- © 2011 Mineralogical Society of America
We measured hydrogen concentrations in 12 olivines using secondary ion mass spectrometry (SIMS and NanoSIMS), cross-calibrated against Fourier transform infrared (FTIR) spectroscopy and nuclear reaction analysis (NRA). Five of these samples are routinely used for calibration in other laboratories. We assess the suitability of these olivines as standards based on over 300 SIMS analyses, comprising 22 separate calibrations. Seven olivines with 0–125 ppm H2O give highly reproducible results; in contrast to previous studies, the data are fit to well-constrained calibration lines with high correlation coefficients (r2 = 0.98–1). However, four kimberlitic megacrysts with 140–245 ppm H2O sometimes yield 16O1H/30Si ratios that have low internal precision and can vary by up to a factor of two even in sequential analyses. A possible cause of this behavior is the presence of sub-microscopic inclusions of hydrous minerals, such as serpentine. In most cases, however, we link the anomalous results to the presence of sub-micrometer to micrometer-scale pores (as small as 100 nm), which we imaged using SEM and NanoSIMS. These pores are interpreted to be fluid inclusions containing liquid H2O, other volatiles (including fluorine), and/or hydrous phase precipitates. Ionization of the contents of the pores contributes variably to the measured 16O1H, resulting in analyses with erratic depth profiles and corresponding high uncertainties (up to 16%, 2σmean). After filtering of these analyses using a simple criterion based on the error predicted by Poisson counting statistics, all the data fit well together. Our results imply that the Bell et al. (2003) calibration can be applied accurately to all olivines with IR bands from ~3400–3700 cm−1, without the need for band-specific IR absorption coefficients.
The accurate analysis of trace concentrations of hydrogen in nominally anhydrous minerals (NAMs) and glasses is a long-standing problem (Aines and Rossman 1984; Rossman 1990, 2006; Ihinger et al. 1994; Hauri et al. 2002), with wide-ranging applications to geology and planetary science (e.g., Bell and Rossman 1992; Smyth and Jacobsen 2006; Saal et al. 2008; Hirschmann et al. 2009). Much of the work on NAMs has concentrated on olivine, the most abundant mineral in the Earth’s upper mantle. Although experimental studies have shown that olivine can incorporate large amounts of hydrous components—up to nearly 1 wt% (as H2O) at pressures of ~13 GPa (Mosenfelder et al. 2006a; Smyth et al. 2006)—typical concentrations in natural crystals are orders of magnitude lower, ranging from 0–20 ppm H2O (by weight, as used throughout this paper) for samples derived from the upper-most mantle (plagioclase- or spinel-lherzolite facies) to 50–400 ppm H2O for xenocrysts from diatremes or kimberlites (garnet-lherzolite facies) (Miller et al. 1987; Bell and Rossman 1992; Bell et al. 2004; Matsyuk and Langer 2004; Koch-Müller et al. 2006; Mosenfelder et al. 2006b).
Fourier transform infrared (FTIR) spectroscopy is the most commonly applied tool for measuring such low concentrations. An increasingly popular and powerful technique is secondary ion mass spectrometry (SIMS), which has benefited from recent improvements in instrumentation, analytical protocols, and sample preparation techniques (Kurosawa et al. 1997; Hauri et al. 2002; Koga et al. 2003; Aubaud et al. 2007). FTIR and SIMS are complementary methods, each with its own advantages and disadvantages. FTIR provides constraints on mechanisms of H incorporation. In glasses, speciation into H2O and OH molecular groups can be readily and quantitatively determined. In minerals, it is possible—though with less certainty than in glasses—to assign different O-H absorption bands to intrinsic defects (point defects, substitutions) or extrinsic defects (inclusions or extended defects comprised of hydrous phases, fluid inclusions). Under ideal conditions, with thick samples (on the order of 1 cm) in transmission mode, the detection limit for IR is very low, less than 1 ppm H2O (Fig. 1). SIMS, on the other hand, offers enhanced spatial resolution (particularly with NanoSIMS), with the ability to examine much smaller, near-surface volumes of material, and the convenience and power of simultaneous analysis of other elements. SIMS can also be used for samples that are too fragile and/or small in grain size to analyze practically using FTIR, which requires double-polished sections to acquire high-quality data.
Recently published SIMS and NanoSIMS calibrations for hydrogen in olivine (Tenner et al. 2009; Kovács et al. 2010; O’Leary et al. 2010) suffer from scatter in excess of analytical precision (Fig. 2), to a greater degree than for other minerals and/or glasses calibrated in the same studies. We assessed the causes of this scatter by conducting extensive SIMS and NanoSIMS measurements on 12 olivine standards, five of which are routinely used for calibration in two other ion microprobe laboratories (Carnegie Institute of Washington and Arizona State University). Data are reported from two sessions each on the Cameca 7f-GEO and Cameca NanoSIMS 50L ion microprobes, both located at the Center for Microanalysis at Caltech. We examined both primary standards, for which H concentrations were determined using nuclear reaction analysis (NRA) (Bell et al. 2003), and “secondary standards”, for which H concentrations were determined using polarized FTIR spectroscopy, referenced to the calibration developed by NRA. Repeated calibrations were conducted in the course of bracketing measurements of hydrogen concentration profiles, at levels of ~10–100 ppm H2O, in experimentally annealed and natural olivines (Le Voyer et al. 2010; Mosenfelder et al. 2010). Critical requirements for such measurements are the establishment of low blanks and attainment of high precision, which prompted us to search for new, suitable secondary standards with H2O concentrations in the 0–50 ppm range.
Based on over 300 individual SIMS and NanoSIMS measurements comprising 22 separate calibrations (i.e., discrete blocks of sequential analyses on standards in between blocks of analyses on unknowns), we assess reproducibility of the standards, explore causes of standard heterogeneity, recommend protocols for ion microprobe analysis, and address the possible influence of spectral variability on IR absorption coefficients (Koch-Müller et al. 2006; Kovács et al. 2010). We also present new data on the correlation between fluorine and hydrogen contents in mantle olivines (Hervig and Bell 2005).
In accordance with previous work (Hauri et al. 2002; Koga et al. 2003; Aubaud et al. 2007), we used epoxy-free preparation methods to reduce H backgrounds for SIMS. Samples were cut with a 50 μm diameter wire saw and attached for polishing to glass plates using ethyl-2-cyanoacrylate (“Super Glue”) and poly(methyl methacrylate) (“Orthodontic Resin” from Denstply International Inc.). These organic compounds are soluble in acetone and toluene, respectively. Sample sections were doubly polished with alumina papers, down to 0.3 μm grit size. After removal from the glass plates and analysis by FTIR (see below), the samples were cleaned in an ultrasonic bath using a sequence modified after Aubaud et al. (2007): three 20 min baths in toluene, three 20 min baths in acetone, and three 20 min baths in isopropyl alcohol. The crystals were then baked overnight at 115 °C in a vacuum oven prior to mounting and pressing (using a pellet press) in indium, which had been melted and pressed into aluminum disks. After all samples were mounted in indium, the mounts were cleaned using three 20 min baths in acetone and three 20 min baths in isopropyl alcohol. In agreement with Tenner et al. (2009), we found the final polishing step adopted by Aubaud et al. (2007) to be unnecessary. After mild baking overnight (50 °C in a vacuum oven), the mounts were sputter coated with 30–60 nm of gold and inserted into the airlock of either the 7f-GEO or NanoSIMS 50L.
