Forsterite, Hydrous and Anhydrous Wadsleyite and Ringwoodite

American Mineralogist, Volume 94, pages 905–915, 2009

29 Forsterite, hydrous and anhydrous wadsleyite and ringwoodite (Mg2SiO4): Si NMR results for chemical shift anisotropy, spin-lattice relaxation, and mechanism of hydration

JONATHAN F. STEBBINS,1,* JOSEPH R. SMYTH,2 WENDY R. PANERO,3 AND DANIEL J. FROST4

1Department of Geological and Environmental Sciences, Stanford University, Stanford, California 94305, U.S.A. 2Department of Geological Science, University of Colorado, Boulder, Colorado 80309, U.S.A. 3School of Earth Sciences, Ohio State University, Columbus, Ohio 43210, U.S.A. 4Bayerisches Geoinstitut, Universität Bayreuth, Bayreuth, Germany

ABSTRACT We present a detailed 29Si NMR spectroscopic study of isotopically enriched samples of forsterite

and of anhydrous and hydrous wadsleyite and ringwoodite (α, β, and γ phases of Mg2SiO4), which complement previous extensive studies of these minerals by XRD and vibrational spectroscopy. VISi is not detected in any of the phases at levels of about 0.1 to 0.5%. When coupled with recent theoretical calculations on ringwoodite, this suggests the possibility of re-ordering of high-temperature octahedral-tetrahedral disorder during cooling. Cross-polarization (29Si{1H} CPMAS) NMR supports the protonation of O1 oxygen atoms in hydrous wadsleyite without formation of significant amounts of Si-OH groups. In contrast, new NMR peaks appear in hydrous ringwoodite that cross-polarize very rapidly, indicating very short Si-H distances and the presence of Si-OH, as expected from models in which much of the H+ substitutes into Mg2+ vacancies. Static NMR spectra provide new constraints on chemical shift anisotropies in wadsleyite and are fully consistent with the cubic structure of ringwoodite. Spin-lattice relaxation in all phases is much better fitted by a stretched exponential function than with

a more conventional “T1” exponential, as expected when relaxation is dominated by paramagnetic impurities. However, the effects of paramagnetic impurity on ion contents on relaxation, and on the formation of newly observed minor peaks that may result from “pseudo-contact shifts,” appear to depend on mineral structure, and will require considerable future study to understand in detail. Keywords: NMR spectroscopy, forsterite, wadsleyite, ringwoodite, water in mantle

INTRODUCTION marily sensitive to short- to intermediate-range structure around Forsterite, wadsleyite, and ringwoodite [α, β, and γ phases particular nuclides and, in principle, can be highly quantitative without matrix-dependent intensity corrections, NMR spec- of (Mg,Fe)2SiO4] are the most abundant minerals in the Earth’s upper mantle and transition zone (Ita and Stixrude 1992) and troscopy can serve as an important complement to diffraction thus have been widely studied by many methods to elucidate methods and other types of spectroscopy. As reviewed recently their structures and geochemical and geophysical properties. (Stebbins and Kanzaki 1990; Stebbins 1995; Phillips et al. 1997; Ashbrook et al. 2005; Xue et al. 2008), NMR of the Mg2SiO4 The incorporation of H2O into these materials is of particular interest in understanding the water content of the mantle and its phases, as well of other high-pressure Mg silicates, has been ensuing major implications for petrology, global tectonics, and investigated in considerable detail. These have included magic 29 Earth history. The long-range crystal structures of the anhydrous angle spinning (MAS), static, and even single-crystal Si and 25 (Fujino et al. 1981; Sasaki et al. 1982) and hydrous (Smyth Mg NMR on forsterite (Derighetti et al. 1978; Weiden and 29 1994; Kudoh et al. 1996, 2000; Smyth et al. 1997, 2003) phases Rager 1985; Stebbins 1997; Ashbrook et al. 2005), Si MAS are well known from diffraction studies. Infrared and Raman NMR on wadsleyite and ringwoodite (Stebbins and Kanzaki 17 spectroscopy have been especially useful in characterizing O-H 1990; Ashbrook et al. 2005), O MAS and multiple-quantum vibrations in the hydrous phases, measuring hydroxyl contents NMR on forsterite, wadsleyite, clinohumite, and chondrodite (Mueller et al. 1990; Ashbrook et al. 2001, 2005) and 1H NMR and H2O solubilities, evaluating the role of hydrogen bonding, and addressing issues of disorder among proton sites (Kohlstedt on hydrous wadsleyite (Kohn et al. 2002). NMR studies of related et al. 1996; Kudoh et al. 1996, 2000; Kohn et al. 2002; Smyth et high-pressure Mg silicates have ranged from MgSiO3 perovskite al. 2003). However, important details of cation order-disorder, and akimotoite (Stebbins and Kanzaki 1990; Kirkpatrick et al. defects, and vacancy structure (all of which may have significant 1991; Stebbins et al. 2006; Ashbrook et al. 2007a), to majorite effects on bulk properties including elastic constants) remain garnet and its solid solutions (McMillan et al. 1989; Phillips et uncertain, but are beginning to be addressed by first-principles al. 1992), as well as the hydrous 10 Å phase (Welch et al. 2006) theoretical calculations (Panero 2008a, 2008b). and phases A, B, D, E, and superhydrous B (Kanzaki et al. 1992; Because nuclear magnetic resonance (NMR) spectra are pri- Xue et al. 2008). Forsterite, doped with varying concentrations of paramagnetic transition metal cations (either naturally or * E-mail: [email protected] experimentally), has also served as a convenient test system

