American Mineralogist, Volume 79, pages 848-861, 1994

The structural role of Mg in silicate liquids: A high-temperaturettMg, 23Na, and 2esi NMR study

Pntrn S. Frsxrr* JoNlrrHx F. SrrnnrxsDepartment of Geological and Environmental Sciences, Stanford University, Stanford, Califomia 94305-2115, U.S.A'

AssrRAcr

We have studied the structural role of Mg in silicate liquids as a function of temperatureand composition using high-temperature 25Mg, '3Na, and 'zesi NMR. To establish therelationship between the isotropic chemical shift (6,..) and the coordination number for25Mg in silicates, we have obtained high-resolution 25MB MAS NMR spectra on isotopi-

cally enriched akermanite (tolMg) and diopside (tulMg) using spin-echo pulse sequences andabsolute-value Fourier-transform techniques. The 6,* for akermanite is 49 + 3 ppm, andthat for diopside is 8 + 3 ppm, demonstrating that the chemical-shift range for "Mg inthe silicates is similar to that established in oxides.

The 0,* for 25Mg in (CaO)o.rr(MgO)o,o(SiOr).,, liquid at 1400 "C is 8 ppm, suggestingthat Mg resides in a structural environment similar to that in diopside. The d,* for 25Mg

in (NarO)0.,r(MgOL,8(SiO2)050liquid shifts from 34 ppm at l l50'C to 29 ppm at 1360

"C, indicating smaller coordination numbers than in the Ca liquid but a relatively largeincrease in the Mg-O bond length or coordination number (or both) with increasing tem-perature. The 6,* for 23Na shows similar, but less pronounced, changes, whereas the 2eSi

6,"o is unchanged over this range of temperatures. This suggests that Mg undergoes thelargest increases in bond distance or CN (or both) ofany ofthe cations in the liquid.

Spin-lattice relaxation times for 2sMg and 23Na were measured in liquid (NarQ)'"-(MgO[,8(SiO2)0 5o over the same range of temperatures. The apparent activation energy ofthe ,sMg relaxation (related to Mg diffirsional motion) is I l9 kJ/mol, intermediate betweenthat of Na and Si. The apparent activation energy for 23Na relaxation is 85 kJ/mol, whichis higher than that observed in sodium silicate liquids of similar polymerization but com-parable to that in mixed alkali silicate liquids, suggesting that Na and Mg may show amixed-cation effect.

IrvrnooucrroxThe structure and dynamical properties of silicate liq-

uids exert a fundamental control on igneous processessuch as mineral-melt equilibrium, elemental partitioning,and transport properties, including diftrsion, viscosity,and density. Although much is known about silicate min-erals, relatively little is known about the liquids fromwhich these minerals form. Much of our existing knowl-edge about silicate liquids is based on inferences madefrom studies of macroscopic properties such as thosementioned above, but these inferences can be nonunique.Thus, direct studies of the structure of silicate liquids athigh temperature hold great promise for elucidating thephysics and chemistry of natural magmas.

Most information about the structure of silicate liquidshas come from the study of silicate glasses, the structuresofwhich represent that ofthe liquid at the glass-transitiontemperature. There are obvious limitations to this ap-

* Present address: Institute ofGeophysics and Planetary Phys-ics, Lawrence Livermore National Iaboratory, L-413, Liver-more. California 94550. U.S.A.

proach. First, the kinetics of structural relaxation in sili-cate liquids makes it difficult to quench glasses with fic-tive temperatures differing by > 100 'C or so (Brandrissand Stebbins, 1988; Dingwell and Webb, 1990). This rangein temperature is often too small to discern temperature-dependent changes in liquid structure. Furthermore, low-silica liquids cannot be easily quenched to glasses. And,by definition, the dynamical phenomena that control thetransport properties in silicate liquids are not observablebelow the glass-transition temperature. To link the largebody of data from glasses to those data measured at hightemperature (and to magmatic conditions), it is criticalto understand how the structure of silicate liquids changeswith temperature.

Recently, several spectroscopic techniques have beenadapted to study the structure of silicate liquids at hightemperature. They include X-ray techniques (XANES,EXAFS: Waychunas et al., 1988; Jackson et al., 1993;Brown et al., 1993), vibrational spectroscopies (McMillanet al., 1992; Mysen and Frantz, 1992), and nuclear mag-netic resonance (NMR) spectroscopy (Stebbins, 1988,l99l; coutures et al., 1990; cot6 etal.,1992). They have

0003-004x/94l09 I 0-0848$02.00 848

FISKE AND STEBBINS: NMR STUDY OF Mg IN SILICATE LIQUIDS 849

allowed for the investigation of the structure of silicateliquids over a wider range of temperatures and at tem-peratures of relevance to magmatic processes.

Mg is a major constituent in most magmas, and, inprimitive, undifferentiated magmas, it can be the mostabundant oxide after silica. Mg also represents an im-portant group ofcations, such as Ca2+,Fez+, Co2+, andNi2+, whose size and charge are intermediate betweenthose of network-forming cations such as Si4+, Alr+, andFe3+ and the charge-balancing, large, monovalent cationssuch as the alkalis.

There have been a number ofspectroscopic and theo-retical investigations of the structure of Mg-bearing sili-cate glasses and liquids. They include X-ray scattering(Waseda and Toguri, 1977 Yin et al., 1983), X-ray emis-sion and molar refractivity (Hanada et al., 1988), vibra-tional spectroscopies (McMillan, 1984; Williams et al.,1989; Cooney and Sharma, 1990, Kubicki et al., 1992),and molecular dynamics (MD) simulations (Matsui et al.,1982; Dempsey et al., 1984; Kubicki and Lasaga, l99l).Most of these studies conclude that Mgz+ resides in adistorted site; however, disagreement exists regarding theactual coordination number for Mg2+.

The possibility of multiple coordination environmentsfor Mgz+ in silicate liquids and glasses is srggested byEXAFS studies of divalent cations of similar radius. Fe2*and Ni,* EXAFS studies show dominantly tetrahedralcoordination of these cations in a number of silicate liq-uids and glasses studied (Waychunas et al., 1988; Jack-son, I 99 l; Galoisy and Calas, 1991, 1992, 1993a, 1993bBrown et al., I 993; Farges et al., 1993, I 994). Those stud-ies suggested that Mg2+, being intermediate in size be-tween Fe2+ and Nir+, may also assume a tetrahedral co-ordination in silicate liquids.

The high heat capacities and thermal expansivities ofsilicate liquids require that they undergo continuousstructural changes with temperature, which may includechanges in the coordination of cations (e.g., Richet andNeuville, 1992). Recent heat capacity measurements ofmagnesium aluminosilicate liquids have demonstrated asimple linear relationship between configurational heatcapacity and Si/(Mg 1 Al) over an extremely wide rangeof composition, implying that the structural changes withtemperature are mainly due to short-range O to cationinteractions (Courtial and Richet, 1993). The high partialmolar heat capacity of MgO in silicate liquids indicatesthat Mg'?+ may undergo particularly large configurationalchanges in silicate liquids as a function of temperature(Stebbins et al., 1984i Richet and Neuville. 1992). Fur-thermore, studies of glasses and liquids in the systemsdiopside-albite and diopside-anorthite have documentedexcess enthalpies of mixing in the liquids that are sub-stantially more positive than in the glasses (Stebbins etal., 19841, Navrotsky et al., 1989). This excess enthalpyprobably reflects a greater number of structural configu-rations in these liquids (Stebbins et al., 1984) and maybe due to mixing of Mg on tetrahedral, as well as octa-hedral, sites.

This study represents the first application of ,sMg NMRto silicate liquids at high temperature, with the goal ofdetermining the structural environment of Mg as a func-tion of temperature and composition. Because so little isknown about the NMR chemical shifts for 25Mg in solids,it was also necessary to study several model compoundswith known structures. Our success in observing NMRsignals in silicate liquids at high temperature suggests thatNMR studies of other nuclides that are difficult to ob-serve such as 47'4eTi and 43Ca may also be possible.

Expnnrn rnnr.lr, pRocEDuREs

The samples used in this study were synthesized usingreagent-grade oxides and carbonates. The ,5Mg isotopi-cally enriched samples were made using 990/o 25Mg-en-riched MgO powder (Oak Ridge National Laboratory).The crystalline compounds were synthesized with 0.2 uto/oCo'z* added to speed spin-lattice relaxation. Samples weresynthesized at high temperature in air using Pt or PtAucrucibles. Careful weight-loss measurements confirmedthat alkali loss was insignificant and that desired stoichi-ometries were reached.

Crystalline akermanite (CarMgSirOr)

The reagent mixture was heated to 1300 "C at 200 "C/h, held for 12 h, quenched, reglound, and annealed at1300 "C for 20 h. XRD spectra for this sample showedonly peaks for akermanite. The powders were deep bluedue to the I4lCo2+ dopant (e.g., Rossman, 1988; Keppler,1992).The,,Si MAS NMR spectra of this sample showeda single asymmetric peak centered at -76.4 ppm, repre-senting the Si site in akermanite (Janes and Oldfield,1985), and a small peak centered at -84.4 ppm, probablyrepresenting diopside (<2o/oby area) (Smith et al., 1983).

