table 4), and other D HFSE (Th, Zr, Hf, U, Mo) vary with bulk composition and temperature (Fig. 4). D Ta is very similar to D Nb , but overall we find that D Ta are systematically higher than D Nb , which is in good agreement with previous data for perovskite, rutile and other Ti-rich oxide minerals [2,3,5,7,8,10,13,[48][49][50]. Our data shows that D HFSE for Ta-pyrochlore (HW1-3) are mostly above 1, only D Mo in the HW runs are systematically below 1. The Zr and Hf partition coefficients are close to 1, with D Zr and D Hf slightly higher than 1 for Tarich microlite, and D Zr and D Hf are slightly lower than 1 for Nb-pyrochlore and fersmite (Table 4).
This compatible in Ta-rich microlite with a partition coefficient of almost 10, c.f., Table 4). Although one   Ta-pyr), and Ca-Nb-pyrochlore (Nb-pyr)), and Ca-Nb-fersmite (fer), and melts. Minerals and quenched melts were analyzed with LA-ICPMS techniques. See text for details might expect U to exhibit similar partitioning behavior as Th, we find that U is slightly incompatible in pyrochlore, microlite and fersmite (Table 4). This can be readily explained by the oxidizing conditions prevailing in our 1-atm experiments, under which U ions occur mainly in the 6 + valence state and, thus, do not fit as well into the mineral structures as U 4+ would [8,10]. As Th 4+ (ionic radius of 0.94 Å) and U 6+ (with its ionic radius of 0.89 Å in octahedral coordination [46]) are too large to replace Nb or Ta (ionic radius of 0.64 Å [46]) in pyrochloregroup minerals, we presume that both elements partition into the Ca-sites (ionic radius of Ca is 1.0 Å in octahedral coordination), despite the large charge difference.

Implications
To interpret the trace element budget of alkaline-rich und undersaturated melts, one needs to understand the trace element partitioning of all mineral phases involved. In the next paragraph we will show some important aspects of how crystallizing pyrochlore-group minerals may affect coexisting melt compositions. Note that our experiments were done in silicate melt systems, and as no other data are available, we assume that our results are also applicable to carbonate melt systems.
Our data show that pyrochlores prefer to incorporate Th over U (D Th > D U ) under oxidizing conditions; hence crystallization of pyrochlore will deplete a melt in Th and enrich it in U. The composition of natural pyrochlores varies dramatically, but many primitive pyrochlores contain superchondritic Th/U (e.g., [51]), which may be caused by the aforementioned uneven partitioning of Th and U. If natural pyrochlores contain a lot more U than Th (e.g. [15]) then either the melt must have had exceptionally high U/Th [52], or, perhaps more common, the pyrochlores were precipitated from or altered by a hydrothermal fluid, which can transport U much more efficiently than Th [15]. A recent paper shows nicely how the U and Th composition of pyrochlore varies with increasing degree of alteration: Whereas primitive pyrochlores contain wt.% of Th and only little U, the U content of pyrochlore increases and the Th content decreases with increasing degree of hydrothermal alteration [51].
Carbonatite and undersaturated silicate rocks are often extremely enriched in REE, so that many carbonatites are mined for the REE. As pyrochlores prefer to incorporate the lighter REE over the heaver ones (D LREE > D HREE ), precipitation of pyrochlore minerals from a melt will lower the light REE concentrations (i.e., La, Ce, etc.) significantly, whereas the concentrations of the heaver REE (e.g., Yb, Lu) will not be depleted so much (D HREE ≈ 1). This may explain why primary bastnäsite, a Cecarbonate, is scarce in carbonatites [26], and only secondary hydrothermal alteration (and breakdown of pyrochlores) may lead to the formation of the formation of REE-carbonates.

Conclusions
We present hitherto unknown experimentally determined trace element partition coefficients between pyrochlore, microlite, fersmite, and silicate melts. The partition coefficients show that pyrochlore, microlite, and fersmite can fractionate U from Th during differentiation of a magma under oxidizing conditions. Furthermore, our REE partition coefficients reveal that pyrochlores prefer to incorporate the lighter REE over the heavy REE, and this may prevent the crystallization of REE-carbonates in many carbonatite melts. Fig. 4 High field strength element partition coefficients between pyrochlore-group minerals, (Ca-Ta-microlite (Ta-pyr), and Ca-Nb pyrochlore (Nb-pyr)), Ca-Nb fersmite (fer), and melts. Minerals and quenched melts were analyzed with LA-ICPMS techniques. See text for details