The neodymium stable isotope composition of the silicate Earth and chondrites

The full-text may be used and/or reproduced, and given to third parties in any format or medium, without prior permission or charge, for personal research or study, educational, or not-for-pro t purposes provided that: • a full bibliographic reference is made to the original source • a link is made to the metadata record in DRO • the full-text is not changed in any way The full-text must not be sold in any format or medium without the formal permission of the copyright holders. Please consult the full DRO policy for further details.


Introduction
The planets of our Solar System are thought to have originated from the accumulation of dust and gas in the young Sun's protoplanetary disk, in which case the initial composition of the ter-* Corresponding author. E-mail addresses: a.j. mccoy-west@durham.ac.uk, alex.mccoywest@gmail.com (A.J. McCoy-West). restrial planets should be similar to that of primitive 'chondritic' meteorites, fragments of material that escaped planetary differentiation. Although it has long been known that Earth's composition cannot be readily ascribed to any particular group of chondrites (e.g. Drake and Righter, 2002), the notion that the Earth has 'chondritic' refractory lithophile element ratios has persisted. The challenge to this assumption came from pioneering studies demonstrating that the 142 Nd/ 144 Nd isotope composition of chondritic meteorites is 18 ± 5 ppm lower than Earth's mantle  (Boyet and Carlson, 2005;Carlson et al., 2007). The measured difference in 142 Nd/ 144 Nd requires the Sm/Nd ratio of the silicate Earth to be ∼6% above the average chondritic value (known to within ±0.3%; Bouvier et al., 2008) providing definitive evidence that Earth's silicate mantle at the present day is non-chondritic. One potential explanation is that Earth's mantle may have experienced an early depletion event resulting in the formation of an incompatible element enriched reservoir with a low Sm/Nd ratio balancing that seen in the mantle at the present-day (Andreasen et al., 2008;Boyet and Carlson, 2005;Carlson et al., 2007). One possibility, is that this reservoir resides at the base of the mantle, in the seismically anomalous D layer overlying the core. Thus far, however, there is little chemical or thermal evidence for such a reservoir (Campbell and O'Neill, 2012). Moreover, because this reservoir must have been formed within the first ∼30 Myr following accretion, while 146 Sm was extant, it has been argued that it would have been destroyed by the Giant-Impact that formed the Moon (Caro and Bourdon, 2010). Alternatively, it has been suggested that the Earth was initially assembled from chondritic material and was subsequently modified during planetary construction through the removal of its early silicate crust through, so-called, collisional erosion (O'Neill and Palme, 2008). Nevertheless, collisional erosion implies that the heat-producing elements are depleted up to 50% of their chondritic values, implying unlikely cooling rates for Earth over geological history (e.g. Campbell and O'Neill, 2012).
An alternative deep reservoir within the Earth is the metallic core, however, experimental data suggests that neither Sm nor Nd are partitioned into Fe-Ni metal at the conditions of core formation, nor is Nd preferentially incorporated over Sm (Bouhifd et al., 2015). Nevertheless, it has long been known that Earth's liquid outer core must be alloyed with ∼10 wt.% 'light' elements (e.g. Birch, 1952) which may have been added late to the core during accretion. An oft cited candidate for that light element is sulfur (e.g. Boujibar et al., 2014;Labidi et al., 2013) and experimental data suggest that Sm and Nd, and other incompatible elements, may be substantially incorporated into sufide or sulfiderich metal (Wohlers and Wood, 2015). In particular, Wohlers and Wood (2015) have shown that the partition coefficients of U, Nd and Sm are strong functions of the FeO content of silicate melts, increasing dramatically (with D becoming >1) as the FeO content decreases below 1 wt.%. Furthermore, D Nd is always significantly greater than D Sm with D Nd /D Sm approaching 1.5 in some cases.
In this case, in a growing planet the segregation of a sulfide (or S-rich metal) from reduced FeO-poor silicate (conditions analogous to core formation) will lead to enrichment of the metallic phase in U and in Nd relative to Sm relative to the residual silicate mantle, providing a significant heat source to the core and a mantle with a superchondritic Sm/Nd ratio. If such a body represented Earth early in its history then the mantle would have developed a positive 142 Nd anomaly relative to chondrites (as observed) and much of the energy deficit identified for core convection (Labrosse et al., 2001) would be supplied by the additional U (and Th), while maintaining a chondritic complement of heat-producing elements for long-term heating of the Earth. This raises the possibility that Earth's missing refractory elements are simply located in a sulfide or sulfur-rich metal phase in the core, rather than being lost through collisional erosion or hidden in the deep mantle. However, this result has been questioned by more recent experimental work which indicates that there is little elemental fractionation between Sm and Nd at higher temperatures, closer to those that pertain to core formation (up to 2100 • C; Wohlers and Wood, 2017).
In parallel, recent high precision Nd and Sm isotope data indicate that, compared to chondrites, the Earth is enriched in Nd produced by the slow neutron capture process (s-process) of nucleosynthesis (Bouvier and Boyet, 2016;Burkhardt et al., 2016). This s-process excess leads to higher 142 Nd/ 144 Nd ratios, relative to chondrites. Therefore, the 142 Nd offset between the Earth and chondrites most likely reflects a higher proportion of s-process Nd in the Earth, rather than 146 Sm decay from a super-chondritic terrestrial reservoir (Bouvier and Boyet, 2016;Burkhardt et al., 2016;Saji et al., 2016). Nevertheless, the possibility remains that some part of the 142 Nd excess seen in the silicate Earth may be due to the presence of a hidden reservoir (up to 5 ± 2 ppm; Burkhardt et al., 2016). These different processes are not mutually exclusive and it is conceivable that the suprachondritic terrestrial 142 Nd/ 144 Nd is a consequence of some combination of nucleosynthetic processes, collisional erosion, and partitioning of Nd into the core.
If there is indeed a sulfide-rich reservoir in the deep Earth that has contributed to the 142 Nd discrepancy between chondrites and the terrestrial mantle, then Nd stable isotopes have the potential to trace this sulfide segregation event. Theory predicts that equilibrium stable isotope fractionation between silicate material (such as the mantle) and metal or sulfide (such as the core, depending on its S content) is driven by contrasts in bonding environment and oxidation state (e.g. Polyakov, 2009;Rustad and Yin, 2009). There is a significant contrast in bonding environment between sulfide and silicate, therefore heavy isotopes should be preferentially incorporated into high force-constant bonds involving rare earth element (REE) 3+ ions in silicate minerals. Preliminary measurements by Andreasen and Lapen (2015) indicate that mantle rocks may indeed possess heavier Nd isotope compositions than chondritic meteorites, consistent with the removal of light Nd into sulfide in the core, driving the residual mantle to heavier values. This study presents high-precision double spike stable Nd isotopic data for chondritic meteorites and terrestrial samples to assess the extent of fractionation between the silicate Earth and chondrites.