Crystalline diopside (CaMgSi,Ou)

The reagent mixture was heated to 1300 "C at 200 .C/

h, held for 12 h, quenched, reground, and annealed at1300 lC for 24 h. XRD spectra of this sample showed itto contain only diopside. The powder was light pink, con-sistent with the t61co2+ dopant (e.g., Rossman, 1988). The"Si MAS NMR spectra of this sample showed a singlesymmetric peak centered at -84.4 ppm, consistent withprevious data (Smith et al., 1983). The widrh ofthis peak,3.4 ppm, is considerably broader than that observed formost crystalline silicates, suggesting that this sample maycontain significant structural disorder.

Magnesium silicate glasses containing Na, It and CaGlasses of the compositions detailed in Table I were

all prepared with 25Mg isotopically enriched MgO pow-der. No paramagnetic dopant was used for these materi-als. The reagent mixtures were slowly heated to temper-atures above the decarbonation point for the respectivealkali or calcium carbonates, held for 4 h, then heated at300 "C/h to 50 "C above their respective melting points,held for 30 min, and then quenched by placing the bot-tom of the crucible in a shallow dish of HrO. Inspection

850

TABLE 1. Compositions of silicate liquids in this study

FISKE AND STEBBINS: NMR STUDY OF Mg IN SILICATE LIQUIDS

Composition Name NBO/T Liquidus f Reference

(Na,O)0,(MgO)0,(SiO,)otu(Na,O).,"(MgO)o,s(SiO,)o 54(K,O)o ,e(MgO)o ,4(SiOr)o 5?(Cao)o *(MgO)o ,4(SiO,)o 5,

NaMglNaMg2KMglCaMgl

1 . 31 .731 . 51 .5

9 1 5 + 5 r c840 + 10'C

1100 + 30 'c1325 + 10qC

Schairer, 1957Schairer, 1957Roedder. 1951Schairer and Bowen, 1942

of the products under high magnification and with crossedpolars showed the glasses to have no detectable crystalsor unreacted material. After the high-f NMR experi-ment, the (NarO)o,r(MgO)o ,'(SiOr)o ro glass (from now onreferred to as NaMg2) was reground and doped with 0.2wt0/0 Co2+ and revitrified. The resulting glass appeareddark blue, indicating that the Co2* resided at least par-tially in tetrahedral coordination.

NMR spectral acquisition procedures

The 'z5Mg MAS NMR spectra were acquired using aVarian VXR-400S NMR spectrometer operating at aLarmor frequency of 24.512 MHz for 'zsMg with a DotyScientific probe. The 25Mg chemical shifts were calibratedagainst an external l-MMg(NO.), aqueous solution. Theliquid 90' pulse length measured on this standard wasabout 30 ps. Samples were contained in a 5-mm diametersapphire rotor and spun at about 9500 Hz. The spectralwidth was 100 kHz, and the digital resolution was 6 Hz.Probe ringing required a dead time of about 85 ps priorto acqursrtron.

The '?eSi MAS NMR spectra of the crystalline com-pounds and the NaMg2 glass doped with 0.2 wt0/0 Co2+were acquired at a Larmor frequency of 79.459 MHz us-ing a Varian MAS probe with a ZrO, rotor spinning at5500 Hz. The '?eSi chemical shifts were referenced to anexternal standard (tetramethyl silane) contained in anidentical rotor. The average isotropic chemical shift forthe '?eSi in the glass was determined by measuring theintegral ofthe central transition and the first set ofside-bands (the sidebands represented about l0o/o ofthe totalsignal), determining the center of gravity of each, andthen taking the weighted average over the entire spec-trum. Several repetitions of this procedure yielded valueswith a range of +0.5 ppm.

The'zsMg NMR spectra are difficult to obtain becauseofthe low natural abundance (10. I 60lo) and low resonancefrequency of 25Mg, which greatly reduces the sensitivity.In addition, the second-order quadrupolar broadening ofthe central transition is roughly nine times greater thanfor 27Al in an identical site (Dupree and Smith, 1988).Good results for both akermanite and diopside were ob-tained on 'z5Mg isotopically enriched samples using a sin-gle pulse with an rf tip angle of about 20-30", a delay of0. I s, and roughly 100 000 sigrral averages. However, spin-echo techniques (see next section) yielded somewhat im-proved results for the enriched materials and were usedto determine the isotropic chemical shifts.

The NMR values, 6,- (isotropic chemical shift), Co (thequadrupolar coupling constant, equal to e2qQ/h), and q

(the asymmetry parameter), for 25Mg in the crystallinematerials were determined in two ways. The quadrupolarcoupling constant can be determined by measuring thedifference between the centers ofgravity ofthe central Yz-Yz transition and the outer +t/z-+3/z and !t/r-+y, transi-tions (Lippmaa eI al., 1986; see Hovis et al., 1992, for anexample), and 6,* can be estimated. Alternatively, thespectra can be simulated directly, yielding values for 6,"",C", and 4. The simulation algorithm used for this studyis based on the procedure of MiiLller (1982). The best fitof the spectra was achieved by visual inspection to max-imize the positions of discontinuities in the line shapesrather than least-squares minimization of the entire spec-trum because these features are less affected by distor-tions of the spectra. The estimated uncertainty is alsobased on the visual comparison of the fit with a range ofvalues for each NMR parameter.

Application of Hahn spin-echo pulse sequence andabsolute-value spectra

For broad NMR peaks, it is difficult to acquire undis-torted spectra using single-pulse techniques. That is becausethe first part of the FID (free induction decay) includes theinformation about the broadest components to the NMRline shape. This information can be lost in the time betweenapplyrng the excitation pulse and turning on the receiver toacquire the FID. In the case of low-frequency nuclides suchas 2sMg, this problem can be worse because probe ringingcan severely distort the first paft of the FID, necessitatingfurther truncation of the data.

More complex pulse sequences can be used to reduce,or even eliminate, these problems. The Hahn spin-echopulse sequence (90rr-180') can be particularly useful. The90" pulse tips the magnetization of the sample into thex-y plane, and after a time r, the 180'pulse is applied.The 180' pulse reorients the magrretization so that, in-stead ofdrifting out ofphase, the individual spins rephaseafter a time 2r. This rephasing produces an echo in theFID, with the top of the echo representing the point atwhich the spins fall completely in phase. Thus, a Fouriertransform ofthe FID starting at the top ofthe echo yieldsa complete NMR signal, free from distortions due totruncations of the FID and due to probe ringing.

Recently, Grandinetti et al. (1993) have developed animprovement to this technique. Rather than truncatingthe FID at the top ofthe echo, they used the entire echoin the Fourier transform. The resulting spectrum can thusbe displayed in absolute-value mode, where each pointof the displayed spectrum is calculated as the square rootof the sum of the squares of the real and imaginary com-

FISKE AND STEBBINS: NMR STUDY OF Mg IN SILICATE LIQUIDS 8 5 1

ponents of each complex data point. This results in acomplete elimination of the baseJine rolls because ofphasing problems and increases the signal to noise of thespectrum by roughly a factor of /2 (Grandinetti et al.,1993). In addition, the spinning sidebands are greatly di-minished. A comparison of the spectra for diopside ac-quired with a single pulse vs. a Hahn spin echo withabsolute-value full-echo transform is presented in Figurel. The best-fit simulations of the central transitions forthe two spectra yielded somewhat different values for theisotropic chemical shift and indicated that neither is com-pletely free of distortion. Both kinds of spectrum showpeaks and shoulders at the same frequencies, but the Hahnspin-echo spectra generally have much narrower peakwidths than the single-pulse spectra. Although the signalto noise ratios for the Hahn-echo spectra are improved,they are lower in absolute intensity by roughly a factor of2. That may be because of the removal of the paramag-netically broadened component of the NMR sigrral, whichmay not refocus during the echo and which can distortthe single-pulse spectra.

25Mg and 23Na relaxation times

For quadrupolar nuclei, spin-lattice relaxation is usu-ally caused by fluctuations in the electric field gradientaround the nucleus. In liquids, especially silicate liquidsat high temperature, these fluctuations are often causedby the diffusional motion of the resonating atom (e.g.,Liu et al., 1987, 1988). Relaxation is most efficient, and?", is shortest, when these motions are at the Larmor fre-quency. When molecular motion is either much sloweror much faster than the Larmor frequency, relaxation isless efficient and T, is longer. In those regions away fromthe Z, minimum, the rate of change of the relaxation timewith temperature can yield an apparent activation energythat may characteize the motion responsible for relaxa-tion (reviewed by Stebbins, 1988).

At temperatures well above the ?n, minimum, the widthof the NMR peak is generally controlled by the spin-Iattice relaxation time where Z, : /,(r") and Zr : Tr. T,is the spin-spin relaxation time, and zu, is the width of thepeak at halfheight in hertz. To confirm that that condi-tion held for our experiments, the spin-lattice relaxationtime was measured directly at 1263 oC using the inver-sion recovery technique (e.g., Fukushima and Roeder,l98l). A nonlinear least-squares regression ofthe peakintensity vs. delay time was used to calculate the relaxa-tion time I,. The measured Z, matched that predictedfrom the line width to within 5ol0.