0003-004X/09/0007–905$05.00/DOI: 10.2138/am.2009.3184 905 906 STEBBINS ET AL.: NMR ON HIGH-PRESSURE Mg2SiO4 PHASES for evaluating the effects of such impurities on linewidths and The wadsleyite was heated for 3 h at 16 GPa, 1200 °C, the ringwoodite for 3 h at 20 in constraining models of spin lattice relaxation (Grimmer et GPa, 1300 °C. The hydrous wadsleyite sample contained a small amount of phase B, as detected by NMR (see below) and EPMA; in the hydrous ringwoodite sample al. 1983; Hartman et al. 2007), both of which can be of critical the only obvious other phase detected by NMR was a trace of phase B. importance in quantifying NMR results. Most samples were analyzed for major and several key minor elements by Relatively recent developments in NMR technology, includ- electron microprobe. As summarized in Table 1, all of the average overall composi- ing fast-spinning MAS probes with small sample volumes, and tions were close to the nominal stoichiometry of Mg2SiO4 with a small excess of 29 progressively higher-field NMR magnets, have facilitated data MgO as planned for the Si-enriched materials. Relatively large standard deviations may reflect the fine particle size of several of the samples and resulting imperfect collection for the small (2–20 mg), high-pressure samples typi- polish, and/or the presence of small inclusions of MgO. In the hydrous samples, 29 cally produced in multi-anvil apparati. For Si spectra, enrich- slightly more excess MgO was expected because of the added Mg(OH)2. We did ment from natural abundances of 4.7% to as high as 95 or 99% not obtain EPMA data on the hydrous ringwoodite sample because of its small can greatly improve spectra. Very recently, we presented such size. Several inclusions with the composition of phase B [Mg12Si4O19(OH)2, 65 wt% MgO, 32 wt% SiO2] were detected by EPMA in the hydrous wadsleyite, as data on several samples of anhydrous forsterite, wadsleyite, and well as in the NMR spectra. These were excluded from the average composition VI ringwoodite (Stebbins et al. 2009). There, we concluded that Si, given in Table 1. Any variations in Mg/Si, caused by incorporation of H2O, were anticipated to be present due to Mg-Si disorder in the spinel- not obvious given the uncertainties of the analyses. Traces of CaO, a common structured ringwoodite at high temperature, could not be detected impurity in MgO reagents, may be present just above the detection limit of about at the 0.1 to 0.5% level, suggesting the possibility of re-ordering 0.02 ± 0.02 wt%; FeO, another common impurity of special interest in some aspects of the NMR spectra, was not significantly detected at a similar limit. CoO was during quench. We also showed the presence of several small detected, and was significantly higher in the doped than in the undoped samples, NMR peaks in forsterite and wadsleyite whose chemical shifts are at roughly 0.06 ± 0.02 wt%. These values were considerably less than the added far outside the known range for Si in silicates, and hypothesized dopant level of about 0.2 wt%, indicating some loss during synthesis. Estimated that these were the consequence of “pseudo-contact shifts” result- contents of other paramagnetic trace element cations, summing contributions from analyzed starting materials, include about 20 ppm Fe (by weight), 10 ppm each ing from trace paramagnetic impurities. In this paper, we extend of Ni and Cr, and 60 ppm Cu. the study to hydrous wadsleyite and ringwoodite, and present Powder XRD on the Co-doped ringwoodite detected only this phase plus detailed results on static (non-spinning) spectra, the effects of <5% wadsleyite. Unit-cell parameters of the hydrous ringwoodite and hydrous paramagnetic ions on spin-lattice relaxation, and results from wadsleyite were determined by single-crystal X-ray diffraction. A single crystal Si{1H} cross-polarization (CPMAS) NMR that help to constrain of the hydrous ringwoodite was oriented using a Bruker APEX II CCD diffractometer and the unit-cell parameter refined from the centering parameters of 48 the substitution mechanisms for H2O. X-ray reflections using a point detector on a Bruker P4 diffractometer. The refined cell parameter was a = 8.0702(5) Å giving a unit-cell volume of 525.59(11) Å3, EXPERIMENTAL METHODS indicating an H2O content of 1.0 ± 0.08 wt% (Smyth et al. 2003). Similarly, the Sample synthesis and characterization cell parameters of the hydrous wadsleyite were refined from the centering of 58 X-ray reflections as a = 5.6959(10) Å; b = 11.457(2) Å; c = 8.2540(14) Å; V = Synthesis and characterization of the anhydrous materials was described 3 538.6(2) Å . This gave a b/a ratio of 2.0114 indicating an H2O content of 0.27 ± previously (Stebbins et al. 2009). To summarize here, >99% 29Si-enriched for- 0.03 wt% (Jacobsen et al. 2005). sterite (Mg2SiO4) starting material was made by grinding together MgO and 29 SiO2 reagents, then heating for 7–12 days at 1500 °C, with several intermediate NMR spectroscopy grinding steps. One batch was made with 0.2 wt% Co3O4 (which decomposes to CoO at high temperature) added to speed spin-lattice relaxation (“Fo-1”); this was 29Si NMR spectra were collected with a Varian Inova 400 spectrometer (9.4 ground even more thoroughly and re-annealed (“Fo-2”) to evaluate the effects of Tesla) at 79.4 MHz, with a Varian/Chemagnetics “T3”-type probe with 3.2 mm further dispersion of the paramagnetic impurities. A second undoped batch (“Fo- zirconia rotors, and are referenced to tetramethyl silane (TMS). Small, high-pressure noCo”), as well as a sample with natural isotopic abundance (“Fo-unen,” 4.7% samples (4 to 11 mg) were spun at 20 kHz in a rotor containing a spacer to place 29Si) were made by similar methods. 29Si-enriched samples were prepared with a the sample in the field center; larger samples (e.g., forsterite, 46 mg) were spun small excess of MgO. at 12 kHz in a rotor with larger capacity. For MAS and static spectra, one pulse High-pressure syntheses were done at the Bayerisches Geoinstitut, using the acquisition was used, with pulse lengths of 0.7 μs (30° rf tip angle) and acquisition 5000 ton multi-anvil apparatus (Frost et al. 2004). For the anhydrous samples, two times of 164 ms. Pulse delays ranged from 0.2 to 1800 s, with numbers of transients runs were made with Fo-1 in unsealed 2 mm diameter Re foil capsules, at nominal ranging from more than 50 000 to as few as 16. Gaussian line broadening of 5 to 20 pressure and temperature of 21 GPa and 1500 °C. The first yielded a sample that Hz was applied during data processing, except for determinations of peak widths, was mostly ringwoodite (“RWD”); an apparent temperature over-run in the second where no apodization was used. The best spectra for the high-pressure samples produced a sample that was mostly wadsleyite (“WDS”). A third experiment at (e.g., S/N >1000 with 5 Hz line broadening) generally took about 12 h to collect, 20 GPa and 1300 °C on Fo-noCo produced a sample that was mostly ringwoodite with pulse delays of 1 to 10 s. Static (non-spinning) spectra were collected with the

(“RWD-noCo”). Hydrous wadsleyite (“WDS + H2O”) and hydrous ringwoodite same probe. Spin-lattice relaxation was studied both by comparison of one-pulse

(“RWD + H2O”) were prepared from the Fo-1 starting material mixed with Mg(OH)2 spectra with varying delay times, and, in more detail, with a saturation-recovery to give nominal total water contents of about 1.25 wt%, then welded into Pt capsules. experiment. In the latter, a train of 64, 90° pulses, spaced at 10 ms, was used to

TABLE 1. Electron microprobe data summary

Fo-unen Fo-1 Fo-noCo RWD RWD-noCo WDS WDS + H2O wt%

SiO2 43.3(1.1) 43.3(0.4) 42.5(1.4) 42.9(0.4) 42.4(0.6) 42.8(0.7) 43.3(0.9) MgO 57.9(0.8) 58.5(0.6) 59.0(1.7) 57.4(0.9) 58.2(0.8) 58.3(1.1) 58.4(0.7) FeO 0.03(0.02) 0.02(0.02) 0.01(0.01) 0.02(0.02) 0.02(0.02) 0.01(0.02) 0.02(0.01) CaO 0.08(0.02) 0.03(0.01) 0.04(0.01) 0.02(0.01) 0.02(0.01) 0.01(0.01) 0.02(0.01) CoO 0.01(0.01) 0.10(0.02) 0.01(0.02) 0.05(0.02) 0.02(0.02) 0.06(0.03) 0.04(0.02) Totals 101.3(1.6) 101.9(0.9) 101.7(1.7) 100.5(1.2) 100.6(1.3) 101.2(1.4) 101.8(0.4) mol%

SiO2 33.4 32.8 32.3 33.1 32.5 32.7 32.8 MgO 66.6 67.2 67.7 66.9 67.5 67.3 67.2 Notes: Standard deviations shown in parentheses; each analysis is the average of 8–12 data points. All samples except “Fo-unen” are 29Si enriched; all enriched samples are Co-doped unless labeled “noCo.” Fo = forsterite, RWD = ringwoodite, WDS = wadsleyite. STEBBINS ET AL.: NMR ON HIGH-PRESSURE Mg2SiO4 PHASES 907 saturate the magnetization, which was then allowed to recover for a time period τ rates from those studied here might contain greater amounts of and measured with a 90° observe pulse. 29Si{1H} cross-polarization magic-angle quenched-in Si-Mg disorder. 1 spinning spectra (CPMAS) were collected with constant-amplitude H CP pulses Also in ringwoodite, several small peaks, which relax at a (75 kHz) and 1H decoupling (50 kHz) during observation, with acquisition times of 41 ms, contact times from 0.2 to 20 ms, and spinning rates of 3.5 kHz. Pulse rate similar to the main peak and correspond to no known phases delays of 2 to 8 s were used, and generally 8000 transients were collected for each (labeled X1, X2, and X3 in Figs. 1 and 2) were observed, which we spectrum. Cross-polarization data were also collected on a sample of high-purity speculated could be related to defects generated during this re- natural talc, which was also useful for optimizing experimental conditions. ordering process or residual from high-temperature disorder. X3 RESULTS AND DISCUSSION is present for the Co-doped anhydrous and hydrous ringwoodite, but not in the Co-free sample, for unknown reasons. In addition, MAS NMR spectra in the forsterite and wadsleyite samples, but not in ringwoodite, Peak positions, widths, and relative areas for 29Si spectra several very small, fast-relaxing “extra” peaks were detected are listed in Table 2, and selected spectra are shown in Figures (arrows in Fig. 2), many of which have chemical shifts well 1 and 2. The 29Si MAS spectra for the forsterite samples, and outside the known ranges of 29Si chemical shifts in silicates. We for the anhydrous wadsleyite and ringwoodite samples, are suggested that these are likely to be caused by “pseudocontact described in our initial report (Stebbins et al. 2009). To briefly shifts,” resulting from through-space dipolar interactions with summarize, the very high signal-to-noise ratio obtained from trace impurities of paramagnetic ions (Grey et al. 1989, 1990). the isotopic enrichment resulted in a detection limit of VISi in The added Co2+, however, was apparently not the cause of these the ringwoodite of about 0.1 to 0.5%. However, this species was unusual features, but did increase peak widths significantly and not detected. Given recent theoretical predictions of 2 to 4% VISi speed relaxation rates as planned in the experimental design at the temperatures of synthesis (Panero 2008a), this suggested (see below). Other paramagnetic trace element cations could be that re-ordering may occur during cooling, as has been observed the source of these features [e.g., Fe, Ni, Cr, and Cu known to at high temperatures for MgAl2O4 spinel (Wood et al. 1986; be present in the starting materials at the level of 10 to 60 ppm Redfern et al. 1999). The contents of VISi (up to 5% of the total (Stebbins et al. 2009)], but only systematic studies of effects Si) suggested by earlier single-crystal XRD studies of anhydrous of trace element doping on spectra will be able to confirm this and hydrous ringwoodite (Hazen et al. 1993; Kudoh et al. 2000) hypothesis. Future studies of the temperature dependence of are much higher than this detection limit, and those results thus such shifted frequencies may also be useful for distinguishing seem inconsistent with the NMR data. This may reflect the close pseudocontact effects from through-bond Fermi contact shifts similarity in X-ray scattering of Si and Mg. It is also possible (Grey et al. 1990). that if re-ordering does occur during quench, defects or vacan- As in our previous report, spectra were collected over wide cies may remain that could affect long-range average polyhedral ranges of pulse delays, from 0.2 to 1800 or even 3600 s. With volumes and thus the interpretation of the X-ray structure. And, one minor exception noted below for the RWD + H2O, no addi- of course, there remains the possibility that ringwoodite samples tional features were seen in the longer-delay, more fully relaxed, synthesized at different temperatures and with different cooling spectra. In most cases we thus present partially saturated spectra

