American Mineralogist, Volume 77, pages 495-506, 1992

Phosphate speciation in potassium aluminosilicate glasses

Hao GaN, Paur, C. HnssDepartment of Geological Sciences, Brown University, Providence, Rhode Island 029 12, U.S.A.

AnsrnlcrParallel- and perpendicular-polarized Raman, MAS, and static 3'P NMR spectra were

collected to investigate phosphate speciation in potassium aluminosilicate glasses. Phos-phate speciation is critically dependent on the K,/(K + Al) (K*) ratio. In peralkaline melts(K* : 0.75), a series of high-frequency Raman bands are assigned to phosphate structuralunits of different degrees of polymerization (from monomer to dimer). The 3'P NMRspectra confirm the assignments. The MAS and static 3'P NMR spectra indicate that Presides in an AIPOo-type environment outside the tetrahedra (SiOo)-K(AlO4) network inperaluminous melts (K* : 0.35). A1OP bonds (A1PO4 type) and KOP bonds (KPO, type)form the dominant structural units in subaluminous melts (K* : 0.5).

INrnooucrroN metaluminous melts are sensitive to the PrO, content

Homogeneous redox equilibria and heterogeneous phase (Ryerson and Hess, 1978). Redox ratios are also impact-

equilibria experiments demonstrate that the alkali-Al ra- ed: the addition of PrO, to granitic melts decreases the

tio is one of the most significant cht atio in peraluminous melts but increases

controlling the solution properties of t n and Hess' 1992; Dickenson and Hess'

ions in silica-rich melts (Gwinn and F runts of PrO, lower the liquidus tem-

and Hess, 1988; Dickenson and Hess, aturated granitic melts (wyllie and Tut-

tel, 1986; Naski and Hess, 1985; Di< len the two liquid solvus in the leucite-

1985; Watson , lgTg). Alkalis in mola lm (Visser and Koster van Groos' 1979;

for example, sharply lower the activity c he solubility of phosphate is greatest in

TiOr, SnOr, FerOr, and PrO, in both < and least in granitic melts under iden-

rated rhyolite melts. Moreover, this s and pressures (Dickinson and Hess'

surely reduces activity coefficients of N '79)' It is imperative' therefore' to un-

and MoOr: Raman spectra of alkali sili )tural role of PrO' in silicate melts over

with these oxides indicate that these < ;hemical and physical conditions' The

complexes with alkalis and largely resi aper is to determine the homogeneous

icatenetwork(EllisonandHess, 1989; I anhydrous peralkaline and in subalu-

verweij, l98l; Brawer and white, l9l luminous granitic melts, using a com-

In addition, certain components al an and NMR spectroscopy to analyze

minous effect wherein the activity co P,O,-bearing glasses in the system Sio'z-

field-strength cation is lowered in mel(molar) greater than unity. This was demonstrated to betrue for P by measuring the solubility of LaPOo (monazite Gr'lSS PREPARATIoN AND EXPERIMENTAL METHODS

structure) in (KrO,AlrO3).4SiO, melts with varying K,/Al Glass compositions were synthesized in three differentratios (Ellison and Hess, 1988). The solubility of LaPOo regions of the 80 molo/o SiO, isopleth in the SiOr-AlrOr-is lowest in subaluminous compositions (K : Al moles) K,O system (Table 1, Fig. l), i.e., peralkaline [K* : KrO/and increases with either an increasing or a decreasing (KrO + AlrOr) molar > 0.51 with K+ in excess of Al3*,K,/Al ratio. Clearly, the existence of a peraluminous and peraluminous (K* < 0.5) with Al3* in excess of K+, andperalkaline effect in PrOr-bearing silicate melts implies a subaluminous (K* : 0.5) with equal amounts of Kt andcomplex solution behavior of P. Al3t cations. K* values of 0.75, 0.5, and 0.35 were chosen

This has obvious implications in the geochemical evo- as representative for each region, and PrO, contents var-lution of natural magmas. For example, the high PrO, ied from 0 to 8 molo/o in peraluminous and subaluminouscontents of certain pegmatites relative to high silica rhy- glasses and from 0 to about 3 molo/o in peralkaline glasses.olites may be a consequence of the peraluminous com- The low PrO, contents of the peralkaline glasses are lim-position of pegmatite (London, 1987). Also, crystalliq- ited by crystallization and phase separation.uid distribution coefficients for rare earth elements in Most of the glasses were prepared by the Corning Glass

0003404x/92l0506-0495$02.00 495

SiOz

496 GAN AND HESS: PHOSPHATE SPECIATION IN GLASSES

Al2o3 0.35 0.5 o.75 Kzo

K*=rqoiG<2O+AlrO)

Fig. l. Triangular plot of the glass compositions. All the pointsstand along 80 molo/o SiO, isopleth.