2sMg crrrcurcAt, SHTFTS AND srrE TNFoRMATToN rNCRYSTALLINE SOLIDS

Few data have been published on 25Mg chemical shiftsin solids (e.g., MacKenzie and Meinhold, 1994a,1994b).The '?sMg data show the same trend to increasing chem-ical shielding with increasing coordination number asother cations such as Si, Al, and Na. The range in chem-ical shift betweenrorMg (in MgAlrOo) and r6lMg (in MgO)

100 50 0 -50 -100 -150 -200 -250 ppm

Fig. I . The ,5Mg MAS NMR single-pulse and Hahn-echo spectrafor crystalline diopside do@ with 0.2 wP/o Co,+. The Hahn-echospectrum is substantially nzurower than the single-pulse spectrumand yields an isotropic chemical shift of 8 + 4 ppm, whereas thesingle-pulse spectrum yields a valw of 22 + 5 ppm.

is 26 ppm (Dupree and Smith, 1988), compared withapproximately 100 ppm for'zeSi in fourfold and sixfoldcoordination, 60 ppm for 27Al in fourfold and sixfold co-ordination (Engelhardt and Michel, 1987), and 20 ppmfor 23Na in sixfold and eightfold coordination (Xue andStebbins, 1993). However, second-nearest-neighbor ef-fects on the isotropic chemical shift may be larger forcations with a lower charge. In the case of 23Na, the degreeof polymerization has nearly as large an effect on di* asthe coordination number (Xue and Stebbins, 1993).Therefore, the limited amount of 'z5MB chemical shiftinformation available from oxides might not be easilyapplied to silicates because ofdifferences in second-near-est-neighbor populations. Thus, it was necessary to studyMg-bearing silicate minerals with different Mg coordi-nations and compare their isotropic chemical shifts withthose known for the oxides.

There are several constraints limiting the choice ofmodel compounds for the calibration of 'sMg NMRchemical shifts in silicates. The magnitude of the quad-rupolar coupling (C") for quadrupolar nuclides in solidsis proportional to the distortion of the atomic site (Ghoseand Tsang, 1973). Because ofthe substantial effect ofsec-ond-order quadrupolar broadening for 25Mg, we selectedminerals with the most regular Mg sites. Compositionplaces additional constraints: the minerals cannot containmore than a few percent of paramagnetic cations such asFe2+ because of problems with paramagnetic broadening,and the proportion of SiO, to the other cations must notdiffer too much from that of the liquids of this study. Theminerals akermanite (Ca,MgSirO,) and diopside (Ca-MgSirOu) were the best match to these criteria. The siteinformation for Mg in these two minerals, as well as inthe oxides, is summarized in Table 2.

The'z5Mg MAS NMR Hahn-echo spectrum for aker-

852 FISKE AND STEBBINS: NMR STUDY OF Mg IN SILICATE LIQUIDS

TABLE 2. Site information for Mg in the oxides and silicates studied

Mg-O bondMineral length (A)

Bond length O-Mg-Ovariation angle variation CN of Mg CN of O linked to Mg

lsotropicchemical

shift (ppm)

Periclase (MgO)Spinel (MgAl,O.)Diopside (CaMgSi,O6)Akermanite (CarMgSirO?)

2.1051.9242.O771 .915

00

17.45.8r.

00' lVo

0

6464

6 (all Mg)4 (3Al,Mg)3,4 (Mg,Ca,Si + Mg)4 (Mg,Si,Ca,Ca)

2552

8 r 34 9 + 4

Note.'site information is from Smyth and Bish (1988). CN : coordination number.

manite, along with a simulation of the central transition,is shown in Figure 2. The central peak consists ofa broaddoublet consistent with an asymmetry parameter of 0.2-0.3, a quadrupolar coupling constant of 2.7-2.9 MHz,and an isotropic chemical shift between 47 and 52 ppm.The extra intensity to higher frequency may represent asmall amount of distortion of the spectrum. Analysis ofthe centers of gravity of the spinning sidebands in thesingle-pulse spectrum suggests a quadrupolar couplingconstant of 2.9-3.0 MHz.

The'?5Mg MAS NMR Hahn-echo spectrum for diop-side, along with a simulation of the central transition, isshown in Figure 3. The central peak has a small, high-frequency shoulder. The central peak is best simulatedwith an asymmetry parameter of l-0.8, a quadrupolarcoupling constant of 2.0 + 0.05 MHz, and an isotropicchemical shift of 8 t 3 ppm. Analysis of the centers ofgravity ofthe spinning sidebands in the single-pulse spec-trum suggests a quadrupolar coupling constant of 2.2MHz.

In comparison, the single-pulse spectrum for diopside(Fig. l) has additional intensity at high and low frequencyand has a wider line width overall. The best-fit simulationof this spectrum has an asymmetry parameter of l-0.8,a quadrupolar coupling constant of 2.3 + 0.1 MHz, andan isotropic chemical shift of 22 + 3 ppm. The high-frequency shoulder is present in both spectra and mayrepresent the center of the +y2-+3/, transition, althoughit is centered slightly lower in frequency than that ex-pected from the sideband analysis. It is unlikely that thepeak represents Mg on the larger M2 site in diopside (oc-cupied by Ca), as this would be expected to produce apeak at lower frequency because ofthe longer bond lengthsand higher coordination number for this site.

The differences in chemical shift between the silicatesstudied and those known for the oxides can be explainedusing empirical trends established for other cations. Bothsilicates have a lower chemical shift (greater chemicalshielding) than the respective oxides with the same co-ordination number for Mg. This follows the trend foundfor 27Al and 2eSi, where increasing the number of Si next-nearest neighbors causes increased shielding (Engelhardtand Michel, 1987; Kirkpatrick, 1988). The much largerdifference in chemical shift between periclase and diop-side than between spinel and akermanite reflects the factthat the two minerals with I61Mg have larger differencesin their next-nearest-neighbor populations than do thetwo minerals with torMg. In the case of periclase, the O

atoms around the Mg are sixfold coordinated, whereas indiopside they are threefold and fourfold coordinated. Incontrast, the O atoms around Mg in akermanite and spi-nel are both fourfold coordinated.

Arguments such as these can also illustrate the diffi-culty in interpreting chemical shift data in terms of spe-cific structural parameters. For example, as stated earlier,increasing Mg coordination (which necessitates an in-crease in the average bond length) corresponds to an in-crease in shielding (decreased chemical shift). However,within each pair of minerals, those with shorter bonds(the silicates) actually have a lower chemical shift. Whereasthe current paucity of data prevents a more complex anal-ysis of chemical shift trends, 25Mg chemical shifts willlikely show the same complex relationship to bond length,bond angle, second-nearest-neighbor population, andother structural parameters as do other NMR nuclides.

In conclusion, the chemical shift trend for Mg in dif-ferent coordination environments in the two silicates doesnot differ greatly from the trend observed in the two ox-ides, and like "Al and 2eSi, coordination number exertsa primary control on 2sMg chemical shifts in solids. Thisenables us to make some limited structural inferences forMg in the silicate liquids on the basis of their observedisotropic chemical shifts.

Hrcn-rrcrrprRATr.IRE 2sMg NMR niSILICATE LIQI'IDS

Observation of a signal from 25Mg in silicate liquids athigh temperature that represents the average isotropicchemical shift is dependent on several related factors.First, the viscosity of the liquid must be sufficiently lowto allow for rapid isotropic reorientation ofthe local en-vironment around the resonating nuclide. This averagesout the entire spectrum for all the transitions (usuallyseveral megahertz in width) and results in a sharp, re-solvable NMR peak, the position of which represents theisotropic (true) chemical shift for the site. For network-forming cations such as Si and Al, the reorientation fre-quency can be crudely estimated from the viscosity usingthe Maxwell equation for viscoelastic behavior: r : n/G-(e.g., Dingwell and Webb, 1989), where r is the shearrelaxation time, 4 is the viscosity in pascal seconds, andG- is the infinite frequency shear modulus. However, forcations that are not strongly bound to the silicate net-work, such as the alkalis, hopping frequencies are notcorrelated with viscosity but must be estimated using datafor tracer diffusivities or other transport properties (e.g.,

FISKE AND STEBBINS: NMR STUDY OF Mg IN SILICATE LIQUIDS

0 -r00 -2O0 ppm

853

+ 100

Fig 2. The xMg MAS NMR Hahn-echo spectrum for akerman-ite. The cental Vz-Vz transition is best fit with a 6* of 49 + 3 ppm,a Co of 2.8 + 0. I MHz, an asymmetry parameter of 0.2-0.3, anda line broadening of 900 Hz.

Liu et al., 1987, 1988). In the case of Mg, for which tracerdiffusion data are lacking, the Maxwell equation providesan initial estimate that is likely to be a lower bound only.For quadrupolar nuclides such as 25Mg, the static NMRpowder pattern for all the 2.I transitions (where 1 is thenuclear spin, in this case t/r) can span several megahertz.Full averaging is achieved at hopping frequencies on theorder of tens of megahertz, which, using the Maxwellequation, correspond to viscosities of < 10, Pa.s.

Another constraint is the relaxation time (?",) for ,5Mg

in the liquid. For an NMR signal to be observable, Z,must fall into a certain range. Relaxation times must beshort enough to allow efficient data collection; the lowsensitivity for the nuclide requires a minimum ofthousands of signal averages. However, if the i", is tooshort, the line width increases, reducing the ratio ofsignalto noise. If Z, falls below tens of microseconds, the signalmay be lost entirely in the instrumental dead time.

