- © 2013 Mineralogical Society of America
Pyroxenes have the capacity to incorporate both hydrogen and fluorine in their structures, and accurate measurement of these volatile elements can be used to constrain geophysical and petrologic processes in planetary bodies. For example, pyroxenes may be used to constrain the volatile contents of melts from which they crystallized. However, the experimental determination of H and F in pyroxenes is difficult, particularly at the relatively low levels present in natural samples. Here we evaluate methods for determining both H and F in orthopyroxene.
We measured trace concentrations of H (~40–400 ppm H2O) and F (<1–17 ppm) in a suite of nine orthopyroxenes from varying geological environments, using secondary ion mass spectrometry (SIMS). The SIMS data for H (measured as 16O1H, referenced to 30Si and 18O) are cross calibrated against Fourier transform infrared (FTIR) spectra, in turn calibrated against either manometry (Bell et al. 1995) or the frequency-dependent molar absorption coefficient derived by Libowitzky and Rossman (1997). Despite the fact that our samples exhibit a wide range of IR band structures, with varying percentages of absorbance split among low (2600–3350 cm−1) and high (3350–3700 cm−1) wavenumber bands, the SIMS data are fit with the same precision and virtually the same regression slope regardless of which IR calibration is used. We also confirm previous suggestions that the matrix effect for SIMS analyses between orthopyroxene and olivine is small (≤20%). Anomalously high yields of 16O1H in some analyses can be attributed to the presence of amphibole lamellae, and these analyses must be filtered out with different criteria than for olivine due to differences in the geometrical relationship of host to inclusion. For F, our derived values are highly dependent on analytical uncertainties related to the use of silicate glasses as standards. Regardless of the accuracy of our calibration, we see systematic differences in F concentrations in orthopyroxenes and olivines depending on their geological context. Samples derived from crustal environments and from Colorado Plateau minette diatremes have very low F (≤3 ppm), while higher contents can be found in megacrysts from South African kimberlites (up to 17 ppm in orthopyroxene and 47 ppm in olivine) and in xenocrysts from the Rio Grande Rift (Kilbourne Hole, 7–9 ppm in orthopyroxene).
It is now firmly established that the major rock-forming minerals of the Earth’s upper mantle—olivine, orthopyroxene, clinopyroxene, and garnet—have the capacity to incorporate hydrogen into their structures in the form of structurally bound hydroxide groups, at levels ranging from trace amounts (Bell and Rossman 1992) up to thousands of ppm H2O (Kohlstedt et al. 1996; Mosenfelder et al. 2006a; Smyth et al. 2006; Mierdel et al. 2007; Withers and Hirschmann 2008; Withers et al. 2011). Accurate modeling of processes influenced by H in these nominally anhydrous minerals (NAMs)—such as deformation and melting—relies on accurate measurement of H in both natural and experimentally produced samples. Some recent studies of natural samples have focused on pyroxenes, which typically have more H than coexisting olivine—sometimes to a degree unexpected from experimental partitioning data (Peslier and Luhr 2006). This implies that pyroxenes better retain H during decompression and cooling compared to olivine. Based on this inference, they are thus a better choice for estimating the water content of the magma from which they crystallized or of the solid source from which they were extracted (Peslier et al. 2002; Wade et al. 2008; Gose et al. 2009; Warren and Hauri 2010; Nazzareni et al. 2011; Sundvall and Stalder 2011). On the experimental side, considerable progress has been made toward determining storage capacities, incorporation mechanisms, and diffusion rates for H in clinopyroxene and orthopyroxene, particularly in the latter (Rauch and Keppler 2002; Stalder 2004; Stalder and Skogby 2002, 2003; Stalder et al. 2005, 2007; Mierdel and Keppler 2004; Mierdel et al. 2007).
In addition to H, recent studies have shown that pyroxenes can contain—apparently as structurally bound entities, not just in inclusions—significant amounts of boron, fluorine, and chlorine (Hervig and Bell 2005; Hauri et al. 2006a; Hålenius et al. 2010; O’Leary et al. 2010; Bernini et al. 2012; Beyer et al. 2012; Dalou et al. 2012). Due to their high solubility in aqueous fluid and/or silicate melts, these light elements are also of interest to geochemists for understanding chemical recycling and melting processes in the Earth and other planetary bodies. Recently reported concentrations in synthetic pyroxenes range up to 2000 ppm B (Hålenius et al. 2010), 600 ppm F (Dalou et al. 2012), and 80 ppm Cl (Dalou et al. 2012). Moderate amounts of F have also been identified in pyroxene (and olivine) megacrysts from kimberlites (Hervig and Bell 2005; Guggino et al. 2007; Mosenfelder et al. 2011; Beyer et al. 2012). These studies used low-blank methods with high sensitivity and the capability to discriminate surface contamination: B was measured by nuclear reaction analysis (NRA), and F and Cl were measured with secondary ion mass spectrometry (SIMS).
In this study, we focus on quantification of H and F in pyroxenes using two techniques, Fourier transform infrared (FTIR) spectroscopy and SIMS. Our results on orthopyroxene and clinopyroxene are presented in part I herein and part II (Mosenfelder and Rossman 2013, this issue), respectively. FTIR is a well-established technique used routinely for measuring H (but not commonly used for F). However, significant problems remain with accurate quantification using this method. The utility of SIMS for measuring H and F in minerals has been long recognized (Hinthorne and Andersen 1975). However, measurement of low H concentrations, such as in many natural NAMs, has only recently become tractable as a result of advances in instrumentation, analytical protocols, and sample preparation techniques (Kurosawa et al. 1997; Hauri et al. 2002; Koga et al. 2003; Aubaud et al. 2007). Here we present SIMS data on simultaneously measured H and F in nine different orthopyroxenes, acquired using the methods outlined in Mosenfelder et al. (2011). We also compare these data to measurements taken on olivine during the same session, as well as calibrations presented by other laboratories. We assess possible sources of analytical uncertainties for both SIMS and FTIR, examine the possibility that the IR absorption coefficient for O-H bonds is frequency dependent, and finally discuss implications of our work for F distribution in the Earth’s mantle and crust.
Sample preparation, electron probe microanalysis (EPMA), and FTIR
The orthopyroxenes and glasses we analyzed are listed in Tables 1 and 2, respectively; details on the olivine samples are in Mosenfelder et al. (2011). Sample preparation and cleaning protocols followed the methods in Mosenfelder et al. (2011). Minerals were oriented for polarized FTIR measurements using optical methods (examination of cleavage, morphology, pleochroism, and extinction/interference figures under cross polarized light). Spindle stage methods, used for orienting clinopyroxene in part II of the study, were not necessary for the orthopyroxenes used here, many of which were large gem-quality crystals. A single exception is sample KBH-1, oriented by Bell et al. (1995). We estimate the accuracy of orientations to be ±2°, with the major source of uncertainty coming from cutting and polishing. Orientations were also confirmed using silicate overtone spectra, examples of which are shown in the Appendix1. We follow the conventions α = nα = X ||  ~ 9 Å, β = nβ = Y ||  ~ 18.3 Å, and γ = nγ = Z ||  ~ 5.2 Å, and use the Greek letters to designate polarized spectra with the E vector parallel to the given optic/crystallographic direction.