Polarized IR spectra were obtained for all samples using Nicolet 60SX or Nicolet magna IR 860 FTIR spectrometers at Caltech. Spectra taken on oriented crystals prior to cutting slices for SIMS have been published previously for most standards (Miller et al. 1987; Bell et al. 2003, 2004; Mosenfelder et al. 2006b); these data were acquired in the main compartment of the spectrometer, using a LiIO3 Glan-Foucault prism polarizer with the E-vector of the radiation aligned parallel to the three crystallographic axes. Spectra were also taken on the thin slices that were cut off the ends of large crystals and actually used for SIMS, using an IR microscope (details in Mosenfelder et al. 2006a), to confirm lack of zoning in hydrogen contents. Crystal orientations were established either by X-ray diffraction (Miller et al. 1987) or optical techniques (Bell et al. 2003, 2004; Mosenfelder et al. 2006b) and are estimated to be accurate to better than 5°.
where Abstot is the total integrated absorbance per centimeter, calculated from polarized spectra in the three principal optic directions α, β, and γ (formalized by Eq. 1 in Bell et al. 2003), using integration limits as discussed in that study. For the samples from Bell et al. (2004), only two out of three directions could be measured and the calibration coefficient was adjusted accordingly (from 0.188 to 0.233 ± 0.015, assuming 6% uncertainty). Final uncertainties in concentrations were assessed in the following manner. For the three primary standards measured by Bell et al. (2003), we used the assigned uncertainties derived from the NRA measurements. Otherwise, we propagated the uncertainty of the calibration coefficient with an estimate of uncertainty of the absorbance measurements. The uncertainty in absorbance is dominated by the uncertainty in baseline correction, as discussed extensively by Bell et al. (1995, 2003, 2004); note that none of the samples used in this study are measurably zoned in hydrogen content. We used a value of ±10%, except for the three ROM samples (±15%) and GRR997 (±20%). The higher uncertainty for the ROM samples (±15%) is discussed by Bell et al. (2004) and is primarily derived from the uncertainty in broad-band (molecular water) contribution to the absorbance and the use of only two out of three polarizations for each sample. For GRR997, the relatively high signal:noise ratio at very low absorbance and steep curvature of the baseline (Fig. 1) led us to assign a higher uncertainty.
SIMS: Cameca 7f-GEO
We followed established procedures (Hauri et al. 2002; Koga et al. 2003; Aubaud et al. 2007) for reducing H (and thus 16O1H) backgrounds by baking the instrument for 24 h prior to each session and keeping samples in the sample exchange airlock for as long as possible prior to analysis (24–72 h). Attainment of ultra-high vacuum (2–4 × 10−10 torr in the sample chamber) was also aided by use of a liquid N2 cold trap. The 16O1H background was assessed periodically by analyzing GRR1017, DSCOL, and/or GRR997 (all standards with less than 1 ppm H2O; Table 1). During the first session, the 16O1H/30Si ratio of GRR997 on two different mounts showed little drift, with an average value of 0.0019 ± 0.0001 (n = 14) (Fig. 3). The background was high at the start of session 2 as a result of needing to open the sample chamber to fix a part prior to pumping down and baking of the instrument, but eventually reached lower levels than in session 1 due to the long duration of the session (12 days total under ultra-high vacuum conditions, without introduction of highly degassing samples such as epoxy mounts).
For all analyses, we used a Cs+ primary ion source that delivered a current of 2–8 nA on the sample surface with an impact energy of 20 keV to sputter the samples and produce negative ions. Charge compensation was provided by a normal-incidence electron flood gun. The beam was rastered to produce craters ~25 × 50 μm in dimension and a 100 μm field aperture was used to collect ions from the central 8 μm of the crater. For each analysis in the first session (calibrations 1 and 2, Table 2), following ~4 min of pre-sputtering (including sputtering during automated beam alignment; see below), we measured 30 cycles through the mass sequence 12C, 16O1H, 19F, 30Si, 32S, 35Cl, and 27Al16O, with counting times of 5, 2, 1, 1, 3, 5, and 1 s, respectively. A mass-resolving power (M/ΔM) of 5000 was used to separate the 16O1H peak from 17O. Ions were detected using an electron multiplier, with counts corrected for detector background and counting system deadtime. Uncertainties in counts and ratios are cited as two times the standard error (2σmean) from the 30 counting cycles.
Our analysis protocol for the second session (calibrations 3–19) was essentially identical to the first except that we increased the pre-sputtering time to 5 min and dropped 27Al16O from the mass counting cycle. Furthermore, for the final set of measurements (calibration 19) we added 18O to the counting cycle and deleted 19F, 32S, and 35Cl.
Primary and secondary ion beam alignment is critical for maximizing analytical reproducibility, particularly for the low 16O1H/30Si ratios studied here. Our protocol for beam alignment differs slightly from previous studies because the 7f-GEO is equipped with software and hardware (deflectors and motors) to control aperture and slit movements, allowing for automated tuning. Therefore, after initial, manual beam alignment on a standard in the middle of the mount (GRR997), we generally used the automated routine, which aligns the secondary beam with respect to the primary beam by scanning the deflectors and motors to center the field aperture, contrast aperture, entrance, and energy slits, and perform mass calibration centering on 30Si or 18O. Where necessary (e.g., farther from the center of the mount), the secondary beam was realigned semi-manually by scanning parameters outside the set range of the automated routine. Analyses with low Si counts (<10 000 cps for our instrumental parameters) were rejected on the assumption that the automated beam alignment routine did not perform well.
Alignment of the electron flood gun is also important for maintaining reproducibility for low-mass elements (Hauri et al. 2002). The electron beam is deflected by changes in the stray magnetic field of the spectrometer magnet, as its B-field switches during mass cycling. The corresponding effect on charge compensation can significantly affect count rates. This problem is partially mitigated by shielding of the magnet in our instrument with a mu-metal plate. Moreover, measurement of 16O1H rather than 1H minimizes this problem substantially (e.g., Tenner et al. 2009).
Surface contamination was assessed by examining the 30-cycle sequence of 12C/30Si and 16O1H/30Si ratios for each analysis. These sequences reflect progressive sputtering into the sample and are thus hereafter referred to as “depth profiles”, while noting that the depth was not calibrated and the depth resolution is poor compared to traditional SIMS depth profiling. Some analyses show a monotonic decrease in both ratios, leading to a plateau. In these cases we arbitrarily set a cut-off value of 0.001 or 0.002 for the 12C/30Si ratio, based on visual estimation. Cycles with ratios above these values were then eliminated, and the ratios recalculated. In rare cases, where an entire profile exhibited ratios above our cut-off value, the whole analysis was rejected. Both raw and re-processed analyses are given in the supplementary material1. Depth profiles were also used to assess heterogeneity within standards not related to organic contamination, and we discuss this phenomenon throughout the rest of the paper.
SIMS: NanoSIMS 50L
As with the 7f-GEO, the NanoSIMS 50L was baked for 24 h prior to the analytical sessions. Vacuum in the sample chamber ranged from 1 to 2 × 10−10 torr. The NanoSIMS 50L is usually used with a “low current” setting (a few pA delivered at the surface of the sample; e.g., Dekas et al. 2009), allowing study of sample areas that are smaller than 1 × 1 μm. However, the current can be maximized using the primary lenses (L0 and L1) to measure very low H contents such as in this study. This “high current” setting (Saal et al. 2008) corresponds to a reading of ~25 nA on the Faraday cup situated on the outside of the primary column. Consequently, the crater size is larger, ranging from 10 to 20 μm in diameter. This compromise between spatial resolution and detection limit proved sufficient for studying low H-content concentration profiles in natural samples (Le Voyer et al. 2010).
For the analyses reported here, a Cs+ primary ion beam of 0.6–1.0 nA was delivered on the sample surface, with charge compensation provided by an electron flood gun. We used a 5 μm diameter beam rastered over a 5 × 5 μm area, divided into 128 × 128 pixels with 200 μs dwell time per pixel, and an electronic gating on the central 3 × 3 μm. Electronic gating provides a function analogous to using the field aperture on the 7f-GEO, mitigating effects of contamination from the crater edge. After 200–327 s of pre-sputtering, the secondary ion beam was automatically centered (using both SIBC and the E0S lens), as well as the position of the 30Si mass peak. Then we collected signals on six separate electron multiplier detectors in multi-collection mode, with a counting time of 1 s for each mass. Various masses were collected for different analyses; here we report only data for 12C, 16O1H, 19F, 27Al, 30Si, and/or 35Cl. The counting cycle was repeated 80–200 times for each analysis. The smallest exit slit (no. 3) was set on the 16O1H detector to separate 16O1H from 17O (mass resolution power of 13 000–14 000 using Cameca NanoSIMS units; ~6000 using the IUPAC definition).