Rrsur.rs FoR LreurDs

The composition of the calcium magnesium silicateliquid [composition (CaO)o rn(Mgo)o,o(SiO,)' s,, hereafterreferred to as CaMgll chosen for this study (Table l) fallsalong the eutectic between tridymite and diopside, withan estimated melting temperature of 1325 "C (Schairerand Bowen, 1942).The '?sMg NMR signals were observedat 1350 and 1400 qC (Fig. 4). The peaks were fittedto singleI-orentzian line shapes, which yielded peak positions of 5+ 15 and 8 a 3 ppm, respectively. The decrease in the linewidth with increase in temperature indicates that the 'z5Mg

i1 in this liquid is increasing with temperature and thatthese temperatures lie above that of the Z, minimum.

Two compositions in the system NarO-MgO-SiO, werestudied at high temperature, along with a single compo-sition in the system KrO-MgO-SiO, (Table l). No signalwas observable in the (NarO)or(MgO)or(SiOr)o.o liquid(hereafter referred to as NaMgl ) up to I 3 l0 "C or in the(KrO)o rn(MgO)0.r4(SiO2)0.57 (hereafter referred to as KMg I )up to 1290 "C, possibly because of the extremely short ?n,for'5Mg in these liquids and correspondingly broad peaks.

However, the NaMg2 liquid yielded observable NMR

+50 0 -50 - r00 -150 ppm

Fig. 3. The rMg MAS NMR Hahn-echo spectrum for diopside.The central Yz-% transition is best fit with a d* of 8 + 3 ppm, a C"of 2.0 + 0.05 MHz, an asymmetry parameter of 0.9-1.0, and a Iinebroadening of 600 tlz.

signals from lll9 to 1368 .C (Fig. 5). The peaks werefitted to single Lorentzian lines to determine the peakposition and line width at each temperature. The peakpositions are plotted in Figure 6. The narrowing of theNMR peak with increasing temperature indicates that,like the calcium silicate liquid, the temperatures at whichthese spectra were obtained lie above the 1", minimumfor 25Mg in the liquid. To confirm that the line widthsare actually controlled by the I,, the latter was measureddirectly at two temperatures using the techniques de-scribed above. The 7", measured matched those estimatedfrom the line widths to within 50/0. The derived activationenergy for the mechanism of 2sMg relaxation (l 19 + 8kJlmol) was determined by calculating the least-squaresfit to the slope of the relaxation time data vs. temperatureon an Arrhenius plot (Fig. 7).

Both 'zeSi NMR and'z3Na NMR spectra were also ob-tained for this sample over the same range in tempera-ture. As with the 25Mg results, the 'zeSi and 23Na peakswere fitted with single Lorentzian line shapes, and thepeakpositionandlinewidthsweredetermined. The isotropicchemical shifts vs. temperature for'zeSi and 23Na are plot-ted in Figure 8. The 23Na relaxation times were derivedin the same manner as for 25Mg and are plotted on anArrhenius plot as well (Fig. 9). The derived activationenergy for the 23Na relaxation is 85 + l0 kJ/mol.

No 25Mg NMR signal was observable from the dopedNaMg2 glass using single-pulse MAS techniques at roomtemperature, probably because of severe quadrupolar orparamagnetic broadening. However, using the Hahn-echosequence, a weak broad peak (centered at -20 ppm, and200 ppm wide) was detected after 10000 signal averages.Its low intensity cleady showed that only a fraction ofthe 25Mg signal was detected.

DrscussroxPrevious spectroscopic and molecular dynamics studies ofmagnesium silicate liquids and glasses

Previous theoretical and spectroscopic studies in Mg insilicate [quids have principally focused on MgSiO, andMg,SiOo glasses and liquids (Table 3). These studies (see

854 FISKE AND STEBBINS: NMR STUDY OF Mg IN SILICATE LIQUIDS

600 400 200 0 -200 -4OO ppm

Fig 4. The 5Mg NMR spectnun for CaMgl liquid at 1400 rC.The single peak is fitted with a I-orentzian line shape at 8 + 3 ppm.

also Navrotsky et al., 1985) suggest that the local structuralenvhonment around Mg is different in these two liquids.

MgSiOr. X-ray scattering studies (Waseda and Toguri,1977; Waseda, 1980) determined the mean Mg-O bondlength to be 2.14 A in enstatite glass and 2.16 A in theliquid at 1700 "C and reported a mean coordination num-ber of 4.5. However, Yin et al. (1983) proposed the meancoordination number of Mg to be 6, on the basis of fittingtheir X-ray scattering data to a quasi-crystalline modelfor the glass. Early molecular dynamics (MD) simulationsby Matsui et al. (1982) yielded simulated RDF patternsthat were similar to those observed in the previous X-rayscattering studies and reported a mean Mg-O bond lengthof 2.1 A. However, more recent MD simulations by Ku-bicki and Lasaga (1991) disagree with the interpretationsof Yin et al. (1983) and suggest a mean Mg-O bond dis-tance of 1.9 A, corresponding to a mean coordinationnumber for Mg close to 4. However, it should be notedthat both MD simulations correspond to extremely highfictive temperatures (roughly 5000 and 2000-5000 K, re-spectively) and may not be directly comparable withspectroscopic studies of glasses (with fictive temperaturesof roughly 1000 K). A relatively low abundance of t61Mg

in enstatite glass is further implied by Raman spectra ofglasses in the system MgSiO.-CaSiOr, which show a pro-gressive loss of a low-frequency peak associated with t6lca

or r6tMg with increasing Mg content (Kubicki et al.,1992).Finally, X-ray emission spectra and molar refractivitymeasurements ofrf-sputtered amorphous films in the sys-tem MgO-SiO, indicate predominantly tetrahedral coor-dination of Mg when the MgO content is 500/o or less(Hanada et al., 1988).

MgrSiOn. The local enwironment of Mgz+ in forsteriteglass seems to be significantly different from that in en-

500 400 300 200 100 0 -100 -300 PPm

Fig. 5. The '?5Mg NMR spectra for NaMg2 liquid fromI I 19 to 1 368 'C. The narrowing of the peak with increasingtemperature indicates that these temperatures are above thatfor the Z, minimum.

statite glass and liquid. MD simulations (Matsui et al.,1982) proposed higlrer coordination numbers and longerMg-O bond distances in forsterite liquids when comparedwith enstatite liquids. In addition, vibrational spectro-scopic studies (Williams et al., 1989; Cooney and Shar-ma, 1990) show that, unlike in the enstatite-wollastonitesystem, Ca and Mg substitute freely in the orthosilicatesystem, suggesting that Mg'z+ is principally octahedrallycoordinated, whereas Fe2+, which is dominantly teta-hedrally coordinated in these glasses (Jackson et al., I 993),does not substitute freely for Mg2*.

2sMg chemical shifts and coordination number

As outlined earlier, it is difficult to make unique struc-tural interpretations from NMR chemical shift data be-cause chemical shielding is a complex function of manystructural parameters such as bond length, bond angle,and coordination number, which are themselves inter-related. The trend to lower chemical shift (increasedshielding) with increased coordination number is well es-tablished for 2eSi, 27A1, 23Na, and the limited set of datafrom this study, 25Mg. This is commonly ascribed to anincrease in the ionicity of the M-O bond with increasingbond length, which increases the positive charge on thecation and thus the overall chemical shielding (Engel-hardt and Michel, 1987). Calculations of the variation ofchemical shielding with structural parameters for simplesilicate molecules confirm that increased Si-O bond dis-

FISKE AND STEBBINS: NMR STUDY OF Mg IN SILICATE LIQUIDS

-101 100 1200 1300 1400

855

37 -7

9 -8cl

F\t r - 9

l rs'- 33.2

;31

K29

27

Temperature in "CFig. 6. The 25Mg chemical shifts for NaMg2 liquid as a func-

tion of temperature.

tances correspond to increased chemical shielding (Tos-sell and l-azzerelti, 1987; Lindsay and Tossell, l99l).

The'?5Mg chemical shift trends established from the oxideand silicate minerals summarized in Table 2 canbe used toestimate the mean coordination number (and with less cer-tainty, the mean Mg-O bond length) for Mg in the silicateliquids studied. The chemical shift for Mg in the calciummagnesium silicate liquid at 1400 lC is identical to that forMg in diopside within the uncertainty of the measurement.This strongly implies that the Mg in this liquid at this tem-perature has a mean coordination of 6, with a mean Mg-Obond distance similar to that in diopside.

The alkali silicate liquid, however, has mean 25Mg

chemical shifts that are intermediate between sixfold andfourfold coordinations. This implies that the Mg in thisliquid has an average coordination ofabout 5. Althoughwe did not study any minerals (e.g., grandidierite, yoder-ite) with t5rMg to calibrate the position of pentahedrallycoordinated Mg, the chemical shifts for ptSi and I5lAl fallabout midway between those for the more common four-fold- and sixfold-coordinated species (Stebbins and Mc-Millan, 1993; Risbud et al., 1987). However, our resultscannot distinguish between a dynamical average of r4lMgand t6lMg vs. t5lMg or a mixture of all three.

To compare the mean Mg-O bond distances in the sil-icate liquids of this study with those derived in previousstudies, the chemical shift data summarized in Table 2

6.5e-4 7 .0e-4 7 .5e-41/T in K

Fig. 7. The '?5Mg relaxation time data vs. temperature. AnArrhenius fit to these data gives an activation energy of I 19 +12 kJ/mol.

can also be used to estimate the correlation between 25Mg

chemical shifts and Mg-O bond distances. Clearly, this isan oversimplification and is difficult to constrain with thepaucity of data from Mg-bearing minerals. Nevertheless,it does provide a means of comparing these results withprevious work.