Works. Batches of 500 g each were made by mixing ap-propriate ratios of reagent grade carbonate and oxides,then melted for 4-6 h at 1650- I 800 "C. Melts were thenquenched by rapidly pouring them into a cold steel plate.The glass of sample P3 was melted at 1550 "C andquenched in cold HrO at Brown University.

In order to test the chemical compositions and ho-mogeneity of the glasses, electron microprobe analyseswere performed at Brown University with a Cameca Ca-mebax (TM) electron microprobe, using a l5-kV accel-erating potential, l0-nA beam current, and 25-pm beamdiameter. Si, Al, K, and P were counted for 10 s each.Table I shows glass compositions determined by electronmicroprobe. K loss occurred in some samples duringmelting, but the compositions are still very close to thenominal values. The glasses were homogeneous in thescale of microprobe beam and free of crystals by XRD.

Raman spectra of glasses were collected with a Spex1403 double monochrometer using the green light (5 14.5nm) of a Spectra-Physics 263 Ar* laser. Entrance and exit

TABLE 1. Chemical compositions of the glasses

spectral slit widths were adjusted to give a 3-cm ' reso-lution. Parallel-polarized (HH) and perpendicular-polar-ized (HV) Raman spectra were obtained by rotating ananalyzer placed between the sample and entrance portwithout varying sample position and laser intensity. In-dividual scans were collected aL 2-cmr increments forscan times of I s/increment. A total of 20-70 scans werecollected sequentially and summed together. All the spec-tra were then normalized to represent a sum of 45 scans.Each corresponding HH and HV spectrum has the sameintensity at 1450 cm-' because the background intensityof 1450 cm ' is subtracted from each. Raman spectra ofseveral crystalline potassium phosphates were collectedfrom their powders in capillary tubes.

Most of the static and MAS 3'P NMR spectra weregathered on a Bruker MSL-300 spectrometer with a 7.1-Tcryomagnet at Brown University. The resonance frequen-cy is 121.5 MHz and the spinning rate is 5.0 KHz forMAS NMR. Pulses of 5.9 ps with a repetition time of15-20 s and a delay time of 10-15 ps were used as theacquisition parameters. The static and MAS 3'P NMRspectrum of sample P3 was collected at the University ofIllinois. The resonance frequency is 145.6 MHz and thespinning rate is 9.5 KHz for MAS NMR. An 850/o aque-ous solution of HrPOo served as the 3'P spectrum refer-ence. The increasing positive chemical shift correspondsto deshielded values. Figures showing stacked spectra inthis paper (both Raman and NMR spectra) are drawn toshow as many features as possible. Therefore, the inten-sity scales vary from sample to sample, and the compar-isons are only valid within the same sample.

RlprlN sPEcrRA

The peralkaline glasses

Raman spectra of peralkaline glasses (nominal K* :

0.75) are given in Figure 2. For convenience, the spectraare divided into three separate regions: the high-frequen-cy region above 800 cm-', the midfrequency region be-tween 500 and 800 cm-r, and the low-frequency regionbelow 500 cm '.

In the high-frequency region (>800 cm-'), P-free po-tassium aluminosilicate glass has a polarized band at 1090cm-r and a noticeable shoulder near I150 cm-' (see alsoDomine and Piriou, 1986). This is very similar to the

PrOt(molo/o)

KrO(mol%)

sio, Alros K.o P"ouSamples K' (v',t"/") (wt%) (wt%) (wt%)

sio,(mol%)

Al2o3(mol7o)

14.4 2.613.9 1 .713.7 0.06.3 8.06.3 3.96.5 2.06.5 0.08.9 7.79.4 4.O9.6 1.7

10.6 0.0

5.4 79.1 3.93.5 79.5 4.90.0 79.9 6.5

15.5 73.2 12.57.9 77.1 12.74.1 76.1 15.40.0 79.9 13.6

15.0 73.9 9.58.0 76.7 9.93 .4 78 .6 1010.0 79.0 10.5

P-3P-9P-l 1P-6P-7P-12P-8P-4P-5P- l3P-10

5.77.49.8't7.3

18.522.620.613.214.415.015.8

19.719.219.28.18.58.79.2

11.412.613.214.7

o.78 69.20.73 69.90.68 71.50.34 59.90.33 66.00.29 65.70.33 71.50.48 60.60.49 65.30.49 68.90.50 70.1