Sample preparation and Nd separation
The samples investigated during this study include eleven carbonaceous chondrites (CI, CM, CO, CK, CV), seven enstatite chondrites (EH, EL) and fifteen ordinary chondrites (H, L, LL). For comparison a range of terrestrial materials were analysed and these include: twelve international rock standards that range from ultramafic to rhyolitic in composition; four optically pristine midocean ridge basalt (MORB) glasses and four abyssal peridotites from the Garrett Fracture Zone, on the East Pacific Rise (Niu and Hékinian, 1997;Wendt et al., 1999); and two spinel lherzolite xenoliths from Kilbourne Hole, New Mexico (Burton et al., 1999;Jagoutz et al., 1980). The majority of samples analysed herein were obtained as powders. When new powders where required samples were coarse crushed and unaltered chips, without visible fusion crust, were selected and then ultrasonically cleaned in distilled water and subsequently powdered using an agate mortar and pestle. Between 0.1 and 1 g of sample was taken to obtain between 100 and 200 ng of natural Nd and then spiked with our 145 Nd-150 Nd double spike, comprising 29% 145 Nd and 66% 150 Nd, with an ideal sample to spike ratio of 60:40 based on the calculations given in Rudge et al. (2009;Fig. A1). Two digestion methods were employed during this study: 1) samples were placed in 15 mL Savillex beakers and dissolved using a concentrated HF-HNO 3 (3:1) mixture on a hotplate at ≥130 • C for 72 h; or 2) samples were placed in Carius tubes with a concentrated HCl-HNO 3 (5:4) mixture and sealed and heated at 220 • C for four days, following cooling all of the sample material was extracted and subsequently underwent a standard HF-HNO 3 dissolution to dissolve the silicate fraction. Carius tube digestions were implemented to ensure the complete digestion of samples containing refractory phases, such as the Cr-spinel present in some mantle samples (see Table ES1). Following dissolution the samples were sequentially brought into solution in concentrated HNO 3 and HCl and refluxed on a hotplate for at least 24 h and dried down prior to being brought completely into solution in dilute HCl.
Neodymium was separated following well-established chromatographic techniques. The REE were separated from the sample matrix using polypropylene R1040 columns (2.5 mL resin capacity) filled with BioRad AG50W-x8 cation exchange resin. Samples were loaded in 2 mL of 1 M HCl, the matrix was then sequentially removed in 10 mL 1 M HCl + 1M HF, 12 mL 2.5 M HCl and 8 mL 2 M HNO 3 with the REE fraction collected in 14 mL of 6 M HCl. The REE fraction was then dried and Nd was separated from the other REE using polypropylene columns (internal diameter 88 × 4 mm) filled with Eichrom Ln-Spec resin. Samples were loaded in 0.5 mL of 0.2 M HCl, another 6 mL of 0.2 M HCl was then eluted, prior to the collection of Nd in the next 6 mL of 0.2 M HCl. This separation protocol results in virtually perfect separation of Sm (isobaric interferences on 144 Nd, 148 Nd, and 150 Nd), however, residual Ce (isobaric interference on 142 Nd) is present, and thus 142 Nd/ 144 Nd ratios are not reported here. Total procedural blanks measured by isotope dilution with every batch of samples vary from 3-18 pg (n = 6) and in all cases were negligible. To confirm the chromatographic procedure had no effect on the stable Nd isotopic ratios repeat aliquots of the JNdi-1 standard were passed through the complete chemical procedure and the δ 146 Nd value obtained is indistinguishable within error from the unprocessed aliquots (Fig. A2).