TABLE 2. 29Si MAS NMR peak locations, widths, and relative areas

Sample Phase or peak label δiso, ppm ± 0.1 FWHM, ppm ±0.02 (major) ±0.1 (minor) Relative area, %* Fo-unen Fo –61.8 0.2 Fo-1 Fo –61.8 0.6 Fo-2 Fo –61.8 0.9 Fo-noCo Fo –61.8 0.2 WDS WDS –78.7 0.6 RWD –81.3 0.5 6 RWD RWD –81.3 0.5 WDS –78.7 0.5 3.5

X1 + X2 (RWD defects?) –80.3 0.5 6 ± 2 (sum) X3 (RWD defect?) –87.5 0.7 0.7 RWD-noCo RWD –81.3 0.2 WDS –78.7 0.4 25

X1 (RWD defect?) –80.1 0.3 6 ± 2 (sum of X1 + X2) X2 (RWD defect?) –80.4 0.3 WDS + H2O WDS –78.7 0.7 B2 (Si2) –64.1 0.6 ≈1.5 B4 (Si4)† –75.1 0.7 ≈1.5 B3 (Si3)† –75.8 0.7 ≈1.5 B1 (Si1) –170.4 0.6 ≈1.5 WDS, WH1‡ –78.1 0.7 RWD + H2O RWD –81.4 0.4 X3 (RWD defect?) –87.8 0.5 0.2 RH1 (RWD Si-OH?) –77.5 0.5 4 ± 2 (sum of RH1 + RH2) RH2 (RWD Si-OH?) –79.7 0.5 ak (?) –180.7 0.3 ≈0.5 Notes: Very small “extra” peaks, thought to result from paramagnetic impurities, are not listed. Fo = forsterite, RWD = ringwoodite, WDS = wadsleyite, B = phase B, ak = akimotoite. * Estimated where possible from long-delay spectra. † Positions from CPMAS spectra, where better resolved. ‡ Clearly detected in CPMAS spectrum only. 908 STEBBINS ET AL.: NMR ON HIGH-PRESSURE Mg2SiO4 PHASES

IVSi VISi

X3 ak RWD+H2O a 300 s

B b 1 s 1 W

c R RWD RH 2 B2 RH1 B1

RWD+H2O d WDS+H O X2 2 X1 e W WDS RWD * RWD-noCo f Fo-noCo

-76 -78 -80 -82 -84 -86 0 -50 -100 -150 -200 ppm ppm FIGURE 1. 29Si MAS NMR spectra (10 s pulse delay) for anhydrous FIGURE 2. 29Si MAS NMR spectra, with 1 s pulse delays, except as ringwoodite with (RWD) and without (RWD-noCo) dopant, and a sample noted, and greatly enlarged vertical scales. Dashed boxes mark known IV VI with added H2O. W marks a minor wadsleyite component; X1 and X2 regions for Si and Si in silicates, except zunyite and Si phosphates. are minor peaks that may be some kind of defect sites; RH1 and RH2 (a) RWD + H2O, with pulse delay of 300 s, showing tiny peak at known mark peaks present only in the hydrous sample, which are accentuated position for akimotoite (“ak”). (b) RWD + H2O, with pulse delay of 1 in CPMAS (see below). s, to enhance S/N and to show very small peak for Si1 site of phase B

(B1). (c) Anhydrous RWD: X3 marks another peak that may be a defect

peak in RWD or an unknown phase. (d) WDS + H2O, showing peaks collected at delay times of 1 to 10 s, to accentuate signal-to- for phase B (B1, B2, with B3 and B4 hidden under the main peak); arrows noise and also to highlight minor, fast-relaxing features. Figure denote “extra” peaks probably related to paramagnetic impurities. (e) 1 shows the IVSi regions of the spectra for the three ringwoodite Anhydrous WDS. (f) Forsterite without Co dopant; “*” marks a spinning samples. As noted previously, the anhydrous samples contain sideband. minor amounts of wadsleyite, as well as the possible “defect” peaks X1 and X2. The hydrous sample contains neither of these labeled B1, B2, B3, and B4, for the four Si sites in the structure; minor features, but has instead two other small peaks at –77.5 the latter two of these are closest to the main wadsleyite peak and –79.7 ppm, which we denote RH1 and RH2. By subtracting a and are off-scale in the expanded-scale plot. At least one of these, fit of the main peak from the experimental spectrum, we estimate B1, can be seen in the spectrum for hydrous ringwoodite, but the the sum of the areas of these components is roughly 4 ± 2% of estimated abundance of phase B in the latter is much smaller, the total Si, in spectra with both short (1 s) and long (300 s) pulse totaling about 0.2 to 0.6% of the Si in relaxed, long-delay spectra. delays. As discussed below, these are greatly enhanced in the In both hydrous ringwoodite and hydrous wadsleyite samples, cross-polarization (CPMAS) spectra, and thus must represent Si the phase B peaks are greatly enhanced in the CPMAS spectra as sites that are especially close to the protons in the hydrous mate- expected from its high H content (see figures later in paper). rial. The IVSi regions for the one-pulse spectra of the wadsleyite Figure 2a also shows the spectrum collected with a 300 s samples are shown with the CPMAS data below. delay for RWD + H2O. Here, a tiny, sharp peak at –180.7 ppm In Figure 2, a wider frequency range is shown, with vertical is probably within uncertainties of that reported for akimotoite scales enlarged about 20 to 40× those shown in Figure 1. Here, of –181.0 (MgSiO3) (Stebbins and Kanzaki 1990). The peak the “extra” peaks, apparently related to paramagnetic impurities, area leads to an estimate of about 0.5 ± 0.2% of total Si in this are clearly seen for the Fo-noCo sample, for both the anhydrous phase. This peak is not visible in the shorter delay spectra with and hydrous wadsleyite, but not for the ringwoodite samples. We much higher signal-to-noise ratios and thus much lower detection suggested previously (Stebbins et al. 2009) that this difference limits, which implies that it is present in a separate phase with a could be the consequence of the much higher symmetry of the slower spin-lattice relaxation rate than that of the ringwoodite. If latter. In the hydrous wadsleyite, four small peaks are seen at the peak is due to akimotoite, it may be related to the formation –64.1, –75.1, –75.8, and –170.4 ppm, which correspond well to of the minor amount of phase B, or may suggest some minor VI those previously reported for phase B [Mg12Si4O19(OH)2] (Phil- compositional heterogeneity in this sample. This Si peak could, lips et al. 1997). From the fully relaxed spectrum, we estimate of course, result instead from a small amount of residual site that about 6% of the Si in this sample is in phase B, and thus disorder in the ringwoodite and only coincidentally resemble the that it contributes about 0.2 wt% to the total water content of the spectrum of akimotoite, but the apparent difference in relaxation sample (i.e., about one fifth of the nominal added H2O). Peaks are rate could be difficult to explain. STEBBINS ET AL.: NMR ON HIGH-PRESSURE Mg2SiO4 PHASES 909