Our procedures for EPMA and FTIR spectroscopy have been described previously (Mosenfelder et al. 2006a, 2006b, 2011). EPMA data are presented in Table 3 and FTIR data are included in Table 1. Most FTIR spectra were acquired in the main compartment of the spectrometer with light polarized by a LiIO3 Glan-Foucault prism; some spectra were taken in the microscope. Furthermore, a CaF2 wire-grid polarizer was used in some cases, which allowed us to measure absorbance in both the silicate overtone and OH-vibrational regions (see Appendix1). Spectra taken with these different experimental setups were consistent where comparisons were made. Details of our baseline correction methods are discussed in the Appendix and the baselines (and corrected spectra) are provided in the supplementary material1.
Hydrogen concentrations (given as ppmw H2O) were determined from FTIR data using the Beer-Lambert law and two different calibrations for the integrated molar absorption coefficient (ɛi): the manometry-based calibration of Bell et al. (1995) and the generic, wavenumber-dependent calibration of Libowitzky and Rossman (1997). The value for ɛi is 80 600 ± 3200 (1σ) L/molH2O/cm2 in the former case, while in the latter ɛi (in L/molH2O/cm2) = 246.6(3753 − ν), where ν is wavenumber in cm−1.
Sample densities were measured for most samples via Archimedes’ method using immersion in toluene, and these values were used in the estimation of H concentrations. For some samples that were too small and/or fractured to measure accurately this way, we assumed the same density as measured by Bell et al. (1995) for KBH-1 (3.318 g/cm3). The measured differences in density between our samples (Table 1) result in very little change in the final estimated H concentrations, on the order of 1–2 ppm H2O compared to assuming the same calibration factor derived for KBH-1.
Final uncertainties in H concentrations calculated using the Bell et al. (1995) calibration were estimated by propagating the 2σ uncertainty in the absorption coefficient (8% relative) with the uncertainty in total absorbance, which was estimated individually for each sample as discussed in the Appendix. Uncertainties in density were inconsequential when propagated and uncertainties in sample thickness were ignored. The resulting uncertainties range from 10–14%. We assumed the same relative uncertainties for concentrations calculated using the Libowitzky and Rossman (1997), for the sake of comparing calibration lines; the actual uncertainty when using this calibration is difficult to assess because the uncertainty in baseline varies depending on wavenumber.
SIMS analyses were obtained on the Cameca 7f-GEO at the Center for Microanalysis at Caltech. We used a mass resolving power of ~5500 (M/ΔM), sufficient to separate the peaks for 16O1H and 19F from 17O and 18O1H, respectively. For all analyses we collected 30 cycles through the mass sequence 12C, 16O1H, 18O, 19F, and 30Si, measuring negative ions sputtered by a 5 nA Cs+ primary beam; other details concerning vacuum conditions, beam alignment, charge compensation, pre-sputtering, raster size, aperturing, and discrimination for surface contamination are outlined in Mosenfelder et al. (2011). Summaries of the data (16O1H/30Si and/or 19F/30Si ratios) are given for glasses in Table 2 and orthopyroxene and olivine in Table 4; errors are given as two times the standard deviation (2σSD) of 2–7 analyses per sample. We use 30Si as the reference isotope throughout the rest of this paper but note that very similar trends and statistics (internal precision, reproducibility, and relative slopes/goodness of fit to calibration data) apply when 18O (which has a higher ion yield than 30Si) is used instead. Complete data [count rates, ratios, and uncertainties for all reported masses, for individual analyses; uncertainties are given as two times the standard error (2σmean) of the 30 measurement cycles] are provided in the supplementary material. Although we only report analyses in this paper from a single analytical session conducted in April 2012, isotopic ratios were consistent within ~10% with values measured on the same samples during the sessions reported in Mosenfelder et al. (2011).
Fluorine concentrations in NAMs were calibrated with reference to silicate glass standards (Table 2). For blank correction of the glass analyses we used synthetic forsterite GRR1017, a well-established blank standard for 16O1H that also yields low 19F counts (Mosenfelder et al. 2011), around 30 cps (a low count rate considering the high ionization efficiency of 19F compared to the other collected secondary ions). Although we cannot be certain that this sample gives a true blank measurement, we note that count rates are within error of each other (when measured sequentially) for GRR1017, synthetic orthopyroxene (see below), synthetic clinopyroxene (see part II), and synthetic zircon (unpublished data); this suggests that we are measuring the background due to F in the vacuum and/or memory effects (from deposition of F on the immersion lens), rather than F within the samples. We report F data for a total of 11 commonly used glass standards of varying composition (ranging from komatiitic to rhyolitic/high-Si synthetic glass), obtained from the U.S. Geological Survey (USGS), National Institute of Standards and Technology (NIST), and Max-Planck Institut (MPI-DING; Jochum et al. 2006). Table 2 lists F concentrations of the standards determined from selected previous studies (Hoskin 1999; Straub and Layne 2003; Jochum et al. 2006; Guggino and Hervig 2010, 2011). The large variations in these values, combined with concerns about heterogeneity in some glasses, prompt us to explore different models for F calibration, as outlined in the results section.
Blank correction for 16O1H and 19F in orthopyroxene was performed using OH-poor and/or F-poor standards, which were measured periodically during the session to monitor the background level. For this purpose we examined a synthetic enstatite (GRR247, grown by Ito 1975 using flux methods) and a natural orthopyroxene that was dehydrated in the lab (ZM1opx-HT) and compared them to GRR1017 forsterite. ZM1opx-HT was dehydrated by heating under reducing conditions at 1000 °C for 72 h in a one atmosphere Deltech furnace, with oxygen fugacity controlled by a CO-CO2 gas mixture at 10−12.2 atm (corresponding to log fO2 approximately one order of magnitude lower than the quartz-fayalite-magnetite buffer; cf. Aubaud et al. 2007). FTIR spectra revealed no detectable OH, with a detection limit of approximately 1 ppm H2O. On the other hand, this sample retained relatively high (although variable) amounts of F, making it unsuitable for that blank correction. Conversely, GRR247 was found to be unsuitable for H correction but excellent for F correction. It contains a small amount of H2O, probably present as fluid inclusions (as evidenced by a weak broad IR band centered at 3400 cm−1 and reflected in measured 16O1H/30Si ratios), but yielded very low 19F counts directly comparable to GRR1017 (~30 cps).
Estimated limits of detection (LOD) and quantitation (LOQ) for H, calculated from analyses of ZM1opx-HT, regression of the data as detailed in the results section, and the simple formulism of Long and Windefordner (1983) as used in Mosenfelder et al. (2011) were 1 and 4 ppm H2O, respectively. The LOD and LOQ for F (calculated in this case from analyses of GRR1017) are both less than 1 ppm regardless of which model for F calibration is used, ranging from 0.1 to 0.25 ppm (LOD) and 0.4 to 0.8 ppm (LOQ). These low values for F correspond well with estimated detection limits in previous studies, which also used synthetic forsterite to measure F backgrounds (Hauri et al. 2002; Jochum et al. 2006).