GAN AND HESS: PHOSPHATE SPECIATION IN GLASSES 497

0 500 1000 1500Raman Shifts (cm-r)

Fig.2. The Raman spectra of peralkaline glasses. Top line isHH, bottom line is HV spectmm for each sample: Pl I (PrO5 :0o/o), P9 (P'O' : l.7o/o),P3 (P,O': 2.6o1o1.

high-frequency bands in the spectra of KrO-nSiO, glasseswith n > 2 (Matson et al., 1983). The 1090-cm-r andI 150-cm ' bands are present also in the spectra of PrOr-bearing glasses, but their intensities relative to the 490-cm-' bands decrease with increasing PrOr. In addition,two strongly polarized bands of roughly equal intensityappear aL964 and 1008 cm ' in the glass with 1.7 molo/oPrOr. The 1008-cm t band becomes more intense thanthe 964-cm-' band in glasses with 2.6 molo/o PrOr.

The P-free glass has two bands in the midfrequencyregion: a weak, narrow polarized band at 588 cm ' and abroad partially depolarized weak band around 770 cm t.

The 1.7-molo/o and 2.6-molo/o PrO, glasses have a weakbroad new band at 708 cm ' of the HH spectrum.

In the low-frequency region (<500 cm r), the PrOr-freeglass has a broad band at 490 cm-' and a shoulder around392 cm' (see also Domine and Piriou, 1986). The 490-

0 500 l@ 1500Raman Shifts (cm'r)

Fig. 3. The Raman spectra of peraluminous glasses. SamplesP8 (PrO' :0o/o),P12 (P,O': 2o/o),P7 (PrO': 3.90/o)' P6 (P'Ot: 8o/o).

cm-' band loses intensity with respect to the band deve-loping around 450 cm ' in glasses with increasing PrO'concentratlon.

The peraluminous glasses

The PrOr-free glass has two intense bands at 474 and436 cm I, two weak bands at 590 and 786 cm ', and twobroad, partially depolarized bands at 1030 and I 128 cm-'(Fig. 3). With increasing PrO', a relatively strong polar-

498 GAN AND HESS: PHOSPHATE SPECIATION IN GLASSES

1250

s00 1000 1500Raman Shifts (cm-t)

Fig. 4. The Raman spectra of subaluminous glasses. SamplesPlO (P,O5 : 0olo), Pl3 (P,O, : l.7o/o),P5 (P,O, : 4o/o),P4 (PrO,: 7.7o/o).

ized band at I I l0 cm-' replaces the two partially depo-laized bands at 1030 and ll28 cm-'. A new broad po-laized, band around 290 cm ' grows in the low frequencyregion, and the intensity of the 474-cm-' band increasesrelative to the 436-cm 'band.

The subaluminous glasses

The Raman spectrum of the P-free subaluminous glassis similar to its peraluminous counterpart (Fig. 4). Themost significant diferences are the lower intensity of the

-200 -150 -100 -50 0 50 100 150 2NChemical Shifts (ppm)

Fig. 5. The 3'P MAS NMR spectra of peralkaline gJasses.Samples P9 (P,O5 : l.7o/o), P3 (P,O5 :2.60/o).

454-crn-t band relative to the 482-cm-' band and a morenoticeable band at 570 cm-'. As PrO, increases, a strongpolarized band grows at 1088 cm-'.

Tnr 3rP MAS NMR spEcrRA

The 3rP MAS NMR spectra for peralkaline, peralumi-nous, and subaluminous glasses are illustrated in Figures5, 6, and 7, respectively. The two peralkaline glasses havean intense peak near 0 ppm (0.4 to -2 ppm) and a shoul-der near - l l ppm (-10.3 to -11.3 ppm). The peralu-minous glasses have only one peak near -27 ppm. Thechemical shift is not strongly influenced by PrO, content.The spectra of the subaluminous glasses have only onepeak, at about -20 to -22 ppm. The static NMR spectraare shown in Figures 8, 9, and l0 for the glasses withindifferent K* regions and will be described and analyzedin the section discussing band assignment.