Static spectra Non-spinning (static) spectra are a straightforward way to constrain the magnitude and symmetry of the chemical shift anisotropy (CSA), although analyses of spinning sideband pat- terns for slow MAS experiments often give good results as well (Smith et al. 1983). Representative static spectra for forsterite, wadsleyite, and ringwoodite are shown in Figures 3 and 4. For- sterite is an interesting test case, as it has been unusually well studied by NMR, including rare single-crystal measurements and theory detailed theoretical calculations, as reviewed recently (Ashbrook et al. 2007b, 2007c). Our results agree well with a previous static spectrum for a sample with natural isotopic abundance, which expt in turn was accurately predicted by density functional theory (CASTEP) calculation, yielding best estimates of the anisotropy fit Δcs of 29.3 ppm and asymmetry parameter η of 0.61 as defined previously; fits to the experimental spectrum yielded 32.3 ppm and 0.57 (Ashbrook et al. 2007b, 2007c). The calculation was 0 -20 -40 -60 -80 -100 -120 -140 ppm based on the optimization of the X-ray structure of an olivine near to Fo90 in composition (Brown 1970). Our spectrum for FIGURE 3. Static 29Si NMR spectra for Co-doped forsterite. “Theory” isotopically enriched forsterite may be slightly broadened by was calculated in a recent publication (Ashbrook et al. 2007b); “exp’t” Si-Si dipole-dipole couplings (expected to be a few hundred shows the experimental spectrum; “ﬁt” is a ﬁt to these data; see text for Hertz), and is best fitted with slightly different parameters, with parameters used to calculate spectra.

δiso from the MAS data at –61.8 ppm, Δcs = 32.3 ± 0.5 ppm, and η = 0.70 ± 0.05, convolved with a Gaussian broadening function of about 500 ± 200 Hz FWHM. For wadsleyite, this same theoretical approach (Ashbrook et al. 2007b) yielded δiso = –80.0 Δ η ppm, cs = –40.5 ppm, and = 0.11. These give a reasonably R good prediction of the experimental peak shape (Fig. 4), which is, however, better fitted by a somewhat narrower CSA, with δiso from the MAS data at –78.7 ppm, Δcs = –37.0 ppm, and η = 0 to 0.1, with a Gaussian broadening of about 1000 Hz. The sign of Δ and the near-axial symmetry (η ≈ 0) are sensible for a Q1 cs theory silicate where the longest Si-O bond is generally to the bridging oxygen (Grimmer et al. 1981). WDS+H O In the spectrum for anhydrous wadsleyite, the narrow ad- 2 ditional peak in the center is due to the minor ringwoodite component, as confirmed by static spectra of the three samples WDS of this phase. For all of the latter, peaks are symmetrical, roughly

Gaussian in shape (although having minor Lorentzian “tails”), fit with FWHM of 760 Hz (9.5 ppm). The Co doping caused little or no broadening of the static line shape. The cubic symmetry RWD-noCo of this mineral indicates that the CSA should be 0, consistent with observation. Some of the residual peak width can be at- 0 -20 -40 -60 -80 -100 -120 -140 tributed to Si-Si dipolar coupling. From the structure (Sasaki et ppm al. 1982) and a simple “Van Vleck” summation of 1/r6 for the FIGURE 4. Static 29Si NMR spectra for hydrous and anhydrous second moment (Abragam 1961; MacKenzie and Smith 2002), wadsleyite, with “theory” and “ﬁt” as in Figure 3, along with experimental the dipolar linewidth can be calculated as about 400 Hz, leaving spectrum for undoped RWD. the remaining width for minor contributions from other factors such as paramagnetic impurities and/or slight disorder. A similar efficiency of this process is enhanced by the number of close 1H 29 conclusion was reached for Si-enriched MgSiO3 perovskite neighbors to a given Si site, depends strongly on their distance (Kirkpatrick et al. 1991). (1/r3), and is affected in complex ways by dynamics and by relaxation of the 1H spin system. Although extraction of structural CPMAS data and H site characterization information from CPMAS spectra can be difficult and generally In the cross-polarization NMR experiment, nuclear spin en- non-quantitative, such data can provide useful constraints on ergy is transferred from 1H to 29Si nuclei. As discussed in detail relative Si-H distances when comparing sites in similar materi- in several recent CPMAS studies of hydrous silicates (Phillips als. Because of the strong enhancement of signals from Si in et al. 1997; Oglesby and Stebbins 2000; Xue et al. 2008), the stoichiometrically hydrous phases or from Si relatively close 910 STEBBINS ET AL.: NMR ON HIGH-PRESSURE Mg2SiO4 PHASES

to H in nominally anhydrous phases with minor water contents, example, comparing data for WDS + H2O for tc = 1 vs. 6 ms CPMAS spectra can often also provide insights into the presence (Fig. 5), it is obvious that the phase B peaks grow in at differ- of minor components. ent rates, as studied in detail previously in a sample composed Figure 5 illustrates representative CPMAS spectra for the primarily of this mineral (Phillips et al. 1997). It is also clear hydrous wadsleyite sample, compared with MAS (non-CP, often that the main wadsleyite peak at –78.7 ppm (W) grows in more denoted as “1 pulse”) spectra for the hydrous and anhydrous slowly than a shoulder at –78.1 ppm (WH1), suggesting that the samples. As noted above, the latter (only) contains a minor latter is a minor component (not obvious in the one pulse spectra ringwoodite component. Also as noted above, the MAS spectra because of its low intensity) representing Si that are closer than 1 of the hydrous wadsleyite reveal four extra peaks (B1–4) known average to included H. to be due to phase B (Phillips et al. 1997), which are greatly Detailed analysis of results required fitting of spectra to enhanced in the CPMAS spectra. Other than the VISi peak for extract the areas of overlapping components. This was done for phase B (B1 in Fig. 2), no signals for octahedral Si were seen in each of the two hydrous samples by first fitting the spectra at any of the CPMAS spectra, which had typical signal-to-noise the longest contact times, which generally had the most intense ratios of about 180 (WDS + H2O) to 40 (RWD + H2O). In the peaks, and then fitting the remainder of the spectra with result- CP experiment, the energy transfer from 1H to 29Si takes place ing peak positions, widths, and shapes (Gaussian fractions) as during a “contact time” (tc), which was adjusted within the range constraints. This provided more robust fits of the noisier data at of about 0.2 to 20 ms. As this time is increased, the observed 29Si shorter contact times, although residuals are somewhat greater magnetization gradually builds up, at a rate related most strongly than less-constrained fits. Plots of magnetization (fitted peak ar- in such materials to the strength of the dipolar coupling to the eas, M) vs. contact time can then be analyzed with a well-known 1 3 nearest H neighbor, which scales with 1/(rSi-H) , where rSi-H is equation for the sum of exponential growth and exponential the distance from Si to H, although all 1H neighbors make some decay, which is often a good approximation in systems of this contribution. [For a fully known structure, a more accurate type (dilute spins, static protons) (Phillips et al. 1997): description can be made in terms of the second moment of the 6 Si-H distribution, which involves summing 1/(rSi-H) for all Si-H M(tc) = [M0 /(1 – TSiH/T1ρH)] [exp(–tc/T1ρH) – exp(–tc/TSiH)] (1) pairs (Phillips et al. 1997).] Spectra collected at different contact times thus may have very different relative peak intensities. For Here, the derived parameters are M0, which is the limiting, infinite time value for the CP magnetization, T , the exponential time B SiH 3 29 constant for the growth of the observed Si signal, and T1ρ(H), that B4 for the decay of the 1H magnetization during the pulse sequence, WH 1 which in turn may eventually begin to reduce the observed signal. W For most of the sites for most of the samples observed here, the latter was long enough to be only roughly estimated, not directly derived from regression of the data. However, this does not B2 greatly affect estimates of the more structurally interesting TSiH. Fitted curves for the phase B peaks in the hydrous wadsleyite

sample, and the sum of the main wadsleyite peak and the WH1 shoulder, are shown in Figure 6, with derived parameters in WDS+H O, CP 2 Table 3. Separation of the two wadsleyite components by fitting a t = 6 ms c was not very robust because of their strong overlap. However, b the two derived TSiH values do at least qualitatively reflect the t = 1 ms c clearly observed differences in their growth rate with contact time as seem in Figure 5.