Table 1 gives integrated absorbance for the three directions measured in each crystal. We also list absorbance for two wavenumber ranges (“low” vs. “high”), separated by an arbitrary dividing line at 3350 cm−1 without performing any peak fitting to separate contributions of overlapping bands. IR spectra for all the samples (except the blank standards) are shown in Figure 1 in order of overall H concentration (Table 1). Note that the y-axis scales are different for each panel in the figure, where the scale for Figure 1c is compressed the most to show all the spectra together. Overall absorbance in most samples follows the order γ > α > β, as observed in many other studies, but the degree of polarization varies greatly between samples (Table 1). The spectra show a wide range of band structures, with the number of different, strong peaks generally increasing with increasing H content. All samples show two or more bands between 3400 and 3560 cm−1. However, the four mantle samples (KBH-1, DE2-1, PMR-54, and GKopxA), which have the highest H concentrations, show an additional band at ~3600 cm−1 in α and β (Figs. 1a and 1b) that is absent or greatly diminished in the other pyroxenes that are thought to be crustal in origin. All samples also contain multiple bands in the range from ~3000–3400 cm−1; the fine details of band structure in this region are difficult to see for all samples in Figure 1 due to scaling (for instance, compare the spectrum of KBH-1 in Fig. 1c with the spectrum in Fig. 2 of Bell et al. 1995).
We looked extensively for OH zoning in all of the pyroxenes, but found none with respect to the major bands shown in Figure 1. On the other hand, some crystals show variable amounts of absorbance at high wavenumbers (>3650 cm−1) that we attribute to inclusions of hydrous phases. These bands were found in optically clear parts of the samples. Skogby et al. (1990) inferred the presence of amphibole and/or pyribole lamellae in many natural pyroxenes via comparison of sharp bands near 3675 cm−1 with corresponding bands in amphibole and pyribole spectra (Skogby and Rossman 1991). Such “chain width defects,” hereafter referred to for simplicity as amphibole lamellae, are also well known from TEM studies (e.g., Veblen 1985). In orthopyroxene, the amphibole lamellae IR bands are strongly polarized in the β direction (Skogby et al. 1990). In Figure 2, we show a comparison of spectra for some of our samples and spectra for hydrous phases from the literature. Bands clearly attributable to amphibole [note the comparison to pargasitic hornblende from Skogby and Rossman (1991)] were found in GRR2334a, JLM46, and JLM14, with average intensities increasing in that order. GRR1650b shows two sharper peaks at 3675 and 3660 cm−1 (Fig. 2), which could be from either tremolite or talc. The high wavenumber bands in GKopxA (Fig. 1b) are more difficult to interpret but we provisionally assign them to amphibole (these bands are also strongly polarized parallel to β). Finally, PMR-54 shows a complicated band structure with a peak at 3685 cm−1 that indicates the presence of serpentine, a common alteration product of orthopyroxene (e.g., Gose et al. 2011). These bands are seen in both α (Fig. 2) and β (Fig. 1) in this sample.
Hydrogen concentrations calculated using either the Bell et al. (1995) or Libowitzky and Rossman (1997) calibrations are shown in Table 1. Note that the high wavenumber bands associated with hydrous inclusions constitute less than 1% of the total absorbance for all samples except JLM14 and were not included in the total absorbance cited in Table 1. When we use the Libowitzky and Rossman calibration, calculated concentrations are 3–14% lower for all samples except KBH-1 and PMR-54, for which the calculated concentrations are higher (by 12% and 2%, respectively). There is a systematic relationship (Fig. 3a) between the two calibrations as a consequence of the differences in ratio of low to high wavenumber absorbance, which vary by up to a factor of two among samples (using our arbitrary dividing line at 3350 cm−1). Overall, however, these subtle discrepancies balance out, leaving a strong, almost 1:1 correlation between concentrations for all samples calculated using both calibrations (Fig. 3b). Similar trends can be seen (Fig. 3b) for data on natural orthopyroxenes presented by Sundvall and Stalder (2011). This result stands in strong contrast to the same exercise performed for clinopyroxene, which shows a greater variation in mean wavenumber as we discuss in Part II.
SIMS: Fluorine calibration
To convert 19F/30Si ratios in NAMs to F, we calibrated using silicate glasses with nominally well-determined F concentrations. For this purpose we multiplied all ratios (both in standard glasses and in minerals) by SiO2 content (Table 2). The precision of our glass analyses was excellent, ranging from 0.3 to 0.9% (2σmean), but reproducibility from 3–7 analyses of each standard varied between 1 and 11% (2σSD). Figure 4a shows our calibration curve for the MPI-DING and NIST glasses. Here we plot F concentrations for MPI-DING samples as recommended by Jochum et al. (2006) based on SIMS measurements conducted by Erik Hauri that are in turn referenced to EPMA data on other standards (Hauri et al. 2002); for NIST standards we also plot the values cited in Jochum et al. (2006). ATHO-G was excluded from this plot because its measured 19F/30Si ratio (2.32) is wholly incompatible with the recommended value for F (0.7 ppm), suggesting substantial heterogeneity in the glass. The data (including the blank standard, GRR1017) were fit with both ordinary least-squares (OLS) and York (1966) regressions and show a large degree of scatter that cannot be explained by the uncertainties of our measurements alone.
Recently, Guggino and Hervig (2010, 2011) published considerably lower values (compared to Jochum et al. 2006; see Table 2) for NIST SRM 610, NIST SRM 612, ML3B-G, and KL2-G, as well as values for the USGS standard glasses BCR-2G and BHVO-2G. Their results for basalts (Guggino and Hervig 2011) are also tied to EPMA data on other glasses (five basalts doped with between 0.2 and 2.5 wt% F), while their values for NIST standards are based on proton induced γ-ray emission analyses (Guggino and Hervig 2010). We plot our data for these six standards against Guggino and Hervig’s values in Figure 4b, along with a York regression fitting only the basalt data. Unfortunately, these points also show considerable scatter. As we discuss below, the reasons for this are likely related more to heterogeneity in the glasses (e.g., due to F loss during homogenization) than matrix effects; for instance, we could not reconcile the data for the four basalts shown in Figure 4b by plotting against other chemical variables such as Mg/(Mg+Fe) ratio (cf. Fig. 1d in Hauri et al. 2002), and the small variations in SiO2 content also cannot explain the scatter.