Mass spectrometry and data reduction
Neodymium isotope measurements were performed using a Thermo-Fisher TritonPlus thermal ionisation mass spectrometer (TIMS) in the Arthur Holmes Geochemistry Labs at Durham University. In preparation for loading and TIMS analysis the solutions were evaporated to dryness and Nd samples individually loaded using 1 μL of 16 M HNO 3 onto Re ionisation filaments using a double filament assembly. Neodymium was measured as a metal ion in static collection mode using eight faraday cups using the fol- The double spike deconvolution used in this study is based on the algebraic resolution method used by Millet and Dauphas (2014). These calculations were confirmed using the Isospike deconvolution program of Creech and Paul (2015) which is based on the algebraic equations presented in Rudge et al. (2009), and the geometric iterative resolution method of Siebert et al. (2001), all three approaches yield identical results within analytical error (with a maximum offset ≤4 ppm). As the radiogenic isotope 143 Nd is not used during double spike deconvolution, the spike proportion, and the geological and analytical fractionation factors resolved during deconvolution can be used to calculate the 143 Nd/ 144 Nd ratios of the samples. Long-term reproducibility of 100-200 ng aliquots of the JNdi-1 standard throughout the period of analysis was δ 146 Nd = 0.003 ± 0.017h and 143 Nd/ 144 Nd = 0.512100 ± 8 (±2σ ; n = 39; Fig. A2). The external reproducibility of five digestions of the USGS rock standard BHVO-1 is δ 146 Nd = −0.029 ± 0.014h and 143 Nd/ 144 Nd = 0.512972 ± 9 (n = 11), and four replicates of the Allende meteorite give δ 146 Nd = 0.057 ± 0.017h and 143 Nd/ 144 Nd = 0.512525 ± 13. Taken together, these data suggest that the long-term reproducibility of the δ 146 Nd measurements is ±0.017h or better. Due to the novel nature of these measurements verifying the accuracy of our δ 146 Nd measurements is difficult. However, the 143 Nd/ 144 Nd obtained herein following double spike deconvolution can be compared directly to previously published values for the same samples, with both techniques agreeing within analytical error (Fig. 1).

Neodymium in chondrites
A general increase in Nd concentration is observed across the different classes of chondritic meteorites from enstatite (0.42-0.61 ppm), to ordinary (0.58-0.85 ppm), to carbonaceous (0.84-1.08 ppm; Fig. 2a; Table 1), consistent with previously recognised trends regarding REE in chondrites (e.g. Nakamura, 1974). Two carbonaceous chondrites Orgueil and Cold Bokkeveld have significantly lower Nd concentrations (Nd = 0.58 and 0.46 ppm, respectively) than the other carbonaceous samples, and three desert finds Happy Canyon, Estacado and Maralinga (Nd = 0.98, 1.34 and 24.1 ppm, respectively) have significantly higher Nd than might be expected. The majority of the chondrites, irrespective of classification, fall within an extremely narrow range in radiogenic Nd isotope compositions with 143 Nd/ 144 Nd varying from 0.5125 to 0.5128 (n = 35; Fig. 2b). Although, the three finds with higher than expected Nd concentrations also exhibit variable Nd isotope ratios Stable Nd isotope variations in chondrites range from δ 146 Nd −0.088h to 0.055h although the majority of the samples possess a significantly more restricted range of compositions ( Fig. 3; Table 1). Of the samples analysed here, enstatite chondrites are the most homogeneous chondrite class ( 146 Nd = 26 ppm) with δ 146 Nd ranging from −0.039h to −0.013h and an average composition of δ 146 Nd = 0.023 ± 0.015h (±2σ ; n = 10). Carbonaceous chondrites span the entire range of observed δ 146 Nd values in chondrites, and even within the CV3 group ( 146 Nd = 117 ppm) two well-characterised meteorites have distinctly different δ 146 Nd values (Vigarano = 0.029h; Allende = −0.057h; Table 1). The average composition of carbonaceous chondrites is δ 146 Nd = −0.026 ± 0.025h (n = 9), excluding all of the CV3 meteorites and the desert find Maralinga. Ordinary H-group chondrites (δ 146 Nd = −0.051h to −0.025h) appear to be systematically lighter than the L-group chondrites (δ 146 Nd = −0.021h to 0.006h; Fig. 3), excluding the desert find Waconda which is significantly offset ( 146 Nd = 58 ppm) from the other L-group samples. The average composition of all ordinary chondrites is δ 146 Nd = −0.025 ± 0.030h (n = 20; excluding Waconda). All three major classes of chondrites (carbonaceous, enstatite and ordinary) have mean δ 146 Nd values that are within error at the 95% confidence level (Table 1), therefore it is possible to calculate an overall chondritic average of δ 146 Nd = −0.025 ± 0.025h (n = 39; 95% s.e. = ±0.004h).