As shown in the table, the measured TSiH values for phase R WDS+H2O, 1 pulse B are systematically smaller than those reported previously c scale x 5 (Phillips et al. 1997), probably because of differences in sample d scale x 1 spinning rate and pulse sequence used. Nonetheless, the relative sizes of TSiH for the four peaks are similar between the two e WDS, 1 pulse studies, supporting at least a rough correlation with closest Si-H distance in this mineral as previously documented (Phillips et al. -60 -65 -70 -75 -80 -85 ppm 1997). The B1 and B2 peaks have the longest TSiH values in both studies, although in our results these values are the same within FIGURE 5. 29 1 (a) Si { H} CPMAS spectrum for WDS + H2O, with error. The corresponding sites have the longest Si-H distances as contact time of 6 ms. (b) CPMAS spectrum with contact time of 1 ms. well, and, given the 1/r3 dependence, the least sensitivity of T to (c) 1 pulse (no CP) spectrum, 1 s pulse delay, vertical scale enlarged SiH ×5. (d) As in c, vertical scale ×1. (e) One-pulse spectrum of anhydrous the distance. In phase B, the shortest bonds for all protons (<0.1 nm) are to O atoms otherwise bonded only to Mg (i.e., “Mg-OH” WDS, showing minor RWD component, R. B2–4 show three of the peaks for phase B; W marks the peak as in anhydrous wadsleyite, WH1 is an groups, in which the O may have more than one Mg neighbor); additional shoulder that appears in the hydrous sample and grows in none are directly connected by short bonds to O atoms in the first more rapidly in CPMAS. shell of Si. Phase B thus has no “Si-OH” groups, noting that this STEBBINS ET AL.: NMR ON HIGH-PRESSURE Mg2SiO4 PHASES 911

2 fraction of Si with a somewhat shorter distance, perhaps about 0.28 nm. These distances are thus typical of those seen for Si

W+WH1 sites near to Mg-OH groups in hydrous Mg silicates, and are considerably longer than is typical for “Si-OH” groups in silicates 1.5 (Oglesby and Stebbins 2000). Thus, the NMR results provide general support for the prevalent model of most H in hydrous wadsleyite being bonded to the O1 oxygen, which is otherwise

B4 bonded only to Mg (Smyth 1994; Kudoh et al. 1996; Smyth et al. 1 1997). The 29Si NMR data do not, however, shed much light on B peak 3 uncertainties in exact proton positions, proton site order/disorder, area or extent of H-bonding. Further studies with 1H NMR (Phillips et al. 1997; Kohn et al. 2002) and 29Si-1H double resonance methods B 0.5 1 (Xue et al. 2008) may help provide such details. CPMAS results for the hydrous ringwoodite are dramatically different from those for wadsleyite. As shown in Figure 7, some B 2 phase B is present, again with peaks accentuated relative to the 0 one-pulse data but considerably smaller than in the hydrous 0 5 10 15 20 contact time, ms wadsleyite, as expected from their much smaller contribution to FIGURE 6. Plot of ﬁtted peak areas vs. contact time for the components the MAS spectra. Again comparing spectra with relatively short (1 ms) vs. relatively long (6 ms) contact times, it can be seen that of the CPMAS spectra for WDS + H2O. Because of incomplete resolution, data for the main W peak and the WH1 shoulder are combined. Typical two new peaks grow in very rapidly. These were seen as small uncertainties are shown on one point on the uppermost curve. features in the one-pulse spectra, RH1 and RH2. Their positions

bracket the W and WH1 peaks in hydrous wadsleyite (also shown TABLE 3. Results of fits to CP curves (peak area vs. contact time in in Fig. 6 for comparison), and are thus clearly distinct. Their CPMAS NMR experiment) , chemical shifts correlate with no known hydrous magnesium Peak (ppm) Phase T , ms* T † Si-H, nm† T ρ , ms*‡ SiH SiH 1 ,H silicates, most of which have been previously studied by NMR WDS + H2O:

B1, –170.4 ph.B, Si(1) 8.8 27 0.44 5000 (see introduction). Their displacement of several parts per mil- B3, –75.9 ph. B, Si(3) 3.9 7.5 0.31 5000 lion from the main ringwoodite peak (R) suggests a significant B4, –75.1 ph. B, Si(4) 1.8 2.9 0.265 160 B2, –64.0 ph. B, Si(2) 8.7 40 0.48 5000 W, –78.9 WDS 5.2§ 5000§

WH1, –77.9 WDS 2.1§ 81§ RH1 W + WH1 WDS 3.3 5000 RH2 RWD + H2O: R, –81.3 RWD 7.7 5000 R RH2, –79.6 RWD?|| 1.2 100 B4 B3 RH1, –77.4 RWD?|| 0.75 30 B3 + B4 ph. B, Si(3,4) 3.1 200 talc: –97.2 talc 5.2 0.315# 430

* This study. Uncertainties in fits of TSiH are roughly 10% of values. † Phillips et al. (1997). B ‡ T1ρ,H could be well constrained by the data only when less than about 300 to 500 2 ms. Values of 5000 were arbitrarily fixed during fitting, but do not significantly affect derived TSiH results. RWD+H2O, CP t = 6 ms § Because of severe overlap of peaks W and WH1, parameters fitted to their a c individual areas are relatively poorly constrained, but TSiH values do illustrate change in their relative proportions with increasing contact time. || These peaks are either Si sites in ringwoodite that are especially close to H, or in an unknown hydrous phase. b tc = 1 ms # Oglesby and Stebbins (2000). designation does not imply that the oxygen involved is bonded RWD+H O, 1 pulse only to one Si and one H. In fact, in silicates with such groups, c 2 there are generally one or more additional mono- or divalent W cation neighbors (Oglesby and Stebbins 2000). New CPMAS WDS+H O, CP data were collected on a sample of natural talc [Mg3Si4O10(OH)2] d 2 using identical methods to those applied to the high-pressure -60 -65 -70 -75 -80 -85 phases. Talc also has Mg-OH groups and an Si-H distance slightly ppm longer than that for Si3 in phase B, and yields a TSi-H value that is 29 1 FIGURE 7. (a) Si { H} CPMAS spectrum for RWD + H2O, with slightly longer than that observed for the B3 peak. We conclude contact time of 6 ms. (b) CPMAS spectrum with contact time of 1 ms. that the mean value of TSiH for the W + WH1 peaks in hydrous (c) One-pulse (no CP) spectrum, 1 s pulse delay. (d) CPMAS spectrum wadsleyite (3.3 ms) corresponds to a mean closest Si-H distance of hydrous wadsleyite, contact time of 6 ms, for comparison of peak of roughly 0.3 nm. The WH1 shoulder may represent a smaller positions. 912 STEBBINS ET AL.: NMR ON HIGH-PRESSURE Mg2SiO4 PHASES difference in their first cation coordination shells, for example the peaks for other components either in the ringwoodite or phase substitution of H+ for Mg2+. The appearance of two discrete new B. Such a complication could result if there was some disorder peaks also suggests complexities to the substitution mechanism, among such sites. which may eventually be elucidated by theoretical calculations The peak in the CPMAS spectra of hydrous ringwoodite that or further experiments. corresponds to the main peak in the MAS spectra (R) grows in

In plots of peak area vs. contact time (Fig. 8), the RH1 and much more slowly, with TSiH of 7.7 ms. Thus, this represents Si

RH2 peaks clearly grow in much faster than any of the phase B in tetrahedra that are at least one or two polyhedra away from H peaks or those in wadsleyite; they can also be seen to begin to dopants (very roughly 0.4 nm). Because this peak is somewhat