Acknowledging the large degree of uncertainty in F measurements by SIMS implied by the above discussion, we use two different models to estimate F in NAMs in Table 4, based on either the Jochum et al. (2006) values or the Guggino and Hervig (2011) values for the standards (not including the NIST standards in the latter case). From simple inversion of the York regressions to the data sets shown in Figure 4 (which differ in slope by a factor of 2.14), we derive model 1 (based on Jochum et al. values):
and model 2 (based on Guggino and Hervig’s data):
SIMS: Hydrogen and fluorine in orthopyroxene
Average measured 16O1H/30Si ratios from 3–7 analyses for each sample are listed in Table 4. Typical internal precision for 16O1H/30Si ratios was 1–3% (2σmean) for samples containing H, and 5–7% for ZM1opx-HT, the blank standard. Out of a total of 47 analyses, only three with much higher uncertainty were rejected based on the Poisson counting statistic criterion we used previously (Mosenfelder et al. 2011); these are not included in Table 4 but are given in the supplementary material. Reproducibility was 10% or better for most samples. However, JLM46 and JLM14 yielded more variable 16O1H/30Si (spread by up to a factor of 2) despite high precision (i.e., flat depth profiles) for individual analyses. This is illustrated in Figure 5, where we plot 16O1H/30Si vs. 19F/30Si for selected samples. For the sake of illustration, we also show an older analysis of JLM14 (from a session in 2010) that was even more discrepant with other analyses. There is a clear linear correlation (r2 = 0.999) between 19F and 16O1H for these two pyroxenes. According to IR spectra, as discussed above, they also contain the highest concentrations of amphibole lamellae. Because of the high probability that we ionized variable amounts of these lamellae, we culled the data by taking the analyses with the lowest measured 16O1H and 19F as an upper constraint on the “true” value of amphibole-free orthopyroxene. This leaves three analyses (out of four) of JLM46 with 16O1H/30Si = 0.0156 ± 0.0001 and four analyses (out of seven) of JLM14 with 16O1H/30Si = 0.0364 ± 0.0034, which we used for the sake of the calibration lines presented next.
Figures 6a and 6b show the average measured 16O1H/30Si ratios (normalized by multiplying by SiO2 as determined by EPMA) plotted against concentrations determined from FTIR using the Bell et al. (1995) and Libowitzky and Rossman (1997) calibrations, respectively. Fits to the data (both OLS and York regressions) have virtually identical slopes and similar intercepts close to zero, despite the fact that H concentrations determined by the two IR calibrations vary significantly for some samples. To illustrate this last point in Figure 6b we use arrows to point to the two samples (KBH-1 and JLM14) whose H concentrations differ the most when applying the Libowitzky and Rossman (1997) calibration. Finally, we note the peculiarity that JLM14, the sample with the most amphibole lamellae, plots farthest from the best-fit line; ironically, if we use the average of all seven analyses listed in Table 4 for JLM14 it would plot exactly on the best-fit line shown in Figure 6a. In other words, this discrepancy is apparently not due to measuring 16O1H in hydrous inclusions, a problem we assessed for olivine in Mosenfelder et al. (2011).
Fluorine concentrations were calculated using the two different models for the calibration on glasses; they range from 1–37 ppm for model 1, and 0.4–17 ppm for model 2. In general, the orthopyroxenes from crustal environments (JLM50, JLM46, GRR2334a, GRR1650b, and JLM14) have low amounts of F compared to mantle samples, which have modest amounts. A single exception is GKopxA, which had the lowest F of any natural orthopyroxene we measured, above the LOD but essentially at the LOQ. In contrast to the recent study of Beyer et al. (2012) on a more limited set of natural samples, we found no correlation between F and Al2O3 (Fig. 7). The lack of correlation is not improved when F is compared to tetrahedral or octahedral Al (as calculated from the EPMA data) or other trivalent cations (e.g., Cr3+) are considered.
SIMS: Hydrogen and fluorine in olivine
Figure 8 shows the olivine calibration conducted in the 2012 session. The level of data scatter and quality of the regression are similar to previous calibrations (Mosenfelder et al. 2011). We also show the same data for orthopyroxene as plotted in Figure 6a (i.e., assuming the Bell et al. 1995 calibration). The difference in slopes between the orthopyroxene and olivine calibration lines is about 20%. This suggests a matrix effect between the two phases that is not simply due to differences in 30Si yield (because we normalized for SiO2). We evaluate this possibility further below, in the context of results from other laboratories and the uncertainties in our regression analysis.
Using model 2 for the F calibration, F concentrations in olivine range from less than 1 ppm to about 50 ppm (or from 1 to 100 ppm for model 1). These values are ~27% lower than the preliminary values for the same samples that we presented in Mosenfelder et al. (2011). They are fairly consistent with recent studies (Guggino et al. 2007; Beyer et al. 2012), which examined a wider range of olivines. Much higher values were reported for Monastery olivines by Hervig and Bell (2005), obviously based on a different calibration. One sample (GRR1695-2) had low F but a very wide spread in 19F/30Si, which we attribute to variable ionization of F-bearing nanometer-scale pores, as documented in Mosenfelder et al. (2011). Although we did not re-analyze some of the other olivines from Mosenfelder et al. (2011) in the session reported here, we can estimate their F concentrations by comparing previously measured 19F/30Si ratios among all the samples. The resulting values for KLV23, ROM250-OL2, and ROM250-OL13 are 6, 10, and 45 ppm, respectively (assuming model 2).
Uncertainties in hydrogen measurements
Our study confirms previous work showing that low-blank, very high precision analyses (as good as 1% 2σ or even better) of trace H and F are attainable using SIMS. For H we depend on cross calibration with FTIR, which is subject to much larger uncertainties deriving from absorbance measurements and the calibration of the absorption coefficient. Unfortunately, these issues have not been resolved since being raised by Bell et al. (1995) and are treated in different ways in different studies.
In the present study, uncertainties due to differences in polarizer efficiency (Libowitzky and Rossman 1996) or imprecision in thickness or density measurements are inconsequential compared to the uncertainty in spectral baselines. Linear baselines, which have the obvious advantage of being inherently reproducible, are commonly used to correct spectra of synthetic enstatite (e.g., Stalder 2004; Prechtel and Stalder 2012). However, this type of correction is not appropriate for Fe-bearing orthopyroxene, because the baseline in this case is a complex, curved function of both Si-O vibrational bands and electronic transitions (Fe2+ in the M2 site, with bands at approximately 11 000, 5400, and 2350 cm−1; Goldman and Rossman 1976), with long tails of absorption. In the Appendix, we detail how we approached this problem.
Aside from uncertainties in absorption measurements, there is clearly further need for additional determinations of the IR absorption coefficient using absolute methods such as hydrogen manometry or nuclear microprobe techniques. An early attempt using NRA determined a value of 508 ppm H2O for PMR-54 (unpublished work of R. Livi reported in Skogby et al. 1990). This is about twice as much as the value we infer for this pyroxene (Table 1) using either of the two IR calibrations. This analysis was obtained before the method was optimized and interferences with other nuclear reactions were recognized (Rossman 2006), and has been ignored in subsequent studies. However, it underscores the potential difficulties associated with achieving low blanks and/or discriminating against “contamination” by hydrous phase inclusions (in this case serpentine, as shown in Fig. 2).
We have relied on the manometry calibration of Bell et al. (1995), who determined a value of 217 ± 22 (2σ) ppm H2O for KBH-1. Support for this value is given by the more recent study of Wegdén et al. (2005), who measured 25 ppm H (equivalent to 223 ppm H2O) in Kilbourne Hole orthopyroxene using proton-proton scattering. On the other hand, O’Leary et al. (2007), employing continuous flow mass spectrometry, measured a significantly lower H concentration in KBH-1. Their value of 165 ± 40 ppm H2O (2σ) is only in weak agreement with Bell et al.’s value. Here we note that the nominal value of 186 ppm H2O cited in O’Leary et al. (2007), Bell and Ihinger (2000), and Koga et al. (2003) represents only the amount of H2O extracted for manometry, not the original concentration, which Bell et al. (1995) calculated by estimating the amount of O-H absorbance remaining in sample chips after extraction.