Stable Nd in terrestrial materials
A range of international rock standards that includes intrusive and extrusive rocks, that vary from basaltic to dacitic in composition, show limited stable Nd isotope variability with δ 146 Nd ranging from −0.046h to −0.011h ( 146 Nd = 35 ppm; Fig. 4; Table 2). Two samples, however, with rhyolitic bulk compositions display systematically heavier δ 146 Nd compositions with increasing SiO 2 (RGM-1 = −0.008h; JG-2 = 0.013h; Fig. 5a). The average composition of the terrestrial magmatic rock standards is δ 146 Nd = −0.029 ± 0.013h (n = 18, excluding the two samples with SiO 2 > 70 wt%). Four MORB glasses from the Garrett Fracture Zone show comparable δ 146 Nd to the standards with values ranging from −0.030h to −0.007h and a slightly heavier average composition of δ 146 Nd = −0.019 ± 0.019h (n = 4). While they are comparatively uniform in their major element compositions, mantle samples possess significantly more variable stable Nd ( 146 Nd = 75 ppm; Fig. 5b) than the other terrestrial rocks measured here, with δ 146 Nd ranging from −0.044h to 0.031h. A significant degree of variability is even observed within a single lithology at the same location, with serpentinised harzburgites from the Garrett Fracture Zone ( 146 Nd = 48 ppm) and spinel lherzolites from Kilbourne Hole ( 146 Nd = 46 ppm) showing large resolvable differences. Although highly variable, an average composition for the mantle based on these analyses can be calculated as δ 146 Nd = −0.008 ± 0.053h (n = 8; 95% s.e. = ±0.022h). Terrestrial reservoirs have broadly similar Nd isotopic compositions allowing an average composition of the bulk silicate Earth to be calculated with δ 146 Nd = −0.022 ± 0.034h (n = 30; 95% s.e. = ±0.006h; Fig. 4).

Previous stable Nd isotope measurements
Stable Nd isotope geochemistry is in its infancy with only a few published measurements of mass-dependent Nd isotopic variations (Ma et al., 2013;Saji et al., 2016;Wakaki and Tanaka, 2012). Wakaki and Tanaka (2012) (Fig. A3), this may result from an incomplete yield during chemical separation with heavy Nd isotopes being lost during Ce removal (see Fig. 1 in Ma et al., 2013) driving the residue to lighter values (as observed), thus we suggest that these data should be considered with caution. Most recently, Saji et al. (2016) used a MC-ICP-MS technique but developed an automated chromatographic procedure using a double-sized Ln-spec column and claim better than 99.5% Nd recovery effectively eliminating problems related to fraction- ation during chemistry. They present results in ε 145 Nd of 3 rock standards (BCR-2, BHVO-2 and GSP-2) and the Allende meteorite and conclude these materials are the same within analytical error (Fig. A3). These measurements are clumped averages of 10 mass spectrometry analyses from 5 different aliquots of the same digestion processed through chemistry (n = 50), with 95% confidence intervals varying from 8 to 31 ppm. The average compositions of 5 different aliquots from the same digestion of BCR-2 are highly variable with 146 Nd = 60 ppm (Fig. A3). It is impossible to verify if the large variability in δ 146 Nd seen in different aliquots of the same digestion of homogeneous rock standards is an artefact from the chemical separation or simply a reflection of the true precision of plasma source technique of Saji et al. (2016). Using the double spike-TIMS technique implemented here 11 measurements of 5 digestions of the rock standard BHVO-1 have a 95% s.e. of 5 ppm and a 146 Nd of 19 ppm. By comparison, using the average of averages approach implemented in Saji et al. (2016), the 146 Nd obtained here is just 7 ppm with the external reproducibility (±2σ ) falling to just 6 ppm. Taken together, these results indicate that the isotope measurements of δ 146 Nd obtained using the double spike TIMS technique here provides data of the best accuracy and precision currently available.

Neodymium isotope variability in chondritic meteorites
The three major classes of chondritic meteorites all have mean δ 146 Nd values within error of each other at the 95% confidence

Terrestrial weathering
The majority of meteorites analysed here are observed falls, with eight samples being hot desert finds (Table 1: Estacado, Gabarato, Happy Canyon, Long Island, Maralinga, NWA1277, NWA2086 and Waconda), though all of the samples may have been subjected to some degree of terrestrial weathering. Exposure to a range of potential weathering agents including water, oxygenrich air, salts and wind may have modified the original δ 146 Nd composition of some of the chondrites. Through analyses of fragments of the Holbrook meteorite collected in the Arizona desert over a 56 year period, Gibson and Bogard (1978) showed that Rb and Sr increased at least two-fold during weathering. Al-Kathiri et al. (2005) analysed up to 50 ka old meteorites collected from the Omani desert and observed a significant increases in Sr and Ba contents during troilite weathering, while pyroxene and olivine remained visibly unaltered in this weathering environment. Significant increases in Sr and Ba concentrations (up to 244 and 300 ppm, respectively; see Table ES3) are observed in both fall and find samples in the meteorite fractions analysed here (Fig. 6a).
Due to the inherent heterogeneity seen between different classes of chondrites, conservatively samples with chondrite normalised ratios <1.5 are considered unaltered. This leaves 14 of the samples analysed here that may have experienced some degree of terrestrial modification, however, no systematic change in δ 146 Nd is seen with increasing Ba/Yb (N) (Fig. 6b), suggesting that alteration does not measurably perturb stable Nd isotopic compositions. Three of the find samples (Maralinga, Happy Canyon and Estacado) have Nd concentrations higher than expected for their class of chondrite (Fig. 2a), similar to light-REE mobility observed by Crozaz et al. (2003) in chondrites in desert environments, and the perturbed radiogenic 143 Nd/ 144 Nd ratios of these samples (Fig. 2b). Only in one case (i.e. Maralinga δ 146 Nd = 0.055h) could this be suggested to have driven the stable Nd isotopic composition away from the chondritic average (Fig. 5b) and this may well be a function of the