“decay,” due to relatively short T1ρH values. There is a strong broader than that in the MAS spectra, it could contain compo- implication of particularly short nearest Si-H distances, as the TSiH nents at similar chemical shifts that have TSiH values (and Si-H values are extremely short, about 1 ms. These peaks are probably distances) that are somewhat lower (i.e., in the range expected for due to Si with direct links to H, i.e., “Si-OH” groups as defined a Si neighbor to a Si vacancy containing H+). This possibility has above. If these Si sites are indeed part of the hydrous ringwoodite been suggested to account for some of the H substitution based structure, this is just what would be expected if H incorporation on X-ray data (Kudoh et al. 2000). Very recent first-principles primarily involves the Mg2+ = 2H+ mechanism. Unless there are theoretical calculations have suggested that 66% of the H in Si vacancies or IVMg defects, all O in the structure are bonded to hydrous ringwoodite are associated with Mg vacancies and thus one Si, thus each added H must be part of an “Si-OH” linkage. would be detected as “Si-OH,” 24% with Si vacancies, and the The predominance of this mechanism is supported by detailed rest associated with other types of defects (Panero 2008b). The X-ray diffraction studies, which generally show that most or CPMAS data alone thus cannot rule out the presence of some H all cation vacancies are at octahedral sites (Kudoh et al. 2000; that are not in “Si-OH” sites. Smyth et al. 2003). Quantifying populations of species from CPMAS spectra is Spin-lattice relaxation problematical because of the complexities of cross-polarization The high-signal to noise ratios provided by the 29Si enrich- dynamics. However, the RH1 and RH2 peaks from the one-pulse ment, together with narrow NMR peaks, facilitated a relatively experiments can be analyzed with more confidence. If each of detailed study of the spin-lattice relaxation behavior of all of the the Si represented by these peaks is assumed to account for one samples. In the “saturation-recovery” experiment used here, an H, then their 4 ± 2% combined area implies 0.04 moles of H per initial train of pulses saturates the magnetization. After a relax- mole of Mg2SiO4, or about 0.25 ± 0.12 wt% H2O. This figure ation delay time τ, the re-growth of the magnetization, caused is well below the nominal water content of 1.25 wt% and the by spin-lattice relaxation, is measured with an “observe” pulse. 1 wt% result from XRD data on a single crystal. This suggests τ from 0.2 to 1800 s were explored. Representative data for the the possibility that a substantial amount of the NMR signal from resulting peak areas, normalized to values at the longest delays, Si-OH sites (or of Si with more distant H neighbors) might be are shown with linear scales in Figure 9. Results for all samples unresolvable from the relatively broad, overlapping bases of the are displayed on a log-log plot in Figure 10 to better display the initial relaxation curves. The trends of the data, as well as fitted curves discussed below and comparison of absolute intensities 2.5 scaled by sample weight, indicate that the magnetization is usu- ally at least 95% recovered after a delay of 1800 s. In the case of

RH2 the ringwoodite samples containing minor wadsleyite (and vice 2 versa), peaks were fitted to obtain areas for the major phases. In systems with rapid physical diffusion (e.g., liquids) or rapid spin diffusion, spin-lattice relaxation generally follows a 1.5 simple exponential behavior, with

peak M(τ) = M {1 – exp[–(τ/T )]} (2) area ∞ 1

1 RH1 where M∞ is the fully recovered observed magnetization (z-

B3 + B4 component) and T1 is the “relaxation time” for all spins in the

0.5 sample (Abragam 1961). (Spin diffusion involves the exchange R of energy via dipolar coupling between nuclear spins that are abundant, relatively close, and of relatively high gyromagnetic ratios γ, e.g., 1H in organic materials or 19F in fluorides.) However, 0 0 5 10 15 20 in solid materials where both physical and spin diffusion are slow, contact time, ms notably rigid oxide materials without abundant, high-γ spins, the 29 FIGURE 8. Plot of fitted peak areas vs. contact time for the relaxation of spin 1/2 nuclides (e.g., Si) is often dominated by direct, through-space dipolar couplings between the observed components of the CPMAS spectra for RWD + H2O. Because of low signal-to-noise, the B3 and B4 peaks are combined here and data for B1 nuclei and the much stronger magnetic dipoles associated with and B2 are omitted. Typical uncertainties are shown on one point on unpaired electron spins (Hartman et al. 2007), for example on uppermost curve. paramagnetic impurity cations such as Fe2+ or Co2+ in the samples STEBBINS ET AL.: NMR ON HIGH-PRESSURE Mg2SiO4 PHASES 913

1.2 Ideally, n should be 0.5, but a mixture of the behaviors given in Equations 2 and 3 can result in fitted exponents between 0.5 and 1.0 (Hartman et al. 2007; Malfait et al. 2008). 1 For simplicity and to allow direct comparisons of the two equations with the same number of adjustable parameters, we 0.8 fixed n = 1 or 0.5 and fitted each data set with only two param-

eters, either M∞ and T1 (Eq. 2) or M∞ and T′ (Eq. 3). Peaks areas τ 1800 (normalized to data at the longest observed ) vs. relaxation de- 0.6 lays (not logs) were fitted by non-linear least-squares regression; M / typical fits are shown for two of the samples with both the linear 0.4 RWD (Fig. 9) and log-log scales (Fig. 10). As expected from several Fo-noCo relaxation time studies, including detailed work on transition- metal doped forsterite (Hartman et al. 2007), Equation 3 gives 0.2 n = 0.5 n = 1 much better fits; derived parameters are listed in Table 4. For

most samples, fitted M∞ values are only 1 to 4% higher than 0 the data at the longest observed τ, suggesting nearly complete 0 500 1000 1500 2000 relaxation delay, s relaxation by 1800 s. However M∞ reaches about 9% above M1800 for the most slowly relaxing sample (Fo-noCo), and is actually FIGURE 9. Linear plot of saturation-recovery data, showing the ratio of the recovered magnetization to that at 1800 s, vs. the recovery time, slightly below M1800 for the Co-doped ringwoodite, presumably for two representative samples, Fo-noCo and anhydrous ringwoodite. simply because of fitting uncertainties. The T1 fits (Eq. 2 or 3 Data for the latter is displaced upward by 0.18 for clarity. Also shown with n = 1) were poor for all samples (χ2 typically at least 10× are curves ﬁtted with Equation 1 (standard “T1”, n = 1) and Equation 2 worse), especially at short delays: the shapes of the fitted curves (stretched exponential, n = 0.5). are clearly inaccurate, as can be seen in both linear and log-log

plots. In Table 4, we include the resulting T1 values to facilitate 0.5 comparisons with other studies, but we re-emphasize that these

are poor approximations. Furthermore, fitted T1’s decrease greatly and systematically as the longest τ data are excluded 0 from the fits. This may be one reason why successively more

extended studies of relaxation in, for example, SiO2 glass, typi-

cally produce longer and longer estimates of T1 (Gladden et al. -0.5 Fo-noCo 1986; Malfait et al. 2008). We also note that better stretched

) Fo-2 exponential fits could be obtained by allowing n to be adjustable Fo-unen 1800 and introducing as well another multiplier term (Hartman et al. RWD-noCo -1 2007), but the physical meaning of the extra parameters may RWD+H O 2 not always be obvious. log (M/M RWD For the 29Si-enriched forsterite, the Co doping has a large WDS effect on the relaxation, reducing the stretched exponential time -1.5 WDS+H O 2 constant T′ by almost an order of magnitude. The curve for the Fo n = 0.5 n = 1 natural abundance forsterite falls in between. It is likely that in the RWD two undoped samples, relaxation is controlled by unknown trace -2 -101234amounts of paramagnetic impurities in the reagents, most likely log (relaxation delay) (s) Fe2+. In contrast, data for all three of the ringwoodite samples,

FIGURE 10. Logarithmic (log10) plot of saturation-recovery data for whether Co-doped or not, are close to each other and to the curve all samples. Fit lines are shown for two samples only (RWD and Fo- for the Co-doped forsterite. Although EPMA data indicate that noCo), and are the same curves as in Figure 9. the Co content of the ringwoodite is somewhat lower than that of its forsterite precursor (possibly through diffusion into the Re studied here. Because the effect of such a dipolar coupling scales high-pressure capsule?), there is clearly a significant effect of as the inverse sixth power of the distance from the nuclide to the electron spin, in the case of dilute impurities there will generally TABLE 4. Fitted parameters for spin-lattice relaxation curves ′ be a wide range of distances and thus a wide range of relaxation Sample T1 (Eq. 1), s T (Eq. 2), s M∞/M1800 (Eq. 2) Fo-noCo 114 (30) 245 (49) 1.09 (0.05) rates throughout the sample. That is, some nuclear spins that Fo-2 13 (5) 26.6 (0.8) 1.04 (0.01) happen to be very close to an impurity will relax very rapidly, Fo-unen 82 (30) 164 (15) 1.04 (0.02) whereas those very far from any impurity may relax very slowly. RWD-noCo 13 (4) 27 (2) 1.01 (0.01) RWD 12 (3) 26 (2) 0.98 (0.01)