Variations like those shown in Figure 1 prompted Bell et al. (1995) to suggest that the molar absorption coefficient probably varies among samples with different types of spectra, making application of their calibration on KBH-1 to other pyroxenes non-ideal. So far there has been only one attempt to conduct absolute calibrations on multiple orthopyroxene samples with different IR spectra. Aubaud et al. (2009), using elastic recoil detection analysis (ERDA), unfortunately concluded that the background H level of the technique was too high to quantitatively determine absorption coefficients for the samples they used (although it could be used in the same study for rhyolitic glasses and synthetic olivines, with much higher H concentrations; in part II, Mosenfelder and Rossman 2013, we also discuss their results for clinopyroxene).
Wavenumber-dependent, site-specific, or uniform IR absorption coefficients?
The suggestion of Bell et al. (1995) that the molar absorption coefficient could vary among samples with different spectra may have partially motivated subsequent workers to prefer the Paterson (1982) or Libowitzky and Rossman (1997) calibrations, which are both wavenumber-dependent functions. The latter calibration has been used especially frequently for pyroxenes (e.g., Stalder and Skogby 2002; Stalder 2004; Nazzareni et al. 2011; Stalder et al. 2012), despite the fact that it was derived from fitting data on stoichiometrically hydrous phases (whereas the Paterson calibration is primarily based on various phases of water and water dissolved in organic solvents, glasses and quartz). Support for a linear correlation between absorption coefficients and wavenumber also comes from theoretical modeling of OH-containing molecules (Kubicki et al. 1993) and hydrous phases (Balan et al. 2008). On the other hand, recent studies have argued that these linear trends as a rule are inapplicable to NAMs (Rossman 2006; Thomas et al. 2009; Koch-Müller and Rhede 2010; Balan et al. 2011). It has also been proposed (Kovács et al. 2010; Balan et al. 2011) that different O-H defect sites within a single mineral can have absorption coefficients varying by much larger factors than those predicted by any wavenumber-dependent calibration; this hypothesis was contested for the case of olivine in our study using SIMS and FTIR (Mosenfelder et al. 2011).
Our calibration for orthopyroxene (Fig. 6) shows that the data can be fit with the same precision and indeed the same regression slope, within error, regardless of whether a wavenumber-dependent or single absorption coefficient is used to calculate H contents by FTIR; a similar conclusion is apparent from the data (see their Fig. 9) of Aubaud et al. (2007). Our samples show a wide range of band structures with variable percentages of low- and high-wavenumber absorbance (Table 1, Fig. 3a). However, these differences balance out as a result of the overall correlation between the two IR calibrations (Fig. 3b). Therefore, we conclude that within the resolution of our methods it is not possible to ascertain which calibration is more accurate. This result stands in contrast to our conclusion for SIMS data on clinopyroxenes, which can be fit better using a frequency-dependent calibration as shown in Part II (Mosenfelder and Rossman 2013).
This assessment also contrasts with recent SIMS/FTIR work by Stalder et al. (2012), who studied pure (Fe- and Al-free) enstatites synthesized at 2.5–8 GPa that show variable absorbance for four major bands over a large wavenumber range (with band centers at 3090, 3360, 3590, and 3690 cm−1). They concluded that applying different absorption coefficients to the four different bands was not meaningful, but that wavenumber-dependent functions fit their data significantly better than using the single absorption coefficient of Bell et al. (1995). A similar conclusion was reached in an earlier study on synthetic clinopyroxenes (Stalder and Ludwig 2007). It should be emphasized that there are many differences between our studies, making straightforward comparison impossible. The SIMS technique used by Stalder et al. (2012) is novel for geological applications, relying on multiple analyses on the same spot at highly varying beam current to correct for the effects of high H background from contamination of the vacuum (Ludwig and Stalder 2007). These studies also rely on tourmaline standards for quantification, neglecting any possible matrix effects. Moreover, while three of the bands in the synthetic samples (3090, 3360, and 3590 cm−1) are similar to bands seen in natural orthopyroxene, the fourth, at 3690 cm−1, has not been encountered as a single sharp band in natural samples; bands at similar frequencies are interpreted to come from hydrous inclusions in natural samples (and are generally less intense) but are thought to represent structurally bound H in the experiments (Prechtel and Stalder 2012). This band contributes an outsized percentage to the total H calculated using the wavenumber-dependent calibration, by the very nature of the function. Thus it is not surprising that the Bell et al. (1995) and Libowitzky and Rossman (1997) calibrations diverge strongly in Figure 3 of Stalder et al. (2012). Therefore it is plausible that a wavenumber-dependent function is superior for synthetic high-pressure enstatites, but not justified for natural orthopyroxenes.
SIMS: Matrix effects
One of the most important concerns with using SIMS to measure H in NAMs is the magnitude of matrix effects, which have been addressed in some studies and ignored in others. Koga et al. (2003) and Aubaud et al. (2007) documented differences in calibration slopes among different NAM, hydrous minerals, and silicate glasses up to a factor of 4.25. Other studies have neglected matrix effects entirely, yet recommended new absorption coefficients for NAMs based on calibration tied to different minerals. In the case of orthopyroxene, Stalder et al. (2005, 2012) suggested that existing IR calibrations systematically underestimate H (by factors of 1.4–1.7). However, the use of tourmaline as the standard for their SIMS analyses makes this conclusion tenuous until the magnitude of matrix effects can be established.
Koga et al. (2003) postulated that the matrix effect between olivine and orthopyroxene should be minimal, based on the similarity in mean atomic weight and chemistry between these minerals. Although they successfully regressed their olivine and orthopyroxene SIMS data together, the H concentrations in most of their orthopyroxenes were imprecisely determined based on adjusting (by a factor of 2) concentrations determined using the Paterson (1982) calibration on unpolarized spectra. Moreover, they did not correct their data for the differences in SiO2 content, as we have done here following typical protocol for evaluating matrix effects (e.g., Hinthorne and Andersen 1975; Ihinger et al. 1994; Aubaud et al. 2007). If their data are normalized in this way, a difference between the calibrations for olivine and orthopyroxene emerges that is similar to the later results of Aubaud et al. (2007), who used the same IR calibrations we have used here to establish a factor of 1.5 difference in calibration slopes for the two phases. A similar difference (factor of 1.4) can be seen in calibration slopes later measured for the same standards by Tenner et al. (2009) using analytical conditions similar to those we used, in particular measuring 16O1H rather than 1H. However, the H concentrations of many of the synthetic olivines used in earlier studies were subsequently revised by Withers et al. (2011), who then found indistinguishable calibration slopes for olivine and orthopyroxene. Another possibility to consider is that there may be a fundamental difference in the effect of the matrix on yield of lighter 1H ions (as measured in the earlier studies) vs. 16O1H.