Nucleosynthetic anomalies
Small variations in the distribution of p-and s-process radionuclides in the inner Solar System has produced measurable differences in the Nd isotopic compositions of the different classes of chondritic meteorites and the Earth (e.g. Andreasen and Sharma, 2006;Burkhardt et al., 2016;Gannoun et al., 2011). Therefore, it is important to assess any possible nucleosynthetic effects on the δ 146 Nd ratios reported here. Both of the isotopes used in the double spike ( 145 Nd and 150 Nd) have observable nucleosynthetic anomalies. The high-precision dataset of Burkhardt et al. (2016) has shown that Nd nucleosynthetic variations are correlated (μ 150 Nd = 2.74 × μ 145 Nd) resulting from the heterogeneous distribution of s-process radionuclides in the solar nebula. We have developed a simple model where a synthetic mixture of 60% JNdi-1, has its isotope composition modified with progressively larger anomalies on 145 Nd and 150 Nd, this was then mixed with 40% double spike and then deconvolved using the geometric iterative resolution method (Siebert et al., 2001) to assess the effects on δ 146 Nd (Fig. 7). In natural samples (bulk meteorites and calciumaluminium rich inclusions) measured μ 150 Nd anomalies range from −91 to 120 (Bouvier and Boyet, 2016;Brennecka et al., 2013;Burkhardt et al., 2016;Gannoun et al., 2011), this level of anomaly results in variations in δ 146 Nd of ≤ ±1.5 ppm, which is less than the smallest internal error on individual measurements (±3 ppm) and significantly less than the reproducibility of four replicates of the Allende meteorite (±14 ppm 95% s.e.; Fig. 7). Due to the con-  (Bouvier and Boyet, 2016;Brennecka et al., 2013;Burkhardt et al., 2016;Gannoun et al., 2011). Also shown (red line) is the effect of a −7 ppm anomaly on 146 Nd as estimated for the Allende meteorite by Saji et al. (2016). (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.) vention of normalising to 146 Nd/ 144 Nd to correct for instrumental mass bias during mass spectrometry, existing data for nucleosynthetic anomalies on 146 Nd and 144 Nd is sparse. By instead normalising to the 148 Nd/ 144 Nd ratio, Saji et al. (2016) suggested that a μ 146 Nd anomaly of −6.3 ppm occurs in the Allende meteorite. The effect of a −7 ppm anomaly on μ 146 Nd results in the shift of δ 146 Nd to lighter values by the same magnitude (Fig. 7), although remaining well within the reproducibility of the stable Nd isotope data. In summary, nucleosynthetic anomalies may be responsible for minor fluctuations in δ 146 Nd values, but they cannot account for the range seen in chondrites.

Aqueous alteration and thermal metamorphism
Variations in the degree of aqueous alteration and/or thermal metamorphism that meteorites experience on their parent bodies may have perturbed their stable Nd isotopic compositions. For whole rock samples such a process requires open system behaviour, however, because of the small volumes of meteorite material used for analysis this may have occurred on a relatively small scale. Chondrites are classified according to their petrologic type and degree of alteration (Van Schmus and Wood, 1967), in this classification type 3.0 represents the most pristine material observed, with type 2 and 1 signifying more intense hydrous alteration. All of the seven recognised CI chondrites are of petrologic type 1 (Weisberg et al., 2006), indicating that they are heavily hydrated meteorites dominated by a fine-grained matrix with few primary chondrules or calcium aluminium-rich inclusions (CAIs). However, this level of intense aqueous alteration appears to have had little discernible effect on the isotopic composition of, for example, the meteorite Orgueil which possesses an δ 146 Nd value of −0.029h (Fig. 3), which is identical to the chondritic average. The majority of CM chondrites are of petrologic type 2 (Weisberg et al., 2006), although the degree of aqueous alteration experienced by different CM chondrites varies widely (Rubin et al., 2007). The meteorites analysed here include those considered both the least (Murchison) and most (Cold Bokkeveld) altered examples of CM2 meteorites (Browning et al., 1996). At  (Hezel et al., 2008), with the 2σ errors on the CAI variability shown at the base of the graph. The grey bars represent the average chondritic composition from Fig. 3, plotted at <0.2% CAI content as this is maximum abundance observed in enstatite or ordinary chondrites (Hezel et al., 2008). The group CV3 carbonaceous chondrites and the anomalous find sample Maralinga are excluded from the chondritic average (hollow symbols). Calculated 2 s.e. errors on the CM-and CV-groups are ±0.037h and ±0.070h, respectively, consistent with higher CAI contents resulting in increased variability in δ 146 Nd. this stage it is not possible to definitively relate the difference in δ 146 Nd between these samples with the intensity of aqueous alteration. Petrologic type numbers rising from 3.1 to 6 designate increasing degrees of thermal metamorphism. Temperatures up to 900 • C are thought to be recorded in type 6 ordinary chondrites (Slater-Reynolds and McSween, 2005), during this heating the recrystallization of minerals combined with open system behaviour could, in principle, perturb their Nd stable isotopic compositions. Mineralogical evidence shows that 30-50% of the Nd in chondrites may be stored in phosphates, with the remainder distributed amongst the silicate phases, thus Nd is not easily mobilized during the early stages of metamorphism (Martin et al., 2013). Seven different group-H ordinary chondrites varying from petrologic type H3/4 to H6 have been analysed here with δ 146 Nd varying from −0.051h to −0.025h (Fig. 3). All H-group ordinary chondrites are indistinguishable from each other within the analytical uncertainties indicating there is no measurable variation in δ 146 Nd with increasing thermal metamorphism.