NanoSIMS ion images of sample KLV23 were acquired using a “low current” setting with only 17 pA delivered to the sample surface, to resolve small features (<1 μm). We used a 10 × 10 μm raster size and 1 h of counting time.
Scanning electron microscope (SEM) and electron microprobe analysis
SEM analyses were performed on a LEO1550VP field emission gun instrument (FE-SEM). For secondary electron imaging we operated at 10 keV using either a lateral detector (“SE2”) or a detector within the beam-focusing lens (“InLens”). The InLens detector facilitates high-resolution imaging of topographic features. Backscattered electron (BSE) imaging and semi-quantitative chemical analysis (using an Oxford INCA energy-dispersive X-ray spectrometer) were performed at 20 keV.
Sub-micrometer scale features were examined on separate sample splits from the plates used for SIMS and FTIR. Fresh surfaces were prepared by cutting and cleaning small chips from the crystals, and then lightly tapping to initiate fracture, typically on the (010) cleavage plane. The fractured surfaces were then coated with 10 nm of carbon. We also examined the craters in the sample mounts corresponding to SIMS analyses.
Electron microprobe analyses of most of the samples have been presented elsewhere (Bell et al. 2003; Mosenfelder et al. 2006b). Other samples (GRR997, GRR1017, and DSCOL) were analyzed with a JEOL JXA-8200 using similar standards and operating conditions as described in these papers. Analyses are compiled and provided in the supplementary material1.
Table 1 gives FTIR absorbances for three different polarizations in each sample in the O-H vibrational region (between 3000–4000 cm−1). In the supplementary material, we provide complete data (count rates, ratios, and uncertainties for all reported masses) for our two SIMS sessions (168 analyses total) and two NanoSIMS sessions (143 analyses total). Fits to these data, which comprise 22 separate calibrations (19 on the 7f-GEO, three on the NanoSIMS), are given in Table 2 and discussed below.
FTIR spectra are shown in Figures 1, 4, 5, and 6. Table 1 lists the major O-H absorbance bands for each sample and classifies the bands according to the scheme used by Mosenfelder et al. (2006b), modified after Bai and Kohlstedt (1993). This scheme is illustrated in Figure 4 (see also Fig. 1 in Mosenfelder et al. 2006b), where we divide the bands into group I (3450–3700 cm−1) and group II (3000–3450 cm−1) and further designate group Ia, comprised of bands at 3573, 3563, 3541, 3525, and 3485 cm−1. Recent experimental and theoretical work (Lemaire et al. 2004; Berry et al. 2005, 2007; Grant et al. 2007; Walker et al. 2007; Balan et al. 2011) provides evidence linking these bands to H associated with Si vacancies (group I), “titanoclinohumite-like point defects” comprised of a Si vacancy coupled to Ti substitution in an octahedral site (group Ia), and Mg vacancies and/or coupled substitution with trivalent cations, including Fe3+ and Al3+ (group II). Some of these attributions are controversial. For instance, the assignment of group I bands to Si vacancies contradicts X-ray diffraction studies on forsterite with high H contents (Kudoh et al. 2006; Smyth et al. 2006) and the thermodynamic model of Kohlstedt et al. (1996). However, the exact nature of the defects is irrelevant to the conclusions of the present study. The salient point, affecting our conclusions with regard to band-specific absorption coefficients, is that the olivines measured in this study show a wide range of spectral features. Some are dominated by group Ia bands (e.g., GRR1695-2, KLV23, and ROM250-OL2); some exhibit almost exclusively group I bands (GRR1629-2 and GRR1784e); some show a mix of groups I and Ia (e.g., ROM177 and ROM250-OL13); and one olivine (GRR999a) has weak group I and Ia bands but a strong group II band.
Several bands at higher wavenumbers were not previously classified specifically as group I bands, as they were not reproduced in the experimental studies upon which the classification was originally based. Some of these bands are tentatively attributed to the coupled substitution of F and H (Fig. 5), whereas others are caused by inclusions of hydrous minerals (Figs. 5 and 6). Justification for these attributions is given in the discussion section.
The internal precision of our SIMS analyses varied from 0.5 to 16% (2σmean). For some olivines—most frequently for those with ≥140 ppm H2O—measured 16O1H/30Si ratios and precision varied widely under the same analytical conditions. This behavior is illustrated in Figure 7, which shows depth profiles for two sequential analyses from calibration 19 on the same standard (ROM250-OL13). The first analysis has a highly erratic profile, with the lowest measured ratios in the range of values measured for the second, highly precise analysis. The extreme variations in 16O1H counts are not correlated with 12C, 30Si, or 18O counts, indicating that the fluctuation is caused neither by organic contamination nor any easily imagined experimental artifact such as tuning or charging instability. We use multiple lines of evidence throughout the rest of the paper to show that variations in the low-precision analyses represent ionization of variable volumes of nanometer-scale, heterogeneously distributed hydrous inclusions (solid and/or fluid).
Representative calibrations are shown in Figure 2 and Figures 8 to 10. For some calibrations, we took the typical approach of collecting multiple, sequential analyses on each standard to derive a single calibration line. In other cases, we measured only one or two points per standard, either on the full set of standards on a given mount or a smaller subset of three to five olivines. For the sake of regression, all data were blank-corrected by subtracting the 16O1H/30Si ratio of either GRR997 or GRR1017. We assess the suitability of these olivines as blank standards in the discussion section on detection limits. Our correction differs from that used by Tenner et al. (2009), who subtracted the 16O1H count rate of their blank standards rather than the ratio. Their method does take into account variations in overall ion yield, which can be significant (for instance, Tenner et al.’s data differ by up to a factor of 2.4 to 4 in 30Si counting rate, perhaps from variations in primary ion beam or electron gun tuning).
Three different regression models were employed in this study (Table 2). We first performed unweighted, ordinary least-squares (OLS) regressions through all the data points. The regressions were not forced through the origin, as in other studies (Aubaud et al. 2007; Tenner et al. 2009; Kovács et al. 2010), and we note that the meaning of the r2 statistic is thus different compared to their constrained regressions. The second model was also an OLS regression, but we fit only selected data discriminated by examining σmean for the 16O1H/30Si ratio of each analysis relative to the error (σPoisson) predicted by Poisson counting statistics (Eq. 23 in Fitzsimmons et al. 2000). For this purpose, analyses with σmean/σPoisson > 5 were discarded; this procedure filters out analyses such as the first one (σmean/σPoisson = 9.4) shown in Figure 7. The third model was a York regression (York 1966) on the filtered data set, weighting uncorrelated errors in FTIR/NRA and SIMS data (cf. Koga et al. 2003). In this case, we averaged the analyses for each standard because York regressions run through all the individual data points sometimes give unreasonable results due to overweighting of low-uncertainty standards with slightly different 16O1H/30Si ratios but identical values of H2O from FTIR/NRA (e.g., multiple blank measurements). For errors in 16O1H/30Si ratios we used the standard deviation (2σSD) for multiple measurements; nearly identical results are obtained if we use the pooled error (cumulative 2σmean) because the error in slope is dominated by the errors in FTIR/NRA measurements.