Using the data in Table 2, the 25Mg chemical shift in thecalcium magnesium silicate liquid correlates to a Mg-O bonddistance of 2.08 A at 1400 C. The data for the alkali silicateliquid, when extrapolated to the same temperature, coffe-sponds to a bond distance of about 2.00 A.

We attribute the difference in bond length primarily tothe replacement of Na for Ca, since the ratio of nonbridg-ing O atoms to tetrahedral sites is similar for the twoliquids. The trend to higher coordination numbers for Mgin the alkaline-earth silicate liquid vs. the alkali silicateliquid is similar to that observed for Ni2* in glasses byEXAFS (Galoisy and Calas, 1992) and, Ti4* (Lange andCarmichael, 1987, 1990; Dingwell, 1992; Paris et al.,1993). This trend has been explained by the decrease innet charge on the coordinating O atom with increasingcovalent bonding caused by the increased charge densityof the modifying cation. The lowering of the charge onthe O atom reduces the effect ofanion repulsion and al-lows for larger coordinations. Stebbins and McMillan(1993) suggested that this effect may contribute to the

Trau 3. Summary of previous studies of Mg2+ in glasses and liquids in the system MgO-SiO,

Ea = 119 !12 kJ/mol

Source Technique Mg-O bond length

Waseda, 1980Waseda, 1980Yin et al., 1983Hanada et al., 1988

Matsui et al., 1982Kubicki and Lasaga, 1991

This studyThis study

MgSi03 glassMgSi03 liquidMgSiOs glassMgO-SiO, films

MgSiOo liquidMgSi03 glass

NaMg2 liquidCaMg2 liquid

X-ray scatteringX-ray scatteringX-ray scatteringX-ray emission and

molar refractivityMD simulationMD simulation

'5Mg NMR,sMg NMR

2.14 A2.16 A2.08 Anot given (CN : 4)

2.1 A (T > 5000 K?)1 . 9 4 ( C N = 4 )(r:2000-5000 K)est. 2.00 A (1400qC)esr. 2.08 A (1400 cc)

+ +++++

856 FISKT AND STEBBINS: NMR STUDY OF Mg IN SILICATE LIQUIDS

B o.s

3 6.0

d

-82.0

g -82.5

a -83.0'j -sr.st O.- -84.0a6' -84.5

7.0

5.01000 I 100 1200 1300

Temperature in oC

+t + ++

b

-4.8

-5.0

e -s.2

i -s.aF\

I -s.6

-5.8

-6.0

Ea = 85 + 10 kJ/mol

6 .5e -4 7 .0e -4 7 .5e '4 8 .0e -4

1/T in KFig. 9. The'?3Na NMR relaxation time data Ys. temperature.

An Arrhenius fit of these data yields an activation energy of 85+ l0 kJ/mol.

magnetic shift similar to that observed in MgO (Fiske etal.. 1994) is subtracted.

In comparison with 2sMg the chemical shifts for "Siand 23Na in the same liquid show much smaller changes:'eSi chemical shifts remain constant, and nNa chemicalshifts drop by about I ppm over the range oftemperaturestudied. Recent high-temperature NMR studies of 'nsi,27A1, and 23Na in alkali aluminosilicate liquids (Stebbinsand Farnan, 1992) showed similarly small, but measur-able, changes in the isotropic chemical shift for all thesecations as a function of temperature.

Second-nearest-neighbor effects cannot be the domi-nant cause ofthese changes because a large change in thesecond-neighbor distribution around Mg should be evi-dent as a temperature effect on 6,* for "Si. Strong, three-dimensional clustering of Mg cations would lead to anincrease in the average number of nonbridging O atomsper Si site (NBO/SD. We estimate that if the observedchange in 6,* for 25Mg in the NaMg2 liquid is caused bya decrease in Mg clustering with increased Z, the averageNBO/Si would have to increase by about 0.2 over 300.C. This should produce an easily observable change of 2ppm in 6,.o for 2eSi. The lack of any detectable tempera-ture effect for "Si thus suggests that local structural changes(coordination number and bond length) have the largesteffect on the '?5Mg temperature dependence. Thus, thedecrease in chemical shift with temperature most likelyreflects changes in the local structure around Mg. If thistemperature dependence persists down to reasonable glasstransition temperatures for this composition (-600'C),it may imply substantially smaller mean coordinationnumbers (roughly 4.5) and smaller Mg-O bond distances(1.95 A) in the glass.

To compare the magnitude of the temperature-depen-dent structural changes for Mg with those observed forSi, Na, and Al, the measured chemical shift changes mustbe scaled relative to each other. Of all the structural pa-rameters that can be correlated to chemical shifts, trendsof coordination number vs. chemical shift are the most

-85.01000 I 100 L200 1300

Temperature in "CFig. 8. (a) The'?3Na isotropic chemical shifts for NaMg2 liq-

uid vs. temperature. (b) The '?eSi isotropic chemical shifts forNaMg2 liquid as a function of temperature.

increase in Si coordination in crystalline silicon phos-phates and alkali silicophosphate glasses, but they ruledit out as an explanation for the concentration of t5tSi inalkali silicate glasses.

Temperaturedependent coordination environment of Mg

One of the most intriguing findings of this study is thestrong temperature dependence in 25Mg chemical shift inthe sodium silicate liquid, from which data could be gath-ered over a range of temperature. Chemical shifts canchange with temperature as a result of structural changes,as well as changes in the electronic character ofthe bondswith temperature. For 25Mg, the latter effect has beencarefully studied in MgO, where increased temperaturecauses a slight deshielding (+2 ppm from 25 to 1300 "C)attributed to the increased orbital overlap caused by ther-mal vibrations at high temperature (Fiske et al., 1994).The relatively large decrease in chemical shift with tem-perature observed for 25Mg in the liquid is in the oppositedirection than that caused by electronic effects and musttherefore be caused by substantial changes in the struc-ture of the liquid. In fact, the structural changes may belarger than observed if a temperature-dependent para-

TAErE 4. Relationships between bond lengths and chemical shiftsfor several cations in silicate minerals

857

TaBLE 5. Change in apparent coordination number for severalcations based on high-f NMR and the relations be-tween chemical shifts and bond lengths in Table 4

46,*' AcN*(ppm/ (per

100 "c) 100'c)

Si (in NaMg2)(1120-1370'C)

Na (in NaMg2)(1 120-1370.C)

Na (in NDS10AL-2)(1000-1340.c)

Al (in NDS10AL-2)(500-1340 €)

Mg (in NaMg2)(1 1 20-1 370 qC)

FISKE AND STEBBINS: NMR STUDY OF Mg IN SILICATE LIQUIDS

cN/db(cN/ppm)

6o(ppm)

Cat-ion Mineral

M-Obond

distancecN (A)

M-O/6b(fuppm)

0.019

0.038

0.16

0.05

Si

AI

Na

Mg

av. of 21 sitesav. of 5 sitesav. of 14 sitesav. of 11 sitesP-Na,Si,Os (Na1)P-NarSiros (Na2)akermanitediopside (Ml site)

4 1.63 -90 0.0026 1 .79 -1954 1.7 4 60 0.00326 1 .91 74 2.42 15.6 0.013-6 2.45 9.44 1 .915 49 0.0046 2.077 8

0

-0.3

-0.3

-0.6

-1.7

0

0.05

0.05

o.02

0.09

this study

this study

Stebbins and Farnan, 1992

Stebbins and Farnan, 1992

this study. The stated value for aNa is based on 1 2 data points (Xue and Stebbins,1993).

consistent [less so for 23Na, for which 6,". is very wellcorrelated to the Na-O bond distance (Xue and Stebbins,1993)]. Therefore, the measured Adt*/Af for 2sMg, 2eSi,

and 23Na from this study and 2esi, 2tAl, and 23Na fromthe study ofStebbins and Farnan (1992) are divided byAD*"/ACN (where CN : coordination number) to providea rough estimate of the change of mean apparent coor-dination number with temperature, ACN/AZ. Using amean value for the M-O bond distance for the cations ineach coordination, an estimate of the M-O b<ind thermalexpansion can also be calculated. The chemical shift datafor 'eSi, 27 Al,2sMg, and 23Na in representative coordina-tions are summarized in Table 4.

The results of this analysis are summarized in Table 5and illustrated in Figure 10. Compared with other cat-ions, Mg shows nearly three times as much structuralchange over a given interval oftemperature as the othercations. The next-largest change is seen in Na. This mayreflect some cooperative changes in coordination for bothNa and Mg: since Si does not change its mean coordi-nation by any measurable amount, the large structuralchange for Mg must necessitate some concomitant changesfor Na.

One way to assess the effect of Mg coordination on Nacoordination is to use simple Pauling bond-strength ar-guments for the cations and anions. For example, chang-ing the coordination of Mgz+ from 4 to 6 lowers the nom-inal Mg-O bond strength by %. The excess charge on theO atom caused by this coordination number increase couldbe offset by the addition of another Na to the coordina-tion sphere around the O atom (r61Na-O bond strength :'/). That'z3Na NMR shows moderate lengthening of theNa-O bond may indicate that the changes associated withMg2+ may actually involve some concomitant changes inNa-O bond length, as well as the addition of anotherNa'+ to the coordination of O. The relaxation time dataalso suggest some cooperative behavior between Mg2+and Nar+ and will be discussed later.