Carbonaceous chondrites and the CAI effect
Carbonaceous chondrites span the entire range of δ 146 Nd values observed in chondritic meteorites ( 146 Nd = 143 ppm) and possess twice the variability of the other classes of chondritic meteorites. This additional variability may simply be the result of their unique constituents with carbonaceous chondrites containing variable proportions of chondrules, matrix and up to 6 wt% CAIs. Significant stable isotope whole rock variability in carbonaceous chondrites has also been observed for a range of other refractory lithophile elements (e.g. Eu: Moynier et al., 2006or Sr: Charlier et al., 2012Moynier et al., 2010). In carbonaceous meteorites CAIs host a significant proportion of the total Nd, a simple mass balance calculation using a range in CAI Nd concentration of 14-25 ppm (Burkhardt et al., 2016) and assuming 1 ppm Nd in the whole rock (see Table 1) and a modal CAI proportion of 3% (Hezel et al., 2008) shows that 42-75% of the Nd budget is controlled by CAIs. Therefore, the presence of significant Nd stable isotope anomalies in fine or coarse grained CAIs and the possible presence of fractionation and unknown nuclear (FUN) effect CAIs could cause the wider range in δ 146 Nd seen in carbonaceous chondrites. This preposition is supported by the progressively larger variability of δ 146 Nd seen in the different groups of carbonaceous chondrites as their modal CAI proportions increase (Fig. 8). Carbonaceous chondrites from the CI-, CK-, and CO-groups with <1% CAIs have relatively uniform δ 146 Nd, but as the proportion of CAIs continues to increases, 1.21% in the CM-group and 2.98% in the CV-group, the dispersion in δ 146 Nd also grows to ±0.037h and ±0.070h, respectively. The exception to this observation is the anomalous CK4 Maralinga, a desert find that contains an extreme >20-fold enrichment of Nd (see Fig. A4) and has also been shown to have an abnormal Zn isotope composition (Pringle et al., 2017).
The CV (Vigarano-like) chondrites, possess a high proportion of matrix relative to chondrules and abundant large CAIs (weighted average 2.98%, but proportions up to 6% are observed; Hezel et al., 2008), isotopically they are the most variable group of carbonaceous chondrites ( 146 Nd = 119 ppm; Fig. 8). Two wellcharacterised meteorites have distinctly different δ 146 Nd values (Vigarano = 0.029h; Allende = −0.057h; Table 1) that diverge in opposite directions from the chondritic mean. For several other lithophile elements including, Ca (Niederer and Papanastassiou, 1984;Simon and DePaolo, 2010), Cd (Wombacher et al., 2008), Eu (Moynier et al., 2006), Sr (Charlier et al., 2012;Moynier et al., 2010;Patchett, 1980) and Zn (Luck et al., 2005), significant enrichments in the light isotopes have been measured in refractory CAIs relative to the meteorite matrix. These data are difficult to reconcile with condensation from a light isotope depleted gas (Richter, 2004), and rather have been attributed to kinetic effects (Richter, 2004) or electromagnetic separation (Moynier et al., 2006). Significantly, Charlier et al. (2012) showed for Sr that CAIs in Allende are isotopically light (δ 88 Sr = −0.07h to −0.35h) and matrix material is heavy (δ 88 Sr = 0.65h) with whole rock compositions simply reflecting a mixture of these components. By comparison, in Vigarano CAIs are significantly heavier (δ 88 Sr = 0.05h) but so is the whole rock (δ 88 Sr = 0.35h). If such an offset between CAIs, matrix and whole-rock also exists for stable Nd isotopes then this may go some way to explaining the heavier whole-rock value for Vigarano relative to Allende observed here.
The CV group has been further subdivided on the basis of petrologic characteristics into one reduced and two oxidised subgroups (Weisberg et al., 1997). Matrix/chondrule ratios increase in the order CV red (Vigarano-like = 0.5-0.6), CV oxA (Allende-like = 0.6-0.7)-CV oxB (Bali-like = 0.7-1.2), whereas metal/magnetite ratios decrease in the same order. A systematic decrease in δ 146 Nd values is observed as the meteorites become progressively more oxidised (e.g. Vigarano +0.029h, Allende = −0.057h, NWA 2086 = −0.088h; this sample is a Bali-like CV on the basis of the chondrule proportion reported in Kereszturi et al., 2014). Mineralogical studies suggest that the differences between the sub-groups are a consequence of varying degrees of late-stage alteration (Krot et al., 1995). Stable Nd isotopes may therefore provide a means to trace aqueous alteration and oxidation in CV group meteorites although the mechanism by which this would occur remains poorly constrained.