It has been shown (Tse and Hartmann 1968; Hayashi 1994; Hart- RWD + H2O 11 (3) 21 (2) 1.02 (0.01) man et al. 2007) that in this common case, relaxation is much WDS 40 (11) 63 (7) 1.03 (0.02) better described by a “stretched” exponential, with: WDS + H2O 40 (11) 66 (9) 1.03 (0.03) Notes: Standard errors from fit are shown in parentheses. Magnetization or “M” values fitted by either Equation 1 or 2 (see text) are the integrated or fitted peak n M(τ) = M∞ {1 – exp[–(τ/T′) ]} (3) areas for the major phase in each sample. 914 STEBBINS ET AL.: NMR ON HIGH-PRESSURE Mg2SiO4 PHASES the dopant on the linewidth (Fig. 1). The apparent lack of effect Polytechnic Institute, Blacksburg. Derighetti, B., Hafner, S., Marxer, H., and Rager, H. (1978) NMR of 29Si and on the relaxation is thus surprising and suggests that Co cannot 25 Mg in Mg2SiO4 with dynamic polarization technique. Physics Letters, 66, be dominating the relaxation, perhaps because of structural dif- 150–152. ferences between the phases, e.g., the much higher symmetry of Devreux, F., Boilot, J.P., Chaput, F., and Sapoval, B. (1990) NMR determination of the fractal dimension in silica aerogels. Physical Review Letters, 65, the ringwoodite (spinel) structure than that of the forsterite. It is 614–617. possible as well that spin diffusion might begin to be important in Frost, D.J., Poe, B.T., Trønnes, R.G., Liebske, C., Duba, A., and Rubie, D.C. some of these phases, given the high level of isotopic enrichment, (2004) A new large-volume multianvil system. Physics of Earth and Planetary Interiors, 143–144, 507–514. although the spinning rate of 20 kHz is likely to mostly quench Fujino, K., Sasaki, S., Takeuchi, Y., and Sadanaga, R. (1981) X-ray determination this process. Spin diffusion was supported by slow MAS NMR of electron distributions in forsterite, fayalite, and tephroite. Acta Crystal- 29 lographica B, 37, 513–518. experiments on Si-enriched MgSiO3 perovskite, resulting in Gladden, L.F., Carpenter, T.A., and Elliot, S.R. (1986) 29Si MAS NMR studies of rapid, simple exponential relaxation (T1 of only 8 s) (Kirkpatrick the spin-lattice relaxation time and bond-angle distribution in vitreous silica. Philosophical Magazine, 53, L81–L87. et al. 1991). This phase has corner-shared SiO6 octahedra and thus Grey, C.P., Dobson, C.M., Cheetham, A.K., and Jakeman, R.J.B. (1989) Studies of a continuous network for Si-Si spin diffusion with Si-Si distances rare-earth stannates by 119Sn MAS NMR. The use of paramagnetic shift probes of 0.343 to 0.345 nm (Horiuchi et al. 1987). Although Si-Si dis- in the solid state. Journal of the American Chemical Society, 111, 505–511. tances are somewhat longer in both ringwoodite and forsterite Grey, C.P., Smith, M.E., Cheetham, A.K., Dobson, R., and Dupree, R. (1990) Y-89 MAS NMR study of rare-earth pyrochlores-paramagnetic shifts in the solid because the SiO4 tetrahedra are not directly connected, the closest state. Journal of the American Chemical Society, 112, 4670–4680. such distance is somewhat shorter in ringwoodite [3-D network Grimmer, A.-R., Peter, R., Fechner, E., and Molgedey, G. (1981) High-resolution 29Si NMR in solid silicates. Correlation between shielding tensor and Si-O of 0.349 nm Si-Si distances (Sasaki et al. 1982)] than in forsterite bond length. Chemical Physics Letters, 77, 331–335. (2-D chains of 0.362 nm distances) (Fujino et al. 1981). Both Grimmer, A.-R., von Lampe, F., Mägi, M., and Lippmaa, E. (1983) Hochauflösende samples of wadsleyite were made from the Co-doped forsterite, 29Si-NMR an festen silicaten; Einfluss von Fe2+ in olivinen. Zeitschrift für Chemie, 23, 343–344. and gave almost indistinguishable relaxation curves (Fig. 10). In Hartman, J.S., Narayanan, A., Rigby, S.S., Sliwinski, D.R., Halden, N.M., and this phase, pairs of SiO4 tetrahedra are corner-shared (Si-Si dis- Bain, A.D. (2007) Heterogeneities in sol-gel-derived paramagnetics-doped tance of 0.298 nm), but there is not a continuous network of close forsterites and willemites-electron microprobe analysis and stretched- exponential 29Si NMR spin-lattice relaxation studies. Canadian Journal of Si neighbors (typical next Si-Si distance of 0.360 nm) (Smyth et Chemistry, 85, 56–65. al. 1997). Added H2O had little or no effect on the relaxation of Hayashi, S. (1994) Effects of magic-angle spinning on spin-lattice relaxations in 1 talc. Solid State Nuclear Magnetic Resonance, 3, 323–330. either ringwoodite or wadsleyite (Fig. 10), indicating that the H Hazen, R.M., Downs, R.T., Finger, L.W., and Ko, J. (1993) Crystal chemistry was too dilute to significantly enhance spin diffusion from 29Si to of ferromagnesian silicate spinels: Evidence for Mg-Si disorder. American paramagnetic centers. Finally, we note that at τ << T′ the limiting Mineralogist, 78, 1320–1323. Horiuchi, H., Ito, E., and Weidner, D.J. (1987) Perovskite-type MgSiO3: Single- slope in a log-log plot of Equation 3 becomes 0.5. This is the crystal X-ray diffraction study. American Mineralogist, 72, 357–360. expected initial slope for relaxation by through-space coupling Ita, J. and Stixrude, L. (1992) Petrology, elasticity, and composition of the mantle to paramagnetic impurities, when the spatial distribution is three- transition zone. Journal of Geophysical Research, 97, 6849–6866. Jacobsen, S.D., Demouchy, S., Frost, D.J., Boffa-Ballaran, T., and Kung, J. (2005) dimensional. In materials such as silica gel and nano-scale phase A systematic study of OH in hydrous wadsleyite from polarized FTIR spectros- separated Li2Si4O9 glasses, this initial slope has been observed copy and single-crystal X-ray diffraction: Oxygen sites for hydrogen storage in the Earth’s interior. American Mineralogist, 90, 61–70. to be reduced, indicating a reduced dimensionality (Devreux et Kanzaki, M., Stebbins, J.F., and Xue, X. (1992) Characterization of crystalline al. 1990; Sen and Stebbins 1994). and amorphous silicates quenched from high pressure by 29Si MAS NMR spectroscopy. In Y. Syono and M.H. Manghnani, Eds., High Pressure Research: ACKNOWLEDGMENTS Applications to Earth and Planetary Sciences, p. 89–100. Terra Scientific Publishing Co., Tokyo. We thank Bob Jones for EPMA data and Clare Grey for helpful discussions of Kirkpatrick, R.J., Howell, D., Phillips, B.L., Cong, X.D., Ito, E., and Navrotsky, A. 29 contact shifts, and two anonymous reviewers whose comments helped to improve (1991) MAS NMR spectroscopic study of Mg SiO3 with perovskite structure. the paper significantly. We acknowledge the support of the NSF (grant EAR- American Mineralogist, 76, 673–676. 0408410 to J.F.S., EAR-0711165 to J.R.S., and EAR-0537813 to W.R.P.) and of Kohlstedt, D.L., Keppler, H., and Rubie, D.C. (1996) Solubility of water in the α, the Ohio Supercomputer Center (PAS0238-1). β, and γ phases of (Mg,Fe)2SiO4. Contributions to Mineralogy and Petrology, 123, 345–357. REFERENCES CITED Kohn, S.C., Brooker, R.A., Frost, D.J., Slesinger, A.E., and Wood, B.J. (2002) Ordering of hydroxyl defects in hydrous wadsleyite (β-Mg2SiO4). American Abragam, A. (1961) Principles of Nuclear Magnetism, 599 p. Clarendon Press, Mineralogist, 87, 293–301. Oxford. Kudoh, Y., Inoue, T., and Arashi, H. (1996) Structure and crystal chemistry of 17 Ashbrook, S.E., Berry, A.J., and Wimperis, S. (2001) O multiple-quantum MAS hydrous wadsleyite, Mg1.75SiH0.5O4: Possible hydrous magnesium silicate in the NMR study of high-pressure hydrous magnesium silicates. Journal of the mantle transition zone. Physics and Chemistry of Minerals, 23, 461–469. American Chemical Society, 123, 6360–6366. Kudoh, Y., Kuribayashi, T., Mizobata, H., and Ohtani, E. (2000) Structure and Ashbrook, S.E., Berry, A.J., Hibberson, W.O., Steuernagel, S., and Wimperis, S. cation disorder of hydrous ringwoodite, γ-Mg1.89Si0.98H0.30O4. Physics and 17 (2005) High-resolution O MAS NMR spectroscopy of forsterite (α-Mg2SiO4), Chemistry of Minerals, 27, 474–479. wadsleyite (β-Mg2SiO4), and ringwoodite (γ-Mg2SiO4). American Mineralo- MacKenzie, K.J.D. and Smith, M.E. (2002) Multinuclear Solid-State NMR of gist, 90, 1861–1870. Inorganic Materials, 727 p. Pergamon, New York. Ashbrook, S.E., Berry, A.J., Frost, D.J., Gregorovic, A., Pickard, C.J., Readman, Malfait, W.J., Halter, W.E., and Verel, R. (2008) 29Si NMR spectroscopy of silica 17 29 J.E., and Wimperis, S. (2007a) O and Si NMR parameters of MgSiO3 phases glass: T1 relaxation and constraints on the Si-O-Si bond angle distribution. from high-resolution solid-state NMR spectroscopy and first-principles calcula- Chemical Geology, 256, 269–277. tions. Journal of the American Chemical Society, 129, 13213–13224. McMillan, P., Akaogi, M., Ohtani, E., Williams, Q., Nieman, R., and Sato, R. (1989) Ashbrook, S.E., Le Polles, L., Pickard, C.J., Berry, A.J., Wimperis, S., and Farnan, I. Cation disorder in garnets along the Mg3Al2Si3O12-Mg4Si4O12 join: An infrared, (2007b) First-principles calculations of solid-state 17O and 29Si NMR spectra of Raman, and NMR study. Physics and Chemistry of Minerals, 16, 428–435.