Our calibration slopes (Fig. 8) for orthopyroxene and olivine, acquired in the same session, are within about 20% of each other. This reinforces the idea that the matrix effect between these two minerals is small. Furthermore, the difference may be even less if more recent values for the IR absorption coefficient for olivine are considered. Thomas et al. (2009) derived a range of ɛi values between ~28 000–47 000 L/molH2O/cm2 for olivines with different spectral characteristics, based on two different techniques (proton-proton scattering and Raman spectroscopy). More recently, using ERDA, Withers et al. (2012) derived an ɛi of 45 200 L/molH2O/cm2 for olivine. Use of the higher values would result in lower amounts of H in olivine, thereby bringing the two calibration lines closer together (in fact, crossing over each other in our particular case).
Matrix effects have also been well documented for solid solution within single mineral groups, but the mechanisms responsible for the effects are still unclear. For instance, Ottolini et al. (2002) showed that H yields decrease with increasing Fe+Mn in a variety of hydrous minerals such as kornerupine, tourmaline, and micas, when using a 16O− primary beam and an energy-filtered secondary ion beam. In an earlier paper (Ottolini and Hawthorne 2001), it was speculated that this effect might be due to oxidation-dehydrogenation reactions (buffered by implantation of oxygen from the primary beam), resulting in liberation of a neutral species (H2O or H2) that goes undetected in the mass spectrometer. This effect cannot be relevant in our study, which used a Cs+ beam (and no energy filtering); indeed, data from Koga et al. (2003) suggest the reverse phenomenon, with H/Si generally increasing with increasing Fe. Moreover, oxidation cannot easily explain the fact that yields of other light elements (Li, B, and F) also decrease with increasing Fe+Mn. More likely, the matrix effect is related to an overall increase in molecular weight of the matrix, which has generally been shown to decrease H/Si ratios (King et al. 2002) as well as fractionate D/H to higher values (Hauri et al. 2006b); in the former case it is unclear whether H yields go down or Si yields go up. Our data are very limited in this respect but suggest that orthopyroxenes with Mg numbers [100 × molar Mg/(Mg+Fe); Table 3] between 79 (GRR1650b) and 99 (JLM50) show no statistically meaningful Fe-related matrix effect for H; a similar conclusion was reached for garnets of varying Mg number by Aubaud et al. (2007). Nevertheless, future work to explore the effect of Fe on H yield is warranted to better quantify H in Fe-rich NAMs from martian and lunar rocks (e.g., Boctor et al. 2003; Liu et al. 2012) as well as experimental samples with high Fe (and considerably more H compared to natural samples) designed to model the martian mantle (Withers et al. 2011).
SIMS: Analysis of hydrous inclusions
A general problem for SIMS measurements of H in NAMs is the possibility of ionizing fluid inclusions or inclusions of hydrous phases. Aubaud et al. (2007) inferred that SIMS could be more effective than FTIR at screening out contributions from H that is not bound in lattice defects, because of its very limited sampling volume compared to FTIR. On the contrary, in Mosenfelder et al. (2011) we emphasized that ionization of hydrous inclusions can result in severely skewed OH/Si ratios, depending on the volume of nanometer- to micrometer-scale inclusions that are sampled. This stochastic phenomenon is also seen here for those samples containing relatively high abundances of amphibole lamellae (JLM46 and JLM14), a common features in many pyroxenes (Skogby et al. 1990). However, there is a significant difference between the results on orthopyroxene and olivine. In the latter case we found that “contaminated” analyses can usually be easily identified by considering the ratio of the standard error of the mean to the error predicted by Poisson counting statistics (σmean/σPoisson). For olivine, we used a cut-off criterion, filtering analyses with σmean/σPoisson > 5. For orthopyroxene, most of our analyses are highly precise with σmean/σPoisson close to 1, even when we infer that amphibole lamellae have been sampled. Our filtering criterion in this case is based on comparison of 16O1H/30Si and 19F/30Si ratios (Fig. 5), making the simplistic assumption that both H and F partition strongly into amphibole. The different behavior for pyroxene analyses presumably reflects differences in the geometrical relationship of the host phase to inclusions; the lamellae may form in wide (but thin) layers that are evenly sampled during sputtering, while the sub-micrometer-scale, ovoid fluid inclusions we imaged in olivines may have a very heterogeneous distribution even on the small scale of a SIMS analysis pit. A similar phenomenon could be envisaged for Ti-clinohumite lamellae or platelets in olivine (Mosenfelder et al. 2006b). An important corollary of this result is that SIMS studies of natural pyroxenes should probably be accompanied by at least qualitative FTIR spectroscopy (e.g., unpolarized spectra taken through polycrystalline samples), because the presence of hydrous inclusions may not be easy to determine by SIMS alone and is also sometimes difficult to detect using optical microscopy.
Uncertainties in fluorine measurements
As there are no well-established matrix-matched standards for F in NAMs, we took the same approach as other workers (Hervig and Bell 2005; Hauri et al. 2006a; Guggino et al. 2007; O’Leary et al. 2010; Dalou et al. 2012) of using F-bearing silicate glasses for calibration. Two other recent studies (Bernini et al. 2012; Beyer et al. 2012) have taken a fundamentally different approach of using ion implanted natural forsterite and enstatite crystals for standardization. Unfortunately, the large variation in regression slopes and high uncertainty envelopes for the different models shown in Figure 4 underscore the large uncertainty in using silicate glasses for F calibration. Possible factors contributing to this uncertainty include: fundamental errors in determinations of the standards; heterogeneity in the standards; matrix effects due to chemistry within the glasses themselves; and matrix effects resulting from the difference in structure between amorphous and crystalline materials.
The large discrepancies in values determined for F in the MPI-DING glasses by Jochum et al. (2006) and Guggino and Hervig (2011) are as yet unexplained but could be related to either or both of the first two issues mentioned above. In both cases, F was determined by calibration on other glasses with concentrations high enough to measure by EPMA, a notoriously difficult analytical method to use for F due to problems including low X-ray peak intensity, volatile migration, and interferences by higher order peaks from transition elements, particularly FeLα1 (for more detailed discussion see Witter and Kuehner 2004). In this respect we presume that the more recent results of Guggino and Hervig (2011) are more reliable because they used multiple synthetic F-bearing glasses as standards and employed peak-integration analysis rather than using peak heights, a more typical method for reducing EPMA data. We note that using this calibration gives a lower value for F in Monastery olivines (e.g., 37 ppm in ROM177, compared to 80 ppm for model 1) that is more consistent with the measurement of 30 ppm F in a Monastery olivine by Beyer et al. (2012).