Inherited compositional differences between parent bodies
Each chondrite group is considered to have sampled a separate parent body (Weisberg et al., 2006). In this case it is possible that variable δ 146 Nd isotope compositions could be inherited from differences in the distribution of Nd in the protoplanetary disk or through the different mechanisms of formation of each these bodies. Ordinary chondrites, are thought to be derived from at least three separate parent bodies (H, L, and LL) on the basis of their major element chemistry, oxygen isotopic compositions and degree of iron oxidation (Yomogida and Matsui, 1984). Significant variability is seen within the ordinary chondrites analysed here, with H-group samples (δ 146 Nd = −0.051h to −0.025h) appearing to be systematically lighter than the L-group chondrites (δ 146 Nd = −0.021h to 0.006h; Table 1). Although the average δ 146 Nd values of these two groups are indistinguishable at the ±2σ level, with H-group ordinaries having δ 146 Nd = −0.036 ± 0.017h (n = 10; 95% s.e = ±0.006h) and L-group ordinaries δ 146 Nd = −0.008 ± 0.019h (n = 9; 95% s.e. = ±0.009h), their means are clearly resolvable at the 95% confidence level (Fig. 3). Systematic variations in the Ni, Zn and Cu stable isotope compositions of the different ordinary chondrites groups (from H to L and LL) have been documented and linked to evaporative processes or differences in parent body formation (Moynier et al., 2007). These systematic differences are not observed in δ 146 Nd. Experimental data indicates that Nd is not strongly partitioned into metal (Bouhifd et al., 2015), and is highly refractory (Lodders, 2003), and hence Nd isotopes are unlikely to be affected by metallic segregation or evaporation processes (i.e. volatile loss). Moynier et al. (2007) attribute variations in Cu isotopes amongst ordinary chondrites to the loss of sulfide, either during evaporation or removal to a metallic core. If light Nd is indeed preferentially partitioned into sulfide, then such a process could explain these variations, however, this is difficult to reconcile with the stable Nd isotope data given that it would fail to explain the lighter values in H-group chondrites than the chondritic mean (Fig. 3) and the fact that different groups of ordinary chondrite possess indistinguishable S contents (Cripe and Moore, 1975;Dreibus et al., 1995). In summary, the small difference in δ 146 Nd between H and L ordinary chondrites observed here are difficult to reconcile with planetary differentiation processes, instead it could simply be a heterogeneity that survives from the earliest stages of nebular condensation.

Modification of stable Nd isotopes during magmatic processes
Magmatic rocks can experience a number of processes including partial melting, fractional crystallisation, mixing and crustal assimilation each of which may have modified their Nd stable isotopic compositions.

Magmatic differentiation
The rock standards analysed here are sourced from widely dispersed localities and a range of intrusive and extrusive rock types including basalt, andesite, dolerite, diorite, granite, rhyolite and tonalite (Table 2; Fig. A5). Samples with basaltic to dacitic major element compositions show minimal stable Nd isotope variability ( 146 Nd = 35 ppm), with an average composition of δ 146 Nd = −0.030 ± 0.014h, and no systematic changes with magmatic differentiation (SiO 2 = 38-69 wt%; Fig. 5a). In contrast, the two most felsic rock standards analysed possess progressively heavier δ 146 Nd as SiO 2 increases above 70 wt% (RGM-1: SiO 2 = 73 wt%; δ 146 Nd = −0.008h and JG-2: SiO 2 = 76 wt%; δ 146 Nd = 0.013h). One cause of this enrichment in heavy Nd in the rhyolitic samples could be the crystallisation of accessory mineral phases. Sphene, allanite and apatite can all show 100-1000 fold enrichments of light REE relative to whole-rock (e.g. Sawka, 1988) and such partitioning might induce an isotopic fractionation due to a contrast in bonding coordination with the melt. This data suggests that in highly evolved rocks, magmatic processes may modify their stable Nd isotopic composition although the exact cause remains unknown.

Partial melting
Neodymium behaves as an incompatible element during mantle melting. Therefore, mass balance dictates that, if any stable isotope fractionation were to occur during this process, it should be recorded in the residue, whereas the original Nd stable isotopic composition of the mantle source will be transferred to the product of melting. Simple peritectic melt modelling (McCoy-West et al., 2015) demonstrates that with just 10% melting >80% of the original Nd budget, in both the spinel and garnet facies mantle (92% and 84%, respectively; Fig. A6), would be in the melt fraction. The broadly similar Nd isotope compositions of basaltic samples measured here suggests that the mantle preserves a homogeneous δ 146 Nd composition. In addition, that 6 out of the 8 mantle samples analysed here yield values that are within error of the basaltic samples indicates that little fractionation occurs during partial melting (Fig. 4). This is further highlighted by comparing MORB glasses and their complementary abyssal peridotites from the Garrett Fracture Zone. The MORB glasses have δ 146 Nd values ranging from −0.030h to −0.007h and an average composition of −0.019h (n = 4), the abyssal peridotites (mantle residues) have slightly more variable δ 146 Nd with values ranging from −0.040h to 0.008h and an average of −0.014h. For the samples measured here, at least, there is no resolvable effect on δ 146 Nd that can be attributed to partial melting.