Figure 8a shows the complete, unfiltered data set for calibration 19, the final calibration we performed on the 7f-GEO. The two data points shown as solid square symbols (ROM250-OL13, with 245 ppm H2O) correspond to the analyses in Figure 7. This plot illustrates that whereas most of the samples yield 16O1H/30Si ratios with low uncertainties (within the symbol size of the graph), three samples (KLV23, ROM250-OL2, and ROM250-OL13) show a high degree of spot-to-spot scatter as well as much higher uncertainties for each individual analysis. Moreover, some of these analyses do not agree with each other even within the high internal errors. The goodness of fit for the OLS regression (r2 = 0.948) is substantially improved (to r2 = 0.997) by our filtering procedure (Fig. 8b). Similar improvements were realized (Table 2) for other calibrations (for instance, from r2 = 0.847 to 0.998 for calibration 1). We emphasize that this filtering method relies only on the internal error of each measurement, not on any circular reasoning argument based on goodness of fit to the calibration line. Figure 8b also shows the data for calibration 19 using 18O as the reference element, rather than 30Si. The goodness of fit of both regression lines shows that the effect on electron gun tuning of jumping the magnet B-field within the range of 17 to 30 amu is insignificant.
Figure 2 shows a comparison between a typical filtered calibration from our work (calibration 1) and two previously published calibrations (Tenner et al. 2009; Kovács et al. 2010). The comparison is not straightforward because the analytical conditions vary between studies; for instance, the calibration slope in Tenner et al. (2009) might differ from ours due to variations in ionization efficiency (generally a factor of 5–10 lower counting rates in Tenner et al.’s measurements). Nevertheless, the comparison illustrates the significant difference in precision between the studies. The scatter in Tenner et al.’s data may be partially caused by inaccuracies in H2O concentration estimates for some experimentally annealed samples that are zoned in water content and/or have a broad band in their IR spectra, representing molecular H2O (Aubaud et al. 2007). Furthermore, the three natural standards common to all of the studies (ROM177, ROM250-OL2, and ROM250-OL13) sometimes give anomalous results and highly imprecise analyses (Figs. 7 and 8a); it is not possible to assess this possibility for the data of Tenner et al. (2009) or the previous study on the same standards by Aubaud et al. (2007) because uncertainties on the measured counts were not given. The SIMS and NanoSIMS calibrations presented by Kovács et al. (2010) and O’Leary et al. (2010), respectively, show even more scatter. In these calibrations, replicate analyses on some standards disagree with each other within mutual uncertainties, as seen also for two samples in Figure 8a. The reasons for this behavior are discussed below, where we characterize each standard individually.
Two of our three NanoSIMS calibrations (20 and 21) are compared in Figure 9. Calibration 22 is not shown but has a nearly identical slope to calibration 21 (Table 2). The NanoSIMS calibration slopes are apparently more sensitive to variations in tuning parameters than for the 7f-GEO, where we see no systematic dependence of calibration slope on beam current (between 2–8 nA). The large difference in calibration slopes shown in Figure 9 is probably only partly due to differences in sample current (0.6 vs. 0.8 nA; calibration 22 was at 1 nA); other tuning parameters in the secondary ion beam column are also important (e.g., aperture widths to the mass spectrometer). Nevertheless, for a given set of tuning parameters, we attain high precision for filtered OLS regressions (r2 = 0.98–0.99) comparable to regressions for 7f-GEO data (r2 = 0.99–1).
An example of the application of York regression is shown in Figure 10, again plotting the filtered data set for calibration 19. Whereas the OLS regression is characterized by a very high correlation coefficient (r2 = 0.997) and low uncertainty (2σ) on the calibration slope (2.52 ± 0.08), the York regression gives a much higher uncertainty on the slope (2.46 ± 0.34). This value, representing the 95% confidence interval (shown as dashed lines in the graph), gives a more conservative and probably realistic uncertainty envelope. Furthermore, close inspection of this particular regression (shown in the inset to Fig. 10) reveals that two samples (GRR999a and GRR1629-2) lie outside the boundaries of the 95% confidence limits. When these two samples are taken out of the regression, the error on the slope is reduced significantly (to 2.44 ± 0.14) and the mean square weighted deviation (MSWD) is improved from 4.8 to 1.3. However, we use both points for the reported parameters in Table 2 because both have σmean/σPoisson < 5. The reasons for this behavior for these two samples, which was reproduced in some other calibrations, are discussed further below.
SIMS detection limits can be determined for most elements from the background counts of the detector (i.e., counts outside the mass peak of interest; e.g., Hinthorne and Andersen 1975). For hydrogen, there can be additional contributions to the background from contamination of the vacuum, residual gas adsorption on the sample surface, H deposition on the immersion lens (the “memory effect”), and H desorption from the electron flood gun. Therefore, it is important to measure the combined background signal on “blank” standards demonstrated to have low water contents. In the following discussion, we first evaluate the blank standards used in the present study, and then assess different methods for calculating detection limits.
Evaluation of blank standards (GRR1017, GRR997, and DSCOL)
GRR1017 is an end-member forsterite synthesized by Shankland (1967) using the flame fusion method. We obtained polarized IR spectra on an oriented boule with path lengths of 11.96 and 12.16 mm in the  and  directions, respectively. The sample was almost completely featureless in the O-H vibrational region (Fig. 1). A small peak at 3320 cm−1 (blown-up region in Fig. 1) could represent OH in the structure but could also be the third overtone of a silicate fundamental at ~880 cm−1 (with first and second overtones at 1679 and 2499 cm−1). In any case, with an integrated absorbance of 0.04/cm, this band would represent only 7 ppb H2O using the calibration of Bell et al. (2003), emphasizing the low detection limit of IR under these conditions. Accordingly, we assign a nominal value of 0 ppm H2O for the OH concentration of the sample (Table 1).
GRR997 is a large olivine crystal from San Carlos, Arizona, with IR spectra originally recorded by Miller et al. (1987), measured through path lengths of 25.06 and 9.53 mm in the  and  directions, respectively. At these thicknesses, this nearly dry sample shows typical absorbance bands for olivine (Table 1) as well as a small peak at 3675 cm−1, which may represent a hydrous phase such as talc. The corresponding H2O concentration is 0.3 ± 0.1 ppm.
The DSCOL standard was prepared by firing San Carlos olivine in a 1 atm furnace at 1300 °C for 24 h, with oxygen fugacity fixed near the Ni-NiO buffer by a CO-CO2 gas mixture (cf. Demouchy and Mackwell 2006). IR spectra reveal no OH, with a detection limit of ~1 ppm (the path lengths for spectra were about 1 mm); we assume a value of 0 ppm (Table 1).
SIMS analyses of all three of these low-hydrogen content standards are characterized by high internal precision and good spot-to-spot reproducibility. GRR1017 yielded systematically lower 16O1H/30Si ratios compared to GRR997 measured at similar times (Fig. 3). It is not clear whether the offset reflects real differences in H content. This seems plausible only in the case of calibration 21, for which the calculated detection limit (0.4 ppm H2O; see below) was very low and the 16O1H/30Si ratio for DSCOL (0.00069 ± 0.00020, 2σSD) was essentially identical to GRR1017 (0.00064 ± 0.00017, 2σSD) but barely within error of GRR997 (0.00097 ± 0.00010, 2σSD).
Calculation of detection limits
We follow the methodology of the International Union of Pure and Applied Chemistry (IUPAC) and American Chemical Society (ACS) for calculating the limit of detection (LOD) and limit of quantitation (LOQ) (Long and Winefordner 1983). The IUPAC and ACS definitions state that
where xL is the measured quantity (here, 16O1H/30Si), xB is the mean value of the blank, σB is the standard deviation of the blank measures, and k is a numerical factor, either 3 for LOD or 10 for LOQ. The value of 10 for LOQ gives a conservative estimate of the lowest level at which a concentration can be accurately quantified, rather than just detected. For σB, we take either the standard deviation of the 30 cycles of one measurement on the 7f-GEO, or the standard deviation of multiple analyses for NanoSIMS measurements. The LOD and LOQ in units of ppm H2O are then calculated from xL by reference to the regression parameters. The resulting values (Table 2) range from 0.4 to 5 ppm H2O for LOD and 1.1 to 16.5 for LOQ (Table 2).