The strong temperature dependence of Mg coordina-tion may reconcile the results of this study at high tem-perature to the studies of glasses, which propose smallermean coordination numbers for Mg. EXAFS studies of

Fe2* and Ni'?+ (Waychunas et al., 1988; Jackson, l99l;Galoisy and Calas, I 99 l, I 992; Farges et aI., 199 3) showedthat these cations are tetrahedrally coordinated in silicateliquids and glasses of similar composition to those in thisstudy. In addition, the strong blue color of the glassesdoped with Co2+ in this study suggests that Co2+ resides,at least partially, in tetrahedral coordination. Hanada et al.(1988) observed a similar coloration in Co'z*-doped glassesin the system MgO-SiOr, which they interpreted to indicatet41co2+. Although divalent cations of similar size may notall reside in the same structural environment in silicate liq-uids and glasses (e.g., Williams et al., 1989; Cooney andSharma, 1990; Keppler, 1992), the results of these studieshave suggested to some workers that Mg3*, being smallerthan Fe2+, and Co2* may be more likely to be tetrahedrallycoordinated in silicate glasses.

On the other hand, our results appear to disagree withrecent high-temperature EXAFS and XANES studies ofNi in an alkali silicate glass and liquid, which suggestedshorter Ni-O distances and a lower Ni coordination num-ber in the latter phase (Farges et a1., 1993). It may be thatthe behavior of Ni is somewhat different from that of Mg,that structural effects with ?" are highly nonlinear, or thatNMR and X-ray absorption sample somewhat differentaspects of the structure.

Iupr,rclrroNs FloR THERMoCHEMTCAL ANDTRANSFORT PROPERTIES OF MAGNESIUM

SILICATE LIQI.IIDS

The strong temperature-dependent behavior of 2tMg

chemical shifts, ifrelated to structural changes such as amean coordination increase, should have significant im-plications for the physical and thermodynamic propertiesof magnesium silicate liquids. The partial molar heat ca-pacity of MgO is the highest ofany major oxide in silicateliquids, indicating that it has the strongest interaction withthe network (Stebbins et al., 1984). Recent heat-capacitymeasurements of silicate liquids in the system MgO-AlrOr-SiO, suggest that the temperature-dependent structuralchanges taking place in the liquid involve changes in thelocal coordination environment of the cations (Courtialand Richet. 1993). Recent dilatometric and calorimetric

Mg

,Na

-Al-si

858 FISKE AND STEBBINS: NMR STUDY OF Mg IN SILICATE LIQUIDS

1.0

0.8

0.6

0.4

0.2

0.0

F oL c gO c )

( . ) 9

Y a )

-0.2600 800 1000 1200

Temperature in "C1400

Fig. 10. Schematic representation of the relative structuralchange (e.g., apparent mean coordination number) forthe cationsin NaMg2 liquid based on observed changes in the chemical shiftsfor these cations vs. temperature from {. Data for rTAl are alsoincluded from Stebbins and Farnan (1992).

determinations of thermal expansivity for silicate liquidsin the system diopside-anorthite showed a strong de-crease in the expansivity of the diopside composition withincreasing temperature compared with anorthite (Knocheet al., 1992). This may indicate some temperature-depen-dent changes in cation coordination.

Our data also show significant compositional effects onMg coordination. However, calculations of the partialmolar heat capacity and volume of MgO in silicate meltshave found constant values (e.g., Stebbins et al., 1984;Courtial and Richet, 1993; Lange and Carmichael, 1987,1990). This may be because of the limited compositionaland temperature range explored by existing bulk propertymeasurements and the difficulty of measuring very Mg-rich liquids. For example, the most complete set of den-sity data for silicate liquids (used to derive partial molarvolumes and expansivities of the oxide components ofnatural silicate liquids) includes data for 13 Mg-bearingsilicate liquids, most of which were Ca-rich (Lange andCarmichael, 1987). Only four samples contained any al-kalis, and in these, the Ca to alkali ratio was >3.

An increase in mean Mg coordination with tempera-ture may seem counterintuitive. Pressure, rather thantemperature, is generally associated with coordinationnumber increases for cations in silicate liquids (e.g., Xueet al., l99l). Strictly speaking, the ,sMg chemical shiftchanges with temperature reflect a lengthening of theMg-O bond, which may or may not lead to a discreteincrease in the coordination number. Although the mag-nitude of Mg-O bond expansion with temperature isroughly three times larger than that observed in solids, itis consistent with the general expectation that bonds ex-pand with temperature. Another explanation may be thatthe Mg environment in the liquid is becoming more sym-metric with increasing temperature. Previous workers (e.g.,Yin et al., 1983; Kubicki and Lasaga, l99l) have pro-

posed that Mg may reside in a distorted octahedral sitein MgSiO, liquid, with two O atoms at longer distancesthan the other four. MD simulations showed that the twolonger Mg-O bonds correlate their vibrations so that oneis closest to the Mg when the other is furthest away (Ku-bicki and Lasaga, l99l). A dynamical average ofthis en-vironment might appear to be a net expansion of the siteas a whole and a mean coordination number increase.For example, in the case of the mineral cryolite, the co-ordination number of the central Na atom increases from8 to 12 at high temperature because of a dynamical av-eraging of the coordinating F atoms (Yang et al., 1993).However, it is unclear if this scenario would account forthe configurational changes suggested by the thermo-chemical data for magnesium silicate liquids.

Lange and Carmichael (1987, 1990) have recentlycompiled data on molar volumes and thennal expansionsof silicate liquids. They report a value for the temperaturederivative of the partial molar volume of the MgO com-ponent, which, when divided by the partial molar vol-ume, results in a value of 2.3 x l0 4/K, which is roughlyfive times greater than the expansivity for crystalline MgO(e.g., Hazen and Finger, 1982). Although some or evenmost of this extra expansion in the liquid may take placethrough the rotation of cation polyhedra, its large valuedoes suggest that additional mechanisms may be present.

An important difference between the effects of temper-ature on a liquid and those on most rigid solids is ofcourse the greater range of structural configurations thations can adopt in the former. As temperature increasesin a silicate liquid, network modifiers in particular maybegin to sample larger sites that are energetically rela-tively unfavorable because of longer bond lengths orcharge imbalance. Some of these sites may be vacancysites at low Z; some may be large sites more frequentlyoccupied by larger alkali cations. The effect on a bulkproperty such as molar volume or on an isotropic chem-ical shift results from the time-weighted average of allsite occupancies.

Cation site exchange may have a particularly importantrole in the configurational entropy, enthalpy, and heatcapacity of a liquid, particularly if exchange does occurwith vacant sites. A close-packed O structure can serveas a crude starting point for estimating the possible mag-nitude of such effects. Such a structure contains two tet-rahedral and one octahedral cation site per O. The mixingof cations on one-fourth to one-half of the larger sitescould contribute 5-6 J/K per mole of O to the entropyof the liquid. This is a substantial fraction of the config-urational entropy increase from the glass transition to themelting point of a composition such as CaMgSirO. [6.7J/K per mole of O (Richet and Neuville, 1992)1. For acation such as Mg2+ that may partially occupy tetrahedralsites, mixing onto octahedral vacancies could also be im-portant, as could exchange with Si. It is likely that siteexchange of network modifiers is closely coupled withsilicate species exchange and Si-O bond breaking, whichat least in high-silica liquids has been shown to be closely

FISKE AND STEBBINS: NMR STUDY OF Mg IN SILICATE LIQUIDS 859

tied to the mechanism of the transition from glass to liq-uid. We have only rather indirect evidence to supportsuch speculations at this time, but such effects should beconsidered in models of melt thermodynamics. It is clearthat these kinds ofeffect have the potential to contributemuch more to the bulk properties of the liquid than doessimple mixing of different coordinations of Mg, of I4rSiand t5tSi (Stebbins, l99l), or of Si species with varyingnumbers of bridging and nonbridging O atoms (Brandrissand Stebbins, 1988).

DvNlmrc BErrAvroR oF MglND Na

As stated earlier, NMR relaxation time measurementscan yield information about the energetics of diflusivecation motion in materials (e.g., Stebbins, 1988). In thecase of silicate liquids, the activation energies derivedfrom NMR data may be in close agreement with thosederived from measurements of macroscopic propertiessuch as tracer diffusion and electrical conductivity (e.g.,L iu et a l . , 1987, 1988).

The activation energy for Na relaxation derived fromthe 23Na NMR data (85 kJ/mol) is in the same range asthat measured in other compositions of approximatelysimilar composition and degree of polymerization (e.g.,Na,Si,O,, 67 kJ/mol; NaKSirOr, 95 kJ/mol; Liu et al.,1988). The fact that the activation energy determined inthe NaMg2 silicate liquid is relatively high and similar tothat for the mixed alkali disilicate liquids studied by Liuet al. (1988) may reflect a partial mixed-cation effect be-tween the Na and the Mg.