Mg2SiO4 polymorphs. Physical Chemistry Chemical Physics, 9, 1587–1598. Mueller, K.T., Wu, Y., Chmelka, B.F., Stebbins, J., and Pines, A. (1990) High- ——— (2007c) First-principles calculations of solid-state 17O and 29Si NMR resolution oxygen-17 NMR of solid silicates. Journal of the American Chemical

spectra of Mg2SiO4 polymorphs (Correction). Physical Chemistry Chemical Society, 113, 32–38. Physics, 9, 5834. Oglesby, J.V. and Stebbins, J.F. (2000) 29Si CPMAS NMR investigations of silanol- Brown, G.E. (1970) The crystal chemistry of the olivines. Ph.D. thesis, Virginia group minerals and hydrous aluminosilicate glasses. American Mineralogist, STEBBINS ET AL.: NMR ON HIGH-PRESSURE Mg2SiO4 PHASES 915

85, 722–731. Physical Constants, p. 303–332. American Geophysical Union, Washington, Panero, W.R. (2008a) Cation disorder in ringwoodite and its effect on wave D.C. speeds in the Earth’s transition zone. Journal of Geophysical Research, 113, ——— (1997) Magnesium site exchange in forsterite: a direct measurement by B10204. high temperature 25Mg nuclear magnetic resonance spectroscopy. American ——— (2008b) First-principles determination of the structure and elasticity of Mineralogist, 81, 1315–1320. hydrous ringwoodite. American Geophysical Union, Fall Meeting, Abstract Stebbins, J.F. and Kanzaki, M. (1990) Local structure and chemical shifts for six- DI13C-03. coordinated silicon in high pressure mantle phases. Science, 251, 294–298. Phillips, B.L., Howell, D.A., Kirkpatrick, R.J., and Gasparik, T. (1992) Investiga- Stebbins, J.F., Du, L.-S., Kelsey, K., Kojitani, K., Akaogi, M., and Ono, S. (2006) 29 27 tion of cation order in MgSiO3-rich garnet using Si and Al MAS NMR Aluminum substitution in stishovite and MgSiO3 perovskite: High-resolution spectroscopy. American Mineralogist, 77, 704–712. 27Al NMR. American Mineralogist, 91, 337–343. Phillips, B.L., Burnley, P.C., Worminghaus, K., and Navrotsky, A. (1997) 29Si and Stebbins, J.F., Panero, W.R., Smyth, J.R., and Frost, D.J. (2009) Forsterite, 1 29 H NMR spectroscopy of high-pressure hydrous magnesium silicates. Physics wadsleyite, and ringwoodite (Mg2SiO4): Si NMR constraints on structural and Chemistry of Minerals, 24, 179–190. disorder and effects of paramagnetic impurity ions. American Mineralogist, Redfern, S.A.T., Harrison, R.J., and O’Neill, H.St.C. (1999) Thermodynamic 94, 626–629.

and kinetics of cation ordering in MgAl2O4 spinel up to 1600 °C from in situ Tse, D. and Hartmann, S.R. (1968) Nuclear spin-lattice relaxation via paramagnetic neutron diffraction. American Mineralogist, 84, 299–310. centers without spin diffusion. Physical Review Letters, 21, 511–514. Sasaki, S., Prewitt, C.T., Saito, Y., and Ito, E. (1982) Single-crystal X-ray study of Weiden, N. and Rager, H. (1985) The chemical shift of the 29Si nuclear magnetic

γ Mg2SiO4. Journal of Geophysical Research, 87, 7829–7832. resonance in a synthetic single crystal of Mg2SiO4. Zeitschrift für Naturforsc- Sen, S. and Stebbins, J.F. (1994) Phase separation, clustering and intermediate hungen, 40A, 126–130. 29 range order in Li2Si4O9 glass: A Si MAS NMR spin-lattice relaxation study. Welch, M.D., Pawley, A.R., Ashbrook, S.E., Mason, H.E., and Phillips, B.L. (2006) Physical Review, 50, 822–830. Si vacancies in the 10-A phase. American Mineralogist, 91, 1707–1710. Smith, K.A., Kirkpatrick, R.J., Oldfield, E., and Henderson, D.M. (1983) High- Wood, B.J., Kirkpatrick, R.J., and Montez, B. (1986) Order-disorder phenomena

resolution silicon-29 nuclear magnetic resonance spectroscopic study of in MgAl2O4 spinel. American Mineralogist, 71, 999–1006. rock-forming silicates. American Mineralogist, 68, 1206–1215. Xue, X., Kanzaki, M., and Shatskiy, A. (2008) Dense hydrous magnesium silicates,

Smyth, J.R. (1994) A crystallographic model for hydrous wadsleyite (β-Mg2SiO4): phase D, and superhydrous B: New structural constraints from one- and two- An ocean in the Earth’s interior? American Mineralogist, 79, 1021–1025. dimensional 29Si and 1H NMR. American Mineralogist, 93, 1099–1111. Smyth, J.R., Kawamoto, T., Jacobsen, S.D., Swope, R.J., Hervig, R.L., and Holloway, J.R. (1997) Crystal structure of monoclinic hydrous wadsleyite

[β-(Mg,Fe)2SiO4]. American Mineralogist, 82, 270–275. Smyth, J.R., Holl, C.M., Frost, D.J., Jacobsen, S.D., Langenhorst, F., and McCam- mon, C.A. (2003) Structural systematics of hydrous ringwoodite and water in the Earth’s interior. American Mineralogist, 88, 1402–1407. MANUSCRIPT RECEIVED DECEMBER 26, 2008 Stebbins, J.F. (1995) Nuclear magnetic resonance spectroscopy of silicates and MANUSCRIPT ACCEPTED MARCH 16, 2009 oxides in geochemistry and geophysics. In T.J. Ahrens, Eds., Handbook of MANUSCRIPT HANDLED BY JOSHUA FEINBERG