Previous SIMS studies have taken different approaches to determining the detection limit or “background” for hydrogen. Perhaps most commonly, the detection limit is estimated from the intercept of a linear regression of data on standards (e.g., Ihinger et al. 1994). Koga et al. (2003) noted that this method can be inaccurate—particularly when background levels are much lower than the hydrogen contents of the standards—because uncertainties in regression can lead to high uncertainties on the intercept. Indeed, detection limits calculated using this scheme are negative for some of our calibrations (Table 2), which has no physical meaning. An alternative method for calculating the detection limit (Hinthorne and Andersen 1975; Hauri et al. 2002; Koga et al. 2003; Saal et al. 2008) or “background” (e.g., Aubaud et al. 2007; Tenner et al. 2009) entails using the calibration slope (forced through the origin and/or calculated with blank-corrected data) to determine the effective H2O concentration of a blank standard based on its uncorrected mass ratio. Our calculated values using this approach (Table 2) range from 1.2 to 18.3 ppm. These values are poorly correlated to those calculated using the IUPAC/ACS method described above, but generally are significantly higher than the LOD and closer to the LOQ. We conclude that this method may be a reasonable approach to estimating the “background” and level of practical quantitation, but usually overestimates the real limit of detection; we prefer the IUPAC/ACS method because it explicitly takes into account the signal:noise of the blank measurement. The utility of this approach is exemplified by calibration 21, for which the LOD (0.4 ppm H2O) is more consistent than the other methods with the fact that we measured values for the two blank standards with 0 ppm that are statistically different from those of GRR997, with 0.3 (±0.1) ppm.
Evaluation of higher water content standards
GRR999a (14 ppm H2O)
This is a gem-quality olivine of hydrothermal origin from Zabargad Island, Egypt. Olivines from this locality exhibit unusual absorption spectra (Fig. 4; see also Miller et al. 1987 and Mosenfelder et al. 2006b) dominated by a prominent band at 3230 cm−1, strongly aligned parallel to . This group II band has been attributed variably to OH groups associated with Si vacancies (Beran and Putnis 1983) or M-site vacancies (Lemaire et al. 2004), molecular water (Matsyuk and Langer 2004), and humite-like defects or (unidentified) inclusions (Mosenfelder et al. 2006b). Although fluid inclusions have been documented previously in olivines from this locality (Clocchiatti et al. 1981), the crystal we examined contained no obvious fluid inclusions and no pores were found on a freshly fractured surface using SEM. TEM studies (Beran and Putnis 1983; Mosenfelder et al. 2006b) also failed to identify hydrous inclusions of any type in Zabargad olivines, although the features imaged by Beran and Putnis (1983), attributed to screw dislocations, could conceivably be voids or inclusions on the order of a few nanometers in diameter.
GRR999a yields 16O1H/30Si ratios with high internal precision (2–5% for the 7f-GEO; 1–2% for the NanoSIMS). However, the analyses follow a bimodal distribution; on the 7f-GEO, half the analyses give blank-corrected 16O1H/30Si = 0.0025 ± 0.0002 and half give 0.00303 ± 0.00004 (2σSD). A similar distribution is seen for NanoSIMS calibrations 21 and 22. When the analyses with lower values are eliminated from the York regressions, the MSWD and uncertainty in slope are significantly reduced, as noted above. Given that about 80% of the absorbance in this sample is from the band at 3230 cm−1, lower than expected 16O1H/30Si is consistent with an increase in absorption coefficient with decreasing wavenumber, as expected from Paterson (1982) and Libowitzky and Rossman (1997). The bimodal distribution and high precision of our analyses, however, prevents us from making this interpretation unambiguously. Although the deviations from best fit amount to ~4 ppm H2O (30% of the total H2O content of the sample), the overall effect on calibration slopes of including or not including GRR999a is minimal. Therefore, for practical purposes this standard is useful for constraining calibrations in this concentration range.
GRR1695-2 (16 ppm H2O)
This olivine from the Black Rock Summit, Nevada, volcanic field is one of the three standards analyzed by NRA (Bell et al. 2003). IR spectra of GRR1695-2 (Fig. 4) show almost exclusively group Ia bands, typical for upper mantle olivines; notably, no bands attributable to hydrous inclusions are present. However, SEM investigation (Fig. 11) revealed dispersed pores ranging in size from about 0.1 to 5 μm in diameter. No precipitates were identified within the pores. The spatial distribution and negative crystal shape of the pores suggest that they are primary fluid inclusions (Roedder 1984). We did not study the fluid composition. Although fluid inclusions in olivine have traditionally been considered to be nearly pure CO2, recent studies indicate that many other volatile species can be present in smaller amounts (e.g., Andersen and Neumann 2001).
GRR1695-2 was analyzed repeatedly on both the 7f-GEO (23 analyses) and the NanoSIMS (15 analyses). The internal precision of the analyses was lower overall than for GRR999a, ranging from 2 to 8%. Some depth profiles show a strong correlation between 16O1H and 19F (Fig. 12), uncorrelated with 12C or 30Si. The analysis shown in Figure 12 was filtered out of the regression (σmean/σPoisson = 7.7). The exceptional variability in 16O1H counts, correlated to 19F, probably reflects ionization of variable volumes of fluid inside the pores that we imaged using SEM. Although analysis of hydrous solid phases is easy to understand as a reason for such variation, analysis of fluids (or vapor) in inclusions via SIMS is an understudied topic. Under ultra-high vacuum, piercing of an inclusion by progressive sputtering might liberate the fluid/vapor phase too rapidly for effective ionization by the primary beam. On the other hand, residual adsorption of H2O and F on the walls of the inclusion might contribute to the overall ionization yield. Ancillary evidence for this possibility comes from the study of Diamond et al. (1990), who reported semi-quantitative SIMS analyses of dissolved alkali content (at very low dilution) in fluid inclusions in quartz.
The lowest precision of our analyses (8%) is significantly better than the uncertainty of the original NRA analysis (30%), which sampled a similar depth range (but with a much larger spot size, about 5 mm in diameter); this suggests that the scatter in the NRA data may be largely tied to experimental issues rather than heterogeneities in the sample. Therefore, despite lower precision compared to GRR999a, this standard is also adequate for constraining calibrations in this concentration range.
GRR1629-2 and GRR1784e (40 and 54 ppm H2O)
These two samples from Buell Park, Arizona, were selected from the study of Mosenfelder et al. (2006b) because their H concentrations fill a gap in previously characterized, natural olivine standards between ~10 and 125 ppm. Moreover, they lack detectable Ti-clinohumite inclusions or lamellae, in contrast to other crystals from this locality. Their IR spectra are dominated by group I bands (Figs. 4 and 5).
The precision of SIMS and NanoSIMS analyses on both standards is excellent (typically 1–2%; no analyses rejected based on Poisson statistics). However, consideration of all the analyses relative to other samples and the best-fit regressions suggests that one or both of the samples suffer from some heterogeneity. Contradicting the measured H2O from FTIR, the 16O1H/30Si ratios of GRR1629-2 are slightly higher than for GRR1784e for three out of eight SIMS calibrations in which they were both measured. The cause of the heterogeneity is unknown, but could comprise very small amounts of Ti-clinohumite not detected by FTIR. SEM investigations of freshly fractured surfaces of the samples failed to reveal any pores. We also took spectra on different sections of the crystals of varying thickness and found no differences in absorption coefficient. Despite this issue, both samples lie within error of the 95% confidence intervals for the York regressions. This result has important implications in the context of our discussion below on band-specific IR absorption coefficients.
ROM177 (125 ppm H2O)
This xenocryst is part of the group 2 (Fe-rich) megacryst suite from the Monastery kimberlite and is one of three samples from Bell et al. (2004) used as a secondary standard for SIMS (Aubaud et al. 2007; Tenner et al. 2009; Kovács et al. 2010). IR absorbance is nearly evenly split between group Ia bands (particularly the strong band at 3572 cm−1) and bands at higher wavenumbers (Fig. 5). Although we refer to these latter bands in Table 1 as “group I”, some of the band positions have not been well reproduced in the high-pressure experimental studies upon which that categorization was based. In particular, the sharp bands at 3668, 3637, and 3620 cm−1, which are commonly found in olivines from kimberlites (Matsyuk and Langer 2004), are rarely seen in experiments; the only exception we know of is the study of Withers and Hirschmann (2008), who show some spectra with a band near 3670 cm−1.