The activation energy for Mg motion derived from the'?5Mg NMR data (l19 kJlmol) is intermediate betweenthat of Na and that likely for Si under these conditions(170 kJ/mol: Farnan and Stebbins, 1990). Again, this im-plies that Mg seems to occupy some intermediate role inthis silicate liquid, with diffusivities and bond strengthsbetween that of the network-forming Si and network-modifying Na cations.

Few measurements of Mg tracer diffirsion in silicate meltshave been made. Thus, it is impossible to compare directlythe relaxation time results with other data. However, MDsimulation of Mg in MgSiO3 and MgrSiOo [quids over atemperature range of approximately 2000-5000 K by Ku-bicki and Iavga (1991) yield values of 76 and 97 kJ/mot,respectively. They are lower than those calculated for Si inthese compositions but are probably higher than those foralkalis in liquids of similar polymerization.

CoNcr,usroxs

The'zsMg NMR spectra, although difficult to collect,can provide valuable information about the structural roleof Mg in silicate minerals and liquids. The chemical shiftrange between t4lMg and tulMg in silicate minerals may besomewhat larger than that observed in the oxides (Dupreeand Smith, 1988), although further work is needed torefine this trend. The use of 25Mg enriched samples, moresophisticated NMR pulse sequences, and, especially,

higher magnetic fields may improve the applicability of'5Mg NMR to a wider range of problems in mineralogyand geochemistry.

Whereas this study has been limited to simplified com-positions and high temperatures, it has demonstrated thathigh-temperature NMR spectroscopy provides a meansof investigating the specific structural contribution fromeach of the constituent cations that make up silicate liq-uids. The direct implications for the structural role of Mgin compositionally complex natural silicate liquids is un-certain at this point. For example, although the coordi-nation number of Ti has been shown to be composition-ally dependent in compositionally simple silicate liquids,it appears to be fixed over the compositional range ofnatural magmas. In the case of Mg, the compositionaldependence on coordination might be similarly subdued,but the greater abundance of Mg in natural silicate liquids(particularly mafic compositions) may cause any struc-tural variation to be physicochemically significant. A moredetailed investigation of the partial molar volume of MgOin a wide range of natural and synthetic silicate liquidswould help to answer this question.

Finally, it has been known for some time that silicateliquids must undergo structural changes with tempera-ture. As most of what is known about the structure ofsilicate melts has come from the study of glasses, thestructure ofwhich represents that ofthe liquid at the glasstransition (usually several hundred degrees below mag-matic conditions), it is critical to assess the structuralchanges that take place as a function of temperature. Thisstudy documents a substantial change in the coordinationof Mg with temperature and suggests that divalent cationssuch as Mg undergo the greatest change in local structurein silicate liquids with temperature. There is no evidencethat the structural changes observed from I100 to 1350"C do not persist down to the glass transition. A deter-mination of the coordination of Mg in these glasses (usingMg-EXAFS for example) would answer this question. Inaddition, high-temperature 25Mg NMR studies of othercompositions would also help to assess over what rangeof composition this temperature dependence is found.

RnrsRExcrs crrED

Brandriss, M.8., and Stebbins, J.F. (1988) Effects oftemperature on thestructures of silicate liquids: 'eSi NMR results. Geochimica et Cos-mochimica Acta. 52. 2659-2669.

Brown, G.E., Jr., Faryes, F., Calas, G., Galoisy, L., Waychunas, G. A.,Xu, N., Schwartz, K., and Dupon, R. (1993) Hieh+emperature XASstudies ofnucleation in glass-ceramics and ofsilica and germanate meltstructure. Stanford Synchrotron Radiation laboratory Proposal no.2155M.

Cooney, T.F., and Sharma, S.K. (1990) Structure of glasses in the systemsMgrSiOn-Fe,SiOo, Mn'SiO"-Fe'SiO., Mg'SiOo-C-aMgSiOo, and MnrSiOo-CaMnSiO.. Joumal of Non-Crystalline Solids, 122, l0-32.

Cot6, B., Massiot, D., Taulelle, F., and Coutures, J.-P. (1992) 'Al NMRspectroscopy of aluminosilicate melts and glasses. Chemical Geology,96,367-370.

Courtial, P., and Richet, P. (1993) Heat capacity of magnesium alumi-nosilicate melts. Geochimica et Cosmochimica Acta, 57, 1267-1275.

Coutures, J.-P., Massiot, D., Bessada, C., Echegut, P., Riffiet, J., and

860 FISKT AND STEBBINS: NMR STUDY OF Mg IN SILICATE LIQUIDS

Taulelle, F. (1990) Etude par RMN ,7Al d'aluminates liquides dans ledomaine 1600-2100 lC. Comptes Rendus de I'Academie des Sciencesde Par is. 310. l04l-1045.

Dempsey, M.J., Kawamura, K., and Henderson, C.M.B. (1984) Moleculardynamics modelling of akermanite- and diopside-composition melts.Progress in Experimental Petrology, 6,57-59.

Dingwell, D.B. (1992) Density of some titanium-bearing silicate liquidsand the compositional dependence of the partial molar volume of TiOr.Geochimica et Cosmochimica Acta, 56, 3403-3407.

Dingwell, D.8., and Webb, S.L. (1989) Structural relaxation in silicatemelts and non-Newtonian melt rheology in geologic processes. Physicsand Chemistry of Minerals, 16, 508-516.

-(1990) Relaxation in silicate melts. European Joumal of Miner-alogy, 2, 427-449.

Dupree, R, and Smirh, M.E. (1988) Solid-state magnesium-25 NMRspectroscopy. Joumal of the Chemical Society: Chemical Cornmuni-cations, 1483-1485.

Engelhardt, G., and Michel, D. (1987) High-resolution solid-state NMRof silicates and zeolites, 485 p. Wiley, New York.

Farges, F., Brown, G.8., Jr., Calas, G., Galoisy, L., and Waychunas, G.A.( I 993) High-temperature Xray absorption spectroscopy of Ni- and Gein glasses and liquids (abs.). Eos, 74, 630.

-(1994) Temperature-induced structural transformations in Ni-bearing silicate glass and melt. Geophysical Research ktters, in press.

Farnan, L, and Stebbins, J.F. (1990) High-temperature,,Si NMR inves-tigation of solid and molten silicates. Journal of the American Chem-ical Society, ll2, 32-39.

Fiske, P.S., Stebbins, J.F., and Farnan, I. (1994) Bonding and dynamicalphenomena in MgO: A high-temperature r?O and ,sMg NMR study.Physics and Chemistry of Minerals, in press.

Fukushima, E., and Roeder, S.B.W. (1981) Experimental pulse NMR: Anuts and bolts approach, 539 p. Addison-Wesley, Reading, Pennsyl-vanla.

Galoisy, L., and Calas, G. (1991) Spectroscopic evidence for five-coor-dinated Ni in CaNiSirOu glass. American Mineralogist, 76, 1777-1780.

- (1992) Network-forming Ni in silicate glasses. American Miner-alogist,77, 677-680.

- (1993a) Slructural environment ofnickel in silicate glass/melt sys-tems. I. Spectroscopic determination of coordination states. Geochim-ica et Cosmochimica Acta, 57, 3613-3626.

- (1993b) Structural environment ofnickel in silicate glass/melt sys-tems. II. Geochemical implications. Geochirnica et Cosmochimica Acta,s7, 3627-3633.

Ghose, S., and Tsang, T. (1973) Structural dependence ofthe quadrupolecoupling constant e2quhfor 21Al and, crystal field parameter D for Fe3*in aluminosilicates. American Mineralogist, 58, 7 48-7 55.

Grandinetti, P.J., Baltisberger, J.H., Llor, A., ke, Y.K., Werner, U., East-man, M.4., and Pines, A. (1993) Pure absorption-mode lineshapes andsensitivity in two-dimensional dynamic-angle spinning NMR. JournalofMagnetic Resonance Series A, 103,72-81.

Hanada, T., Soga, N., and Tachibana, T. (1988) Coordination state ofmagnesium ions in rf sputtered amorphous films in the system MgO-SiO,. Journal of Non-Crystalline Solids, lO5,39-44.

Hazen, R.M., and Finger, L.W. (1982) Comparative crystal chemistry,231 p. Wiley, New York.

Hovis, G.L., Spearing, D.R., Stebbins, J.F., Roux, J., and Clare, A,. (1992)X-ray powder diffraction and '?rNa,

"Al, and ,,Si MAS-NMR investi-gation of nepheline-kalsilite crystalline solutions. American Mineralo-etsr,77 , 19-29.

Jackson, W.E. (1991) Spectroscopic studies offerrous iron in silicate liq-uids, glasses and crystals, 162 p. Ph.D. thesis, Sanford University,Stanford, California.

Jackson, W.E., Mustre de kon, J., Brown, G.E., Jr., Waychunas, G.A.,Conradson, S.D., and Combes, J.-M. (1993) High-temperature XASstudy ofFerSiOo liquid: Reduced coordination offerrous iron. Science,262,229-233.

Janes, N., and Oldfield, E. (1985) Prediction ofsilicon-29 nuclear mag-netic resonance chemical shifts using a group electronegative approach:Applications to silicate and aluminosilicate structures. Joumal of theAmerican Chemical Society, 107, 6769-6775.

Keppler, H. (1992) Crystal field spectra and geochemislry of lransition

metal ions in silicate melts and glasses. American Mineralogist, 77,62-75.

Kirkpatrick, R.J. (1988) MAS NMR spectroscopy of rninerals and glasses-In Mineralogical Society of America Reviews in Mineralogy, 18, 341-404.