Even assuming that Guggino and Hervig’s values for basalts are correct, we still have considerable scatter in our calibration (Fig. 4b). One concern with glasses, which has been particularly well documented for the NIST samples (Hoskin 1999; Eggins and Shelley 2002), is that volatile elements (purposely doped in the case of NIST) can be lost during fabrication, for instance escaping along “cord structures” that form during melting. We were unable to locate cord structures in our NIST standards using optical methods (Eggins and Shelley 2002) and elected to just cut small slabs from the middle of the as-provided wafers, the region least likely to have suffered from volatile loss. As for the MPI-DING and USGS glasses, more work is clearly needed to establish levels of homogeneity. The loss of other elements besides F has been documented in some of the MPI-DING glasses. For instance, Borisova et al. (2010) found significant depletion of many elements in ATHO-G in an anomalous split, and attributed the discrepancy either to loss to the crucible wall or development of compositional cords. This could easily explain the severe discrepancy between our measured 19F yield for ATHO-G and the much lower reference value from Jochum et al. (2006), as well as scatter among the other glasses we used.
The extent to which matrix effects are important for F in glasses has yet to be elucidated and is the subject of ongoing studies (Rose-Koga et al. 2008; Guggino and Hervig 2010, 2011). Although these effects might be expected to be very small as a result of the high ionization efficiency of F (Williams 1992), Guggino and Hervig (2011) estimate matrix effects of ~50% between low- and high-silica glasses. This difference is much larger than a previous estimate of 15% from the same lab for a range of phases (Jamtveit and Hervig 1994). Sensible trends in this regard are difficult to decipher from the data shown in Figure 4a not only because of data scatter, but because the F in the standards was established in the first place under the assumption that matrix effects are small—within the assigned, overall uncertainties (precision and accuracy) of 10% (Jochum et al. 2006). As for possible differences in ion yield due to the fundamental structural differences between amorphous glasses and the crystalline silicates of primary interest to use here, there are very few constraints currently, but we would point out that differences in H yield can be quite large between silicate glasses and NAMs (Koga et al. 2003; Aubaud et al. 2007). All of these issues point to the need in future studies to establish matrix-matched standards for NAMs, perhaps by synthesis at high pressures and temperatures or by ion implantation (Bernini et al. 2012; Beyer et al. 2012).
Fluorine in nominally fluorine-free minerals
Fluorine is readily incorporated in the same sites as OH in hydrous minerals such as amphiboles, micas, and the humite-series, as well as in apatite. Analyses of these phases form the basis for classic studies of F distribution within the crust and mantle (Aoki and Kanisawa 1979; Aoki et al. 1981; Smith et al. 1981; Edgar et al. 1996). One of the implicit and explicit (Smith et al. 1981) assumptions of this work is that the common upper mantle NAMs (olivine, garnet, ortho- and clinopyroxene) do not contribute significantly to the bulk F content of the mantle. A very similar view was held about H until increasing evidence was provided from IR studies that contradicted this idea (e.g., Aines and Rossman 1984; Bell et al. 1992).
The F concentrations we measured for olivine and orthopyroxene are indeed quite low (less than ~50 ppm) compared to F in phases such as phlogopite or amphibole from the studies mentioned above. However, when considered in the context of modal abundance, these amounts may be more than sufficient to account for the inferred F budget of the mantle (Beyer et al. 2012). In general, we found that F is much lower in crustal-derived orthopyroxene and olivine, compared to mantle samples; importantly, we can rule out the presence of amphibole (which might preferentially partition F) in the latter pyroxenes because IR is very sensitive to small concentrations of lamellae (Skogby et al. 1990). However, we see a distinct trend of higher F in megacrysts from South African kimberlites (PMR-54 and DE2-1 orthopyroxene; Monastery olivines) compared to xenocrysts from minette diatremes in the Colorado Plateau (GKopxA orthopyroxene; GRR1629-2 and GRR1784e olivine). For the Monastery kimberlite, we also confirm the distinct difference depending on petrologic context that was reported by Hervig and Bell (2005): F is lower in group 1 or “main silicate trend” olivine (ROM250-OL2) compared to Fe-rich “group 2” olivines (ROM177 and ROM250-OL13). These results imply that different mantle reservoirs have different amounts of F (possibly related to the degree of differentiation in the Monastery samples), and that F is not necessarily correlated with other volatile concentrations (H concentrations are inferred to be high in the mantle below both the Colorado Plateau and South Africa). Beyer et al. (2012) reached a different conclusion, inferring similar amounts of F in different mantle reservoirs based on measurements in natural olivines (nine samples) and orthopyroxenes (three samples) from different localities. Of course, both of our limited surveys have unavoidable sampling bias. For instance, we would not be surprised to see much higher F in orthopyroxenes associated with F-rich phlogopite from ultrahigh-temperature granulite terranes (Motoyoshi and Hensen 2001), or in highly metasomatized mantle xenoliths such as those from West Eifel, Germany (Edgar et al. 1996). Moreover, Beyer et al. (2012) only studied one olivine derived from the garnet-lherzolite facies, so their inference of homogeneity among mantle reservoirs only reasonably applies to the upper-most mantle (plagioclase- and spinel-lherzolite facies).
Experimental studies have shown that NAMs can incorporate approximately an order of magnitude more H—at very high pressures—than their most hydrous counterparts found so far in nature. By the same token, it is not surprising to see that experimental studies are finding the same phenomenon for F. Several recent experimental studies bear on this topic, although the wide pressure range that has been examined for NAMs has yet to be fully explored for studies of F. In Fe-free systems at moderate pressures (1–2.6 GPa), Bromiley and Kohn (2007) and Bernini et al. (2012) measured up to 4500 and 1900 ppm F respectively, in forsterite; the latter study also documents up to 336 ppm F in enstatite and 1110 ppm F in pyrope. In more complex systems, Hauri et al. (2006a) and O’Leary et al. (2010) measured F, Cl, S, and H in NAMs equilibrated with basaltic melts from 0.5–4 GPa and 1000–1370 °C under water-saturated conditions. While the focus of these two studies was on H partitioning, they measured up to 13, 16, 88, and 138 ppm F in olivine, garnet, orthopyroxene, and clinopyroxene, respectively. Hauri et al. (2006a) also found that orthopyroxene/melt partition coefficients for F (DFopx–melt) were about twice as high (ranging from 0.015 to 0.0448) as those for H. More recent experiments by Dalou et al. (2012) and Beyer et al. (2012) were conducted at similar pressures and temperatures, but under nominally anhydrous conditions and with deliberately doped, higher amounts of halogens as in the studies in Fe-free systems. Consequently, F contents of orthopyroxene in some of these experiments are much higher (up to 432 and 571 ppm measured by Beyer et al. and Dalou et al., respectively). Dalou et al. (2012) ascribed a very large variation of DFopx-melt (0.0158–0.1841) in their experiments to a dependence on the calculated viscosities of the investigated melts, while Beyer et al. (2012) measured a much more restricted range of partition coefficients (DFopx-melt = 0.031–0.037). Beyer et al. (2012) also documented a correlation between Al and F in both natural and synthetic orthopyroxenes that was not seen by Dalou et al. (2012) and is not seen in the natural samples from the present study (Fig. 7), underscoring the need for further work to assess the incorporation mechanisms for F in orthopyroxenes as well as other NAMs.