Possible explanations for these higher frequency bands include the high Fe content of the sample, coupled substitution with other elements not present in the experimental studies, or the presence of included hydrous phases. We can dismiss the first possibility because the spectra of ROM177 are very similar to those of SI-387, a Fo89 olivine from the study of Matsyuk and Langer (2004), and at least two of the three bands are also present in sample GRR1012-2 (Fo91). Coupled substitution is considered later in the paper, in the context of the correlation between hydrogen and fluorine concentrations. As for the third hypothesis, possible hydrous phase inclusions that could be attributed to the band at 3668 cm−1 include talc and/or 10 Å phase (Khisina et al. 2001; Matsyuk and Langer 2004; Koch-Müller et al. 2006). We have not identified any such hydrous phases using SEM (TEM is probably needed for this purpose). On the other hand, Bell et al. (2004) identified optically visible inclusions 1–3 μm in diameter that likely correspond to pores imaged using SEM on a fractured surface (Fig. 13). The pores range in size from ~200 nm to ~5 μm in diameter and have both rounded (Fig. 13a) and negative crystal shapes (Fig. 13b). They are aligned in sub-linear arrays (Fig. 13a), as opposed to the randomly dispersed pores in GRR1695-2; this suggests they are mostly secondary (or “pseudo-secondary”) origin fluid inclusions (Roedder 1984). Some of the pores are partially filled with precipitates (Fig. 13b), the composition of which we were not able to determine using EDS; these precipitates may be analogous to hydrous phases imaged with TEM in other olivines (Khisina et al. 2001).
Despite the features described above, ROM177 was one of our most reproducible standards on the 7f-GEO, yielding an average blank-corrected 16O1H/30Si ratio of 0.031 ± 0.004 (2σSD) with internal precision for most analyses in the range of 1–2%. For all 12 calibrations in which it was measured, ROM177 falls well within the uncertainty envelope of the York regression (Fig. 10). These results stand in sharp contrast to those of Tenner et al. (2009) and especially Kovács et al. (2010), who measured much lower 16O1H/30Si ratios on this standard: 0.017 and 0.007 ± 0.001 (3 analyses), respectively. Moreover, the analyses of ROM177 are very poorly fit by the proposed calibration lines in those studies. Analogous behavior is seen in the NanoSIMS calibration of O’Leary et al. (2010).
Our NanoSIMS analyses of ROM177 also had high internal precision, but reproducibility was poorer than expected based on the consistency of our 7f-GEO measurements (Fig. 9). For calibrations 21 and 22, ROM177 is poorly fit by the York regressions, falling to lower than expected average 16O1H/30Si values. However, the deviation from the best fit is markedly less than in other studies.
The above observations suggest that the ROM177 megacryst may be highly heterogeneous. However, we note that we have collected FTIR spectra on multiple sections of ROM177, including the chip we used for NanoSIMS, and that they are indistinguishable from the original spectra collected by Bell et al. (2004). Based on our results alone, we thus judge this to be an excellent standard. For useful inter-laboratory comparison, however, the OH concentrations for the chips used by other laboratories should be re-measured using FTIR.
KLV23 (140 ppm H2O)
KLV23 from the Kaalvallei kimberlite, South Africa, was analyzed by NRA (Bell et al. 2003) and is also used by the Carnegie lab as a SIMS standard. The IR spectra of KLV23 (Fig. 6) are dominated by group Ia bands at 3572 and 3523 cm−1. However, there is also a band at 3708 cm−1, which Bell et al. (2003) assigned to hydrous inclusions of an unspecified sheet silicate. The band constitutes about 10% of the total absorbance and was included in the integration, under the assumption that the near-surface NRA analysis sampled the inclusions, whose spatial distribution was unknown.
The band at 3708 cm−1 can be matched to bands for several hydrous phases. Pure phlogopite shows a narrow peak at ~3710 cm−1 (Serratosa and Bradley 1958); however, even small amounts of Fe or Al substitution result in the appearance of other bands at lower wavenumbers. Moreover, we failed to detect any K in the sample above the detection limit (~200 ppm) of WDS and EDS analyses. A band system at 3709–3711 cm−1 in olivine was assigned by Matsyuk and Langer (2004) to Mg-rich amphibole (pargasite or edenite), but amphiboles also exhibit strong bands at lower wavenumbers (at ~3675 cm−1) that are present in their samples but not in KLV23. Another possibility is brucite, which has a band at ~3700 cm−1 (Fig. 6; Shinoda and Aikawa 1998).
We examined a polished plate as well as two freshly fractured surfaces of KLV23. No included phases were detected using BSE imaging on the polished plate. The fractured surfaces, on the other hand, contain numerous pores ranging in diameter from <100 nm to ~2 μm (Fig. 14a). Some of the pores are partially filled with precipitates (Fig. 14b). We attempted to measure the composition of the precipitates using EDS, but as with ROM177 they are too small (less than ~100 nm) to permit unambiguous analyses unaffected by the surrounding olivine. Moderate concentrations of S (2–5 wt%) were detected in some analyses. Elevated S suggests the presence of sulfides, which are common precipitates in fluid inclusions in olivine (Andersen and Neumann 2001).
These features were also imaged using NanoSIMS. Ion images of previously sputtered and analyzed pits show “hot spots” of elevated 16O1H (Fig. 14c) and 19F (Fig. 14d) counts. The size of the features is consistent with the pores imaged using SEM.
We speculate that the precipitates in the pores are multiphase aggregates, with Mg(OH,F)2 brucite being the dominant phase. Brucite has not been reported previously as an inclusion in olivine and is not expected from known phase relations. However, it could be precipitated from an Mg-rich fluid trapped at high pressures. Experimental studies show that aqueous fluids in equilibrium with peridotite assemblages abruptly change from Si-rich to Mg-rich compositions at pressures above ~3 GPa (e.g., Kawamoto et al. 2004).
KLV23 is one of our least reproducible SIMS standards. Depth profiles in the sample are highly erratic, resulting in low internal precision (as poor as 16%). Only one of 19 analyses met our criterion to avoid filtering for regression method 3. 16O1H is correlated strongly to 19F, both in bulk analyses (as seen in depth profiles) and on the scale of the NanoSIMS images of previously analyzed craters (Fig. 14). Based on the low precision of all KLV23 analyses in this study and the poor fit with other standards, we recommend that KLV23 be abandoned for use as a primary SIMS standard, despite the fact that it gives sensible results using NRA (Bell et al. 2003).
ROM250-OL2 (183 ppm H2O)
This olivine is from the “main silicate trend” of the Monastery megacryst suite (Bell et al. 2004). IR spectra of ROM250-OL2 (Fig. 5) look similar to KLV23 (Fig. 6) except that the peak at 3708 cm−1 is barely above the detection limit. Pores as small as 200 nm in diameter were identified with SEM on a fractured crystal surface but were much less abundant than in other samples discussed above. No precipitates were found within the pores.
The internal precision of SIMS analyses for ROM250-OL2 ranged from 6–15%, and all analyses were rejected from regression method 3 based on Poisson counting statistics. The poor performance of this standard is exemplified by the wide spread in 16O1H/30Si values—disagreeing within 2σ uncertainties—among five sequential analyses for calibration 19 (Fig. 8a). Kovács et al. (2010) also report a large spread in 16O1H/30Si for this standard (Fig. 2). As with previously discussed samples, the extreme variations in 16O1H are strongly correlated to 19F, suggesting that variable volumes of hydrous inclusions were ionized. The sample face polished for SIMS probably contained a higher density of pores than the cleavage face that we imaged using SEM. The high uncertainties on both IR (Table 1) and SIMS data for this standard make it a poor choice for calibration.