Removal of light Nd to the core during planetary formation?
Recent experimental work suggests that the outer core may contain a significant amount of sulfide added during the final stages of accretion (Wade et al., 2012), and that at reduced conditions analogous to core formation sulfide can incorporate a substantial quantities of refractory lithophile and heat-producing elements (e.g. Nd and U; Wohlers and Wood, 2015). Therefore a substantial amount of Nd may have been separated into a sulfide or a sulfur-rich metal phase during the final stages of planetary differentiation, possibly providing a mechanism to resolve the 142 Nd mismatch between chondritic meteorites and the Earth. Neodymium stable isotopes have the potential to provide a tracer of sulfide segregation, because there is a significant contrast in bonding environment between sulfide and silicate, where heavy isotopes should be preferentially incorporated into the high forceconstant bonds involving REE 3+ ions in silicate minerals (i.e. mantle). At first sight, however, based on the stable Nd data presented here it seems unlikely that a significant amount Nd could have been sequestered into the core or else REE incorporation into sulfide in the core was not accompanied by stable Nd isotope fractionation. The average 146 Nd/ 144 Nd isotope composition of chondrites (δ 146 Nd = −0.025 ± 0.004h) is indistinguishable from that of the bulk silicate Earth (δ 146 Nd = −0.022 ± 0.006h) at the 95% confidence level (Fig. 4). This result is consistent with recent high precision Nd and Sm isotopic data that suggest, rather, that a heterogeneous distribution of s-process radionuclides in the Solar Nebula is responsible for the 142 Nd discrepancy between chondrites and the Earth (Bouvier and Boyet, 2016;Burkhardt et al., 2016). Although a small fraction (∼5 ppm) of the original 142 Nd anomaly between the Earth and enstatite chondrites remains unaccounted for Burkhardt et al. (2016).
The mantle samples analysed here display significantly more variability in δ 146 Nd than the other terrestrial reservoirs measured ( 146 Nd = 75 ppm; Table 2), and on average (δ 146 Nd = −0.008h) are slightly heavier than the chondritic meteorites. An analysis of variance test shows that the mantle samples possess a resolvably different population to both the chondritic meteorites and the terrestrial rock standards at the 99% significance level (p-value <0.001), whereas the rock standards and chondrites are statistically the same. This small difference (17 ppm) between mantle samples and chondrites is consistent with two possible interpretations: 1) the peridotites are indeed heavier and Nd could have been fractionated during the sequestration of light Nd into the core; or 2) the crustal magmatic rocks are more representative of the bulk Earth, due to the incompatible nature of Nd mass balance suggests that melts will dominate the Nd budget after c. 5% melting (Fig. A6). A number of processes may be responsible for varying the δ 146 Nd of peridotites, these include partial melting, melt metasomatism, mineralogical variations, and serpentinisation or alteration during sample emplacement. Two ostensibly identical pristine spinel lherzolites from Kilbourne Hole having significantly different δ 146 Nd ( 146 Nd = 46 ppm). Until a more systematic study of mantle rocks is undertaken, to better understand the processes causing variability in δ 146 Nd, it is impossible to say definitively if the δ 146 Nd of the mantle was indeed perturbed due to core formation or some other secondary process.

Conclusions
Here we present the first comprehensive Nd stable isotope data for chondritic meteorites and a range of terrestrial rocks using a double spike TIMS methodology.
The major classes of chondritic meteorites, carbonaceous, enstatite and ordinary chondrites all have broadly similar isotopic compositions allowing the calculation of an overall chondritic mean of δ 146 Nd = −0.025 ± 0.025h. Enstatite chondrites yield the most uniform stable isotope composition ( 146 Nd = 26 ppm), with considerably more variability observed within ordinary ( 146 Nd = 72 ppm) and carbonaceous meteorites ( 146 Nd = 143 ppm). The effects of terrestrial weathering, nucleosynthetic anomalies and parent body metamorphism produce no resolvable variations on the δ 146 Nd of chondrites. A resolvable offset ( 146 Nd = 28 ppm) between H and L group ordinary chondrites is probably the result of an inherited heterogeneity from the earliest stages of condensation preserved in their different parent bodies. The larger variations observed in carbonaceous chondrites, specifically the CM-and CV-groups correspond to varying modal proportions of calcium-aluminium rich inclusions.
The terrestrial samples analysed here include rocks ranging from basaltic to rhyolitic in composition, MORB glasses and mantle restites. All of these terrestrial reservoirs possess a broadly similar Nd isotope compositions allowing the calculation of a mean for the bulk silicate Earth composition of δ 146 Nd = −0.022 ± 0.034h. Magmatic differentiation appears to have minimal effect on the δ 146 Nd of basaltic to dactitic magmas, with resolvably heavier δ 146 Nd values only observed in highly evolved magmas with >70 wt% SiO 2 .
The average δ 146 Nd of chondrites and the silicate Earth are indistinguishable at the 95% confidence level. Mantle samples possess highly variable stable Nd isotope compositions ( 146 Nd = 75 ppm) with an average δ 146 Nd of −0.008h, if these heavier values represent the true composition of the mantle then it is not possible to completely rule out some contribution from core formation in the offset between the mantle and chondrites. Although, due its highly incompatible behaviour Nd mass balance suggest that melting products strongly dominant the Nd budget. Thus on balance, these results are consistent with the 142 Nd mismatch between the Earth and chondrites being best explained by a higher proportion of s-process Nd in the Earth, rather than partitioning into sulfide or S-rich metal in the core.