Although there are discrepancies between these experimental results that remain to be rectified, when considered together they indicate that F is a much more compatible element in NAMs than H. One consistency between the studies of Hauri et al. (2006a) and Dalou et al. (2012) is that they both found that the compatibility of F on average followed the order of clinopyroxene > orthopyroxene > garnet > olivine. This result contrasts with our results on natural mantle-derived NAMs, where overall we find the highest F contents in clinopyroxene (part II of this study, Mosenfelder and Rossman 2013), followed by olivine, orthopyroxene, garnet, and zircon (unpublished data for the latter two phases). Clearly more work is needed on both ends, to systematically explore the thermodynamics of F incorporation in experiments while documenting coexisting phases in xenoliths, which we have not done here, to gain better constraints on F partitioning behavior.
We acknowledge funding to G.R.R. from NSF grant EAR-0947956 and the White Rose Foundation. The Caltech center for Microanalysis, partially supported by the Gordon and Betty Moore foundation, also provided some support for the SIMS analyses.
We thank Yunbin Guan for assistance with the ion microprobe, Chi Ma for assistance with the electron microprobe, and John Beckett for assistance with the gas-mixing 1-atm furnace. Zachary Morgan donated the mineral separate used to make the ZM1opx-HT sample, and David Bell has kindly allowed continued use of samples from his dissertation work at Caltech. Discussions with Richard Hervig and Marion Le Voyer on the topic of fluorine measurements were enlightening and encouraging. Finally, we thank the associate editor (Roland Stalder) and an anonymous reviewer for their comments that helped greatly to clarify the manuscript.
Appendix 1. FTIR methods
Silicate overtone spectra
The dependence of polarized IR spectra in the silicate overtone region on orientation has been documented for olivine (Lemaire et al. 2004; Asimow et al. 2006) and used by several workers to verify orientations of grains for FTIR analysis. Asimow et al. (2006) derived a mathematical model for using these Si-O overtone spectra to determine H concentrations in populations of randomly oriented grains, and this method has also been used in some subsequent studies (e.g., Withers et al. 2011). In Appendix Figure 1, we show two examples of spectra in the same region for well-oriented orthopyroxene crystals, which are sometimes used in our lab as a crosscheck to confirm the orientations of crystals measured by FTIR.
Although similar spectra were already published by Prechtel and Stalder (2012), the spectra in Appendix Figure 1 demonstrate an important point: although the overall shapes of the band structures are similar for all samples studied here (and easily distinguished for the three principal orientations), differences in chemistry induce small shifts in peak locations and moderate changes to peak heights. For instance, as highlighted by the dashed line in Appendix Figure 1, the main peak in β is 10 cm−1 lower in GRR1650b (Mg no. 78.8, 0.73 wt% Al2O3) compared to KBH-1 (Mg no. 90.5, 4.67 wt% Al2O3); it is also significantly weaker in KBH-1, which we suggest is largely related to tetrahedral Al replacing some of the Si giving rise to these vibrations. Therefore, application of the Asimow et al. (2006) method to orthopyroxene will require either standard spectra for specific compositions close to the unknowns, or modification of the model to take composition into account. Baseline correction for these spectra is also not as straightforward as in olivine, for which straight-line fits are easily applied with consistency between different crystals; for orthopyroxene the tail of absorbance from the fundamental Si-O vibrational region is steeper, complicating that simplistic approach. Another contribution to the uncertainty in baseline comes from the electronic transition band for Fe2+ in the M2 site at ~2350 cm−1 (Goldman and Rossman 1976), which could affect the peak heights in the silicate overtone region due to its large half-width (~1100 cm−1).
Aside from uncertainties in calibration, uncertainty in the baseline is by far the largest contributor to uncertainty in derived H concentrations (Bell et al. 1995, 2003). Under ideal conditions, precision of 3% or better can be achieved (Bell et al. 1995). In this study, however, we have assumed more conservative overall uncertainties, arrived at from the following considerations.
As mentioned previously, the baseline for Fe-bearing orthopyroxenes is a complex function of both Si-O vibrational bands and electronic transitions with long tails of absorption. These effects are difficult to separate for orthopyroxene, primarily because of the Fe2+ band at ~2350 cm−2—which is absent in garnets, olivines, and clinopyroxene. This electronic transition band—which is strongest in γ, the polarization that also shows the strongest O-H absorption—has a very large full-width at half height (about 1250 cm−1) compared to vibrational bands, with the tail of absorption not only overlapping with the Si-O overtone region but extending as far as the frequency range of fundamental O-H vibrations. Furthermore, in β the Si-O overtones feature a strong peak at relatively high wavenumber (linear absorbance of ~50–65/cm at ~1940 cm−1; Appendix Figure 1). The second overtones of these Si-O vibrations manifest as weak yet not trivial absorbance in the region from ~2600–3000 cm−1, which interferes with the O-H bands in some samples. This phenomenon would not be significant in samples with high H concentrations, such as synthetic enstatites from high-pressure experiments, but is noticeable for samples with low H such as JLM50, which has a broad band at 3100 cm−1 overlapping these second overtones.
With these issues in mind, baseline correction was performed using the built-in routine in Nicolet’s OMNIC software. An ideal method of baseline correction involves computer subtraction of the spectrum of a dehydrated sample from that of the sample of interest (Paterson 1982; Bell et al. 1995). In practice this is difficult for two reasons. First, strict application of this method requires dehydration of multiple samples with compositions corresponding to each wet sample, which we have not undertaken. Second, the dehydration process itself can induce changes in the baseline as a result of changes in oxidation state and/or site occupancy for Fe2+ that shift the Fe2+ peaks. Therefore, we performed manual correction by applying spline fits with a concave curvature, to mimic the approximate shape of the dehydrated sample spectra shown by Bell et al. (1995); these spline fits require placement of tie points below the O-H vibrational bands. We also corrected by eye for the absorption caused by the second Si-O overtones in β explained above. Finally, we subtracted the estimated component in β corresponding to hydrous phase inclusions (amphibole and/or talc and/or serpentine). This last correction is only unambiguous for the one sample (GRR1650b) that showed sharp bands at 3675 and 3660 cm−1, well separated from the intrinsic O-H bands in this orientation. Fortunately the correction for hydrous phases is small overall for all samples, as a result of the low absorbance in this direction compared to γ and α.
The corrections outlined above entail unavoidable subjectivity. Rather than assign a blanket uncertainty to the total absorbance, as in previous studies (e.g., Mosenfelder et al. 2011), we performed a subjective evaluation of uncertainties for each individual spectrum, taking into account the various factors described above. An upper bound on this uncertainty for a given spectrum can be calculated from the difference between the curved fit we chose and a simple linear fit between the bounding points of the integration. This discrepancy can be as high as 30% and is clearly unrealistically high in most cases. Our final estimates of uncertainties in absorbance range from 7 to 25% for individual spectra, as reflected in Table 1; when propagated in order to calculate total absorbance (in three directions) by summing in quadrature, the final uncertainties range from 6 to 12%. Further propagation with the uncertainty in the molar absorption coefficient, as described in the analytical methods section, results in final uncertainties in H concentrations of 10 to 14%.
↵1 Deposit item AM-13-043, data sets: SIMS analyses FTIR spectra. 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 July 10, 2012.
- Manuscript Accepted January 2, 2013.