GRR1012-2 (220 ppm H2O)
This xenocryst from the Kimberley mine, South Africa is the highest H2O content standard measured by NRA and is also used by the Carnegie lab as a standard. IR spectra of GRR1012-2 (Fig. 5) are dominated by group Ia bands but there are also prominent bands at 3685, 3673, 3636, and 3622 cm−1. The first of these bands has a long tail of absorption (with a shoulder at ~3710 cm−1) and is attributed to serpentine (Miller et al. 1987; Matsyuk and Langer 2004). SEM investigation failed to reveal distinct zones of serpentine or pores in the sample; TEM would be necessary to characterize its spatial distribution. We also note that another xenocryst from the same locality (GRR920, Fig. 1 in Miller et al. 1987) has a milky appearance and an exceptionally intense, broad, underlying absorption band characteristic of liquid H2O, which Miller et al. (1987) attributed to the presence of submicroscopic fluid inclusions.
Most of our 7f-GEO analyses of GRR1012 have errors in the range of 0.6–4%, with blank-corrected 16O1H/30Si ratios of 0.054 ± 0.005 (2σSD). However, some analyses with higher errors (4–6%) also yield higher 16O1H/30Si (0.07 ± 0.02); these analyses failed our criterion for inclusion in the York regressions. Analogous behavior is seen in the NanoSIMS analyses. Unlike other standards, 16O1H is not correlated with 19F. This suggests heterogeneities caused by hydrous inclusions not containing F, most likely the serpentine seen in the IR spectra. GRR1012 gives the highest spread in 16O1H/30Si ratios for the calibration of Kovács et al. (2010), ranging from 0.048 to 0.110 with analyses disagreeing within error bars. Our results suggest that this standard is useful for SIMS calibration as long as the Poisson counting statistics are considered carefully.
ROM250-OL13 (245 ppm H2O)
The chemistry and IR spectra (Fig. 5) of this megacryst are similar to ROM177. Bell et al. (2004) observed abundant micrometer-sized inclusions in the sample. SEM investigation of a freshly fractured surface revealed isolated pores, albeit with lower density than ROM177. We also imaged sub-micrometer sized pores within one of the sputtered craters on the polished slab used for SIMS. The corresponding analysis, which was rejected based on Poisson error, had an 16O1H/30Si ratio of 0.110 ± 0.008 (blank-corrected, 2σmean), almost twice the value of five other analyses that had internal precision of 0.5–5.6% and were not rejected (average 16O1H/30Si = 0.060 ± 0.006). Another analysis that was rejected, with 16O1H/30Si = 0.072, is shown in Figure 7 and discussed above. Overall, this standard provided SIMS data comparable to GRR1012-2; however, the much higher uncertainty on the H2O concentration from FTIR makes it less useful for precisely constraining calibration lines.
Fluorine and hydrogen incorporation in olivine
Previous studies have shown correlations between hydrogen and fluorine in olivine and other nominally anhydrous minerals (Hervig and Bell 2005; Guggino et al. 2007; Bromiley and Kohn 2007). Depth profiles for some analyses in this study also show such a correlation (Fig. 12), and we have postulated that this reflects analysis of non-intrinsic phases in olivine—either fluids or solid hydrous mineral inclusions. On the other hand, apparently “uncontaminated” analyses also reveal high levels of fluorine that may be structurally bound. We did not calibrate for fluorine on the 7f-GEO, but calibrations were conducted on the NanoSIMS using basalt glasses as standards. In the following discussion we refer to 19F/30Si ratios, but for context the calibrated fluorine concentrations of ROM177, KLV23, and GRR1012-2 on the NanoSIMS are 50, <3, and 65 ppm, respectively (Le Voyer et al., unpublished data).
Figure 15 shows 19F/30Si as a function of hydrogen content (as wt% H2O) for calibration 1. When all standards are considered, the correlation is very poor. However, it is excellent (r2 = 0.996) when only three olivines with high F/Si ratios are considered: ROM177, GRR1012-2, and ROM250-OL13. As previously discussed, these three olivines are distinguished from the other samples by strong absorption peaks near 3668, 3637, and 3620 cm−1. The latter two bands are also present, but close to the detection limit, in KLV23 and ROM250-OL2 (Figs. 5 and 6, Table 1). We also show in Figure 5 the spectrum from Sykes et al. (1994) of GRR1677, an olivine containing exceptionally large concentrations of boron (1.1–1.4 wt% B2O3) and fluorine (0.3–0.6 wt%). The peaks at 3670 and 3637 cm−1 are more intense in this sample than in any other olivine measured in the literature; a peak at ~3620 cm−1 might also be present on the shoulder of the 3637 cm−1 band. This suggests that one or more of these O-H vibrations represent H in sites affected by nearby F and/or B. Sykes et al. invoked the coupled substitution B(F,OH)Si−1O−1, but a boron-free, coupled substitution mechanism is also possible. Although B was not measured in the present study, Kent and Rossman (2002) found very low levels of B in mantle-derived olivines; for instance, they report 0.06 ppm B in GRR920, which is from the same locality and has spectra similar to GRR1012-2.
Hervig and Bell (2005) suggested that inconsistencies in F-H correlations such as that seen in Figure 15 might be due to preferential loss of H during ascent. Although most olivines from kimberlites appear to have suffered little H loss (Peslier et al. 2010), Bell et al. (2004) documented differences in hydrogen content between petrographically different grains in a sample from Monastery (ROM250-OL42) indicating hydrogen loss during recrystallization. However, the samples studied here show no such evidence. Another possibility is that the inconsistencies in correlations simply reflect differences in availability of the respective volatile elements within the source region.
Band-specific IR absorption coefficients
The wide frequency range of O-H absorption bands in olivine calls into question the validity of using a single absorption coefficient (ɛ)—as in this study—because global IR calibrations (Paterson 1982; Libowitzky and Rossman 1997) demonstrate that there is a significant wavenumber dependence to ɛ for OH in minerals. We address this question here in the context of two previous studies (Koch-Müller et al. 2006; Kovács et al. 2010).
Koch-Müller et al. (2006) studied a suite of olivines from Russian kimberlites, using SIMS and FTIR. Comparing H2O concentrations based on their SIMS calibration to the absorbance of the samples (Fig. 16), they derived a value of ɛ (37 500 ± 5000 L/mol/cm2) larger, within stated uncertainties, than the value (28 450 ± 1830 L/mol/cm2) derived by Bell et al. (2003). They ascribed this discrepancy to the difference in “mean wavenumber” between their samples and those studied by Bell et al. and recommended that the new value be used for olivines with spectral characteristics similar to the ones they studied. Most of their olivines have spectra similar to ROM177 and ROM250-OL13, with absorbance nearly evenly split between group Ia bands and the higher wavenumber bands that we attribute to fluorine incorporation. However, the SIMS calibration was performed using a suite of garnets analyzed with NRA by Maldener et al. (2003), who derived an absorption coefficient for pyrope garnets much different than that measured by Bell et al. (1995) using manometry. Moreover, standardization using garnet alone fails to take into account matrix effects, which can be substantial; other SIMS studies document differences of 2–5 in calibration slopes between pyrope and olivine (Koga et al. 2003; Aubaud et al. 2007; Tenner et al. 2009; Mosenfelder et al., unpublished data). We also note that the presence of hydrous phases (serpentine, talc, and 10 Å phase) may contribute to scatter in the data of Koch-Müller et al. (2006), as reflected by the error bars (1σSD, Fig. 16 inset).