Knoche, R., Dingwell, D.B., and Webb, S.L. (1992) Temperature-depen-dent thermal expansivities of silicate melts: The system anorthite-di-opside. Geochimica et Cosmochimica Acta, 56,689-699.

Kubicki, J.D., and Iasaga, A.C. (1991) Molecular dynamics simulationof pressure and temperature effects on MgSiO, and MgrSiOo melts andglasses. Physics and Chemistry of Minerals, 17 , 661-673.

Kubicki, J.D., Hemley, R.J., and Hofmeister, A.M. (1992) Raman andinfrared study of pressure-induced structural changes in MgSiO,,CaMgSiOu, and CaSiO, glasses. American Mineralogist, 77,258-269.

lange, R.L., and Carmichael, I.S.E. (1987) Densities of NarO-KrO-CaO-MgO-FeO-Fe,O.-Al,Or-TiO,-SiO, liquids: New measurements and de-rived partial molar properties. Geochimica et Cosmochimica Acta, 5 l,2931-2946.

-(1990) Thermodynamic properties of silicate liquids with an em-phasis on density, thermal expansion and compressibility. Mineralog-ical Society of America Reviews in Mineralogy, 24,25-64

Lindsay, J., and Tossell, J. (1991) Ab initio calculations of'?O and nINMR parameters (nT : 3tP,

"Si) in HTTOTH, dimers and TrO, tri-meric rings. Physics and Chemistry of Minerals, 18, 191-198.

Lippmaa, E., Samoson, A., and Miigi, M. (1986) High-resolution "AlNMR of aluminosilicates. Journal of the American Chemical Society,108 ,1730 -1735 .

Liu, S., Pines, A., Brandriss, M., and Stebbins, J.F- (1987) Relaxationmechanisms and effects of motion in albite (NaAlSi3Os) liquid andglass: A high temperature NMR study. Physics and Chemistry of Min-erals, 1 5, 1 55-162.

Liu, S., Stebbins, J.F., Schneider, E., and Pines, A. (1988) Dffilsive mo-tion in alkali silicate melts: An NMR study at high temperature. Geo-chimica et Cosmochimica Act^.52. 527-538.

MacKenzie, K.J.D., and Meinhold, R.H. (1994a) Thermal reactions ofcrysotile revisited: A ,,Si and ,5Mg MAS NMR study. American Min-eralogist, 79,43-50.

- (1994b)'?5Mg nuclear magnetic resonance spectroscopy ofmineralsand related inorganics: A survey study. American Mineralogisr, 79,250-260.

Matsui, Y., Kawamura, K., and Syono, Y. (1982) Molecular dynamicscalculations applied ro silicate systems: Molten and vitreous MgSiO,and Mg,SiOo under low and high pressure. In S. Akimoto and M.H.Manghnani, Eds., High pr€ssw€ research in geophysics: Advances inearth and planetary sciences, vol. 12,p.5ll-524. Reidel, Boston, Mas-sachusetts.

McMillan, P. (1984) A Raman spectroscopic study of glasses in the systemCaO-MgO-SiOr. American Mineralogist, 69, 645-659.

McMillan, P., Wolf, G., and Poe, B.T. (1992) Vibrational spectroscopyof silicate liquids and glasses. Chemical Geology, 96, 351-366.

Miiller, D. (1982) Zur Bestimmung chemischer Verschiebungen der NMR-frequenzen bei Quadrupolkemen aus den MAS-NMR-spektren. An-nalen der Physik, 39, 451-460.

Mysen, B.O., and Frantz, J.D. (1992) Raman spectroscopy of silicate meltsat magmalic temperatures: Na,O-SiO,, K,O-SiO,, and LirO-SiO, bi-nary compositions in the temperature tunge 25-1475 qC. ChemicalGeology, 96,321-332.

Nawotsky, A., Geisinger, K.L., McMillan, P., and Gibbs, G.V. (1985)The tetrahedral framework in glasses and melts: Inferences from mo-lecular orbital calculations and implications for structure, thermody-namics, and physical properties. Physics and Chemistry of Minerals,rr,284-298.

Navrotsky, A., Ziegler, D., Oestrike, R., and Maniar, P. (1989) Calorim-etry of silicate melts at 1773 K: Measurement of enthalpies of fusionand of mixing in the system diopside-anorthite-albite and anorthite-forsterite. Contributions to Mineralogy and Petrology, l0l, 122- 130.

Paris, E., Dingwell, D.8., Romano, C., and S€ifert, F.A. (1993) X-rayabsorption study of Ti in silicate melts (abs.). Ens,74, 347.

Richet, P., and Neuville, D.R. (1992) Tbermodynamics of silicate melts:Configurational properties. In S.K. Saxena, Ed., Thermodlnamic dau:Systematics and eslimation. Advances in Physical Geochemistry, 10,132-t61.

FISKE AND STEBBINS: NMR STUDY OF Mg IN SILICATE LIQUIDS 86 t

Risbud, S., Kirkpatrick, R.J., Taglialavore, A.P., and Montez, B. (1987)Solid-stare NMR evidence of 4-, 5-, and 6-fold aluminum sites in roll-er-quenched SiOr-Al,Or glasses. Journal of the American Ceramic So-ciety, 70, Cl0-C13.

Roedder, E.W. (1951) The system KrO-MgO-SiO,. I. American Joumalof Science, 249, 8l-130.

Rossman, G.R. (1988) Optical spectroscopy. In Mineralogical Society ofAmerica Reviews in Mineralogy, 18,207-254.

Schairer, J.F. (1957) Melting relations of the common rock-forming ox-ides. Journal ofthe American Ceramics Society, 40, 215-235.

Schairer, J.F., and Bowen, N.L. (1942)The binary system CaSiOr-diop-side, and the relations between CaSiO. and akermanite. AmericanJournal of Science, 240, 7 25-7 42.

Smith, K.A., Kirkpatrick, R.J., Oldfield, E., and Henderson, D.M. (1983)High-resolution silicon-29 nuclear magnetic rcsonance spectroscopy ofrock-forming silicates. American Mineralogist, 68, 1206- l2l 5.

Smyth, J.n., and Bish, D.L. (1988) Crystal structures and cation sites ofthe rock-forming minerals, 332 p. Alten and Unwin, I-ondon.

Stebbins, J.F. (1988) NMR spectroscopy and dynamic processes in min-eralogy and geochemistry. In Mineralogical Society of America Re-views in Mineralogy, 18, 405-430.

- (199 I) Nuclear magnelic resonance at high temperature. ChemicalReviews. 91. 1353-1373.

Stebbins, J.F., and Farnan, I. (l 992) Effects ofhigh temperature on silicateliquid structure: A multinuclear NMR study. Science, 255, 586-589.

Stebbins, J.F., and McMillan, P. (1993) Compositional and temperatureeFects on five coordinated silicon in ambient pressure silicate glasses.Journal of Non-Crystalline Solids, I 60, | | 6-125.

Stebbins, J.F., Carmichael, I.S.E., and Moret, L.K. (1984) Heat capacitiesand entropies of silicate liquids and glasses. Contributions to Miner-alogy and Petrology, 86, l3l-148.

Tossell, J., andLazzerctti, P. (1987) Ab initio calculations ofoxygen nu-clear quadrupole coupling constants and oxygen and silicon NMRshielding constants in molecules containing Si-O bonds. ChemicalPhysics, I 12, 205-212.

Waseda, Y. (1980) The structure of non-crystalline rnaterials, 326 p. Mc-Graw-Hill, New York.

Waseda, Y., and Toguri, J.M. (1977) The structure of molten binary sil-icate systems CaO-SiO, and MgO-SiOr. Metauurgical Transactions, 88,563-568.

Waychunas, G.A., Brown, G.E., Jr., Ponader, C.W., and Jackson, W.E.(1988) Evidence from X-ray absorption for network-forming Fez+ inmolten alkali silicates. Nature. 332. 251-253.

Williams, Q., McMillan, P., and Cooney, T.F. (1989) Vibrational spectraof olivine composition glasses: The Mg-Mn join. Physics and Chem-istry of Minerals, 16,352-359.

Xue, X., and Stebbins, J.F. (1993) ,3Na chemical shifts and the local Nacoordination environments in silicate crystals, melts and glasses. Phys-ics and Chemistry of Minerals, 20,297-307.

Xue, X., Stebbins, J.F., Kanzaki, M., McMillan, P.F., and Poe, B. (1991)Pressure-induced silicon coordination and tetrahedral structural changesin alkali oxide-silica melts up to 12 GPa: NMR, Raman, and infraredspectroscopy. American Mineralogist, 7 6, 8-26.

Yang, H., Ghose, S., and Hatch, D.M. (1993) Ferroelastic phase transitionin cryolite, Na.AlFu, a mixed fluoride perovskite: High temperaturesingle crystal X-ray diffraction srudy and symmetry analysis ofthe tran-sition mechanism. Physics and Chemistry of Minerals, 19,52E-544.

Yin, C.D., Okuno, M., Morikaw4 H., and Marumo, F. (1983) Structure,nal-ysis of MgSiO, glass. Journal of Non4rystalline Solids, 55, l3l-141.

MeNuscnrr nrsvm hNumY 26, 1994MeNuscnm AoCEPTED Mev 24, 1994