Kovács et al. (2010) conducted a combined FTIR and SIMS study of synthetic olivines and one natural sample showing an even wider variation in spectral characteristics. Most of their samples are dominated by group I bands, but one Ti-doped olivine shows group Ia bands and other samples contain a mix of group I and II bands. The natural sample, from Pakistan, was described as having [Si] defects (group I), but in fact samples from this locality have IR bands unlike those assigned in other samples to [Si] defects; the bands have a different polarization (strongest in the α orientation rather than γ; Kovács et al. 2008; Gose et al. 2010) and slightly higher full-width half maxima (FWHM). Moreover, these crystals and olivines with nearly identical spectra from Kingiti, Tanzania, (Miller et al. 1987 and unpublished data) have unusually high boron concentrations (Kent and Rossman 2002; Gose et al. 2010), suggesting that the OH vibrations should be assigned to coupled B3+-H+ substitution rather than [Si] defects.
Kovács et al. (2010) concluded from multiple regression of their data that there is a factor of 3 difference in absorption coefficient for group I vs. group Ia bands, even though the differences in average wavenumber between these groups are not large. Furthermore, they determined that the absorption coefficient for group Ia bands was identical to that of the group II bands associated with trivalent cations. A simple linear regression of the data just for samples with group I bands (Fig. 16) suggests an even larger difference compared to group Ia. The slope of the regression line is 3.8 times that of the line defining the Bell et al. calibration. If true, this difference in absorption coefficients would have a remarkable impact, not fully assessed by Kovács et al. (2010). Natural olivines dominated by group I bands (e.g., Mosenfelder et al. 2006b) would contain significantly more water than previously recognized. Moreover, nearly all olivines synthesized or annealed in experimental studies above ~3 GPa are dominated by group I bands. The implied H2O concentrations for samples recovered from the highest pressures (Mosenfelder et al. 2006a; Smyth et al. 2006) using the new absorption coefficient would be in the range of 1.5–3 wt% and thus strain credulity for concentrations within the structure of olivine, given that concentrations in the high-pressure polymorphs of olivine (wadsleyite and ringwoodite) are within this same range.
The conclusions of Kovács et al. (2010) conflict with the present study because we have also measured olivines with a wide variation in spectral characteristics and all the data for any given calibration are well fit by a single line (after consideration of data filtering to account for analysis of hydrous inclusions, where necessary). This discrepancy cannot be explained by the lower precision of the olivine SIMS calibration presented by Kovács et al. (2010) because any absolute difference in concentrations is irrelevant to the relative differences (i.e., ratio of absorption coefficients) they measured among olivines with different spectra. Another possible explanation invokes the difference in spectroscopic methods between our study (using polarized IR spectra) and theirs (using unpolarized spectra and an averaging method with corrections applied; Kovács et al. 2008). This is also unlikely to explain the difference, however, because of the extreme differences in measured concentrations between their SIMS values and the values expected from the Bell et al. calibration. Yet another possibility is that there is a problem in extrapolation of the calibration from the relatively low concentrations measured in standards (<250 ppm H2O) to the concentrations of the experimental samples studied by Kovács et al. (up to 4400 ppm H2O). However, other calibrations in the literature (Koga et al. 2003; Aubaud et al. 2007; Tenner et al. 2009) that employed very “wet” experimental samples as secondary standards—while less precise than the calibrations in the present study—do not suggest significant non-linearity in calibration slopes over the relevant concentration range.
The most likely explanation for the discrepancy between this study and Kovács et al. (2010) is that some of their analyses are affected by ionization of hydrous inclusions and/or surface contamination. Neither of these possibilities is easy to assess quantitatively given the lack of primary data presented in the paper (i.e., no 12C counts or 16O1H/30Si ratios with uncertainties for individual analyses). However, contamination of analyses by hydrous inclusions is likely given that four out of five of the highest H2O concentration samples shown in Figure 16 exhibit significant bands assigned to serpentine (as a quench phase?); indeed, the spectrum of one sample (p15_1100, shown in their Fig. 1) shows that ~50% of the absorbance is from serpentine and our experience suggests such a large volume of inclusions would be difficult to avoid during crater sputtering in the SIMS.
We conclude that, within the errors of SIMS calibrations—which are dominated by the uncertainties in H2O concentrations of the standards—there is no evidence for a wavenumber dependence of the absorption coefficient for olivines dominated by group I vs. group Ia bands. On the other hand, in the section on sample GRR999a we presented marginal evidence for such dependence for group II bands at much lower wavenumbers. Verification of this possibility will require further study of multiple samples with group II bands, at higher overall H2O concentrations.
Recommended protocols for SIMS analysis of hydrogen
Low-blank analysis of hydrogen using SIMS is now a mature technique, with well-established sample preparation and analytical protocols (Hauri et al. 2002; Koga et al. 2003; Aubaud et al. 2007). Nevertheless, our study shows that there is significant room for improvement in careful characterization of standards and assessment of data quality.
We have documented sub-micrometer to micrometer scale heterogeneities—caused by hydrous inclusions—in previously used standards. Because of the scale and distribution of the heterogeneities, SIMS analyses can be influenced more significantly (and randomly) than FTIR or NRA analyses of the same standards, which average over much larger volumes. This explains the apparent contradiction that samples such as KLV23 are appropriate standards for NRA but not always accurate for SIMS and we thus disagree with the implication that the NRA standards alone define the true calibration curve (Kovács et al. 2010; Hauri and Hermann, pers. comm.). There are many more possible mechanisms for heterogeneous hydrogen distribution in olivine caused by nanometer-scale hydrous inclusions (e.g., Khisina et al. 2001; Matsyuk and Langer 2004; Mosenfelder et al. 2006b), and such heterogeneities are not unique to this phase. For instance, nanometer-scale amphibole lamellae are a common occurrence in pyroxenes (Skogby and Rossman 1991) that can “contaminate” SIMS analyses (Mosenfelder et al., unpublished data). Simultaneous measurement of 19F facilitates assessment of these possibilities because F co-substitutes with H in most hydrous minerals. Measurement of 12C is also critical to determining contributions of H from organic contamination on sample surfaces or cracks. 19F, 32S, and 35Cl, used in other studies as additional discriminants for contamination, are redundant and/or less sensitive compared to 12C, particularly in the case of olivines with high amounts of fluorine.
We also emphasize the need to consider the Poisson counting statistics for individual analyses; this is standard protocol for many SIMS applications such as stable isotope measurements (Fitzsimmons et al. 2000), but is not widely mentioned or possible to assess in retrospect for SIMS studies of hydrogen that do not report primary data and/or analytical uncertainties. We have derived a simple, empirical criterion for rejecting analyses based on Poisson statistics that provides more consistent results than in previous work.
Finally, we recommend that the uncertainty of calibration curves—which should be propagated into the uncertainty of hydrogen analyses on unknowns—is best determined using a method such as York regression (cf. Koga et al. 2003), which takes into account the precision of both SIMS analyses and the “known” H content of the standards. The uncertainties for our York regressions are dominated by scatter in the data and/or the uncertainties in H2O concentration determined by FTIR or NRA; individual uncertainties on the 16O1H/30Si ratios for filtered data have little effect on the end result. Therefore, improvement in the precision of SIMS measurements in the future will depend primarily on improving the precision of absolute concentrations for primary standards (and thus improving the IR calibration, for secondary standards). Further efforts toward absolute calibrations, using manometry or nuclear methods such as NRA, elastic recoil detection analysis, or proton-proton scattering (see Rossman 2006 for review of these methods), are still highly desirable.
Financial support for this research was provided by the Gordon and Betty Moore Foundation, the White Rose Foundation, and by NSF grant EAR-0947956 to George Rossman.
We thank Erik Hauri, Jeremy Boyce, and Richard Hervig for helpful discussions; Chi Ma for assistance with the electron microprobe and SEM; and Jörg Hermann and an anonymous reviewer for helpful suggestions that improved the manuscript.
↵1 Deposit item AM-11-058, Supplementary materials. Deposit items are available two ways: For a paper copy contact the Business Office of the Mineralogical Society of America (see inside front cover of recent issue) for price information. For an electronic copy visit the MSA web site at http://www.minsocam.org, go to the American Mineralogist Contents, find the table of contents for the specific volume/issue wanted, and then click on the deposit link there.
- Manuscript Received February 17, 2011.
- Manuscript Accepted July 5, 2011.