R-llr.lN AND 3rP MAS NMR spEcrRA oFCRYSTALLINE PHOSPHATES

Band assignments in the Raman and NMR spectra ofglasses typically rely upon comparisons with the spectraof crystals of similar composition (e.g., Brawer and White,1975; Dupree et al., 1988a). A number of observationssuggests that certain high-frequency vibrational modes ofsilicate glasses, and glass in general, are highly localized,which in turn suggest that the high-frequency vibrationalspectra ofcrystals and their corresponding glasses reflectthe existence of similar structural units. The reader isreferred to the papers of Brawer and White (1975), Fu-

GAN AND HESS: PHOSPHATE SPECIATION IN GLASSES 499

P7 (P2O5

-200 -r50 -100 -50 0 50 100 150 200Chemical Shifts (ppm)

Fig. 6. The 3'P MAS NMR spectra of peraluminous glasses.Samples Pl2 (P,O5 :2o/A,P7 (P,O,: 3.9o/o),P6 (P,O,:3070;.

rukawa et al. (1981), McMillan (1984), Dowty (1987),and Tallant and Nelson (1986) for comprehensive dis-cussions of these questions. Low-frequency vibrationmodes, however, appear to be delocalized and presentsome difficulties in band assignment, since the nature ofthe vibrations responsible for these bands is not well un-derstood. Systematic and detailed comparison of thespectra ofseries interrelated glasses suggests that even thelow-frequency bands, however, are useful in identifoingimportant structural features of glasses. More completeanalyses of the low-frequency spectra are found in Mat-son et al. (1983), McMillan (1984), Mysen et al. (1985),and Domine and Piriou (1986).

The basic principle used in interpreting NMR chemicalshifts is that the shielding of a given nuclei increases withthe increasing ionicity ofthe bonds that the cation makeswith nearest-neighbor O (Kirkpatrick, 1988). The chem-ical shift is also a function ofthe electron redistributionin these bonds caused by changes in next-nearest neigh-bors. These changes, however, are much smaller thanthose due to the primary coordination polyhedron. Theapplication of MAS NMR to glasses, in general, has been

-200 -150 -100 -50 0 50 100 150 2@Chemical Shifts (ppm)

Fig. 7. The 3'P MAS NMR spectra of subaluminous glasses.Samples P5 (P,O, : 4o/A,P4 (P2O': 7.7o1o1.

reviewed by Kirkpatrick et al. (1986). Dupree et al. (1988a,1988b, 1989) provide MAS NMR spectra of variousphosphate-bearing glasses, as well as the spectra of someof their corresponding crystals.

In order to identify the Raman bands that are associ-ated with P-O vibrations, spectra of three crystalline po-tassium phosphates were collected (Fig. I l). These in-clude potassium orthophosphate (K3POo), potassiumpyrophosphate (lLoPrO'), and potassium metaphosphate(KPOJ. These compounds are listed in order of increas-ing polymerization of the POo tetrahedron, i.e., from themonomer to dimer to infinite chain. They contain P0, Pt,and P2 phosphate tetrahedra, respectively, where the P'notation identifies the number of POP (bridging) bondsper phosphate tetrahedron. The Raman spectra of crys-talline phosphates show the following significant features:(l) The most intense peak in the high-frequency regionsmoves toward higher frequency with an increasing degteeof polymerization of the phosphate tetrahedra. Thesepeaks reflect the symmetric vibration of the KOP non-bridging bond (Tallant and Nelson, 1986). (2) The mod-erately intense peak near 700 cm-' exists only in the di-mer and more polymerized phosphates and is absent forthe orthophosphate spectrum. This peak probably rep-resents the symmetric vibration of the POP bridging bond(Tallant and Nelson, 1986).

The 3'P MAS NMR chemical shifts ofthese phosphatesobtained from the literature are listed in Table 2. Thechemical shifts are more shielded with increasing degreesof polymerization of the phosphate. Chemical shifts of

500 GAN AND HESS: PHOSPHATE SPECIATION IN GLASSES

-40.6

P3 (P2O5

-30.8

-400 -300 -200 -t00 0 100 200 300 400Chemical Shifts (ppm)

Fig. 8. The 3'P static NMR spectra of peralkaline glasses.Samples P9 (P,O5 : l.7o/o), P3 (P,O5 : 2.60/o).

AIPO, and SiPrO? are also given in the same table, andother relevant phosphates will be cited in the text.

BA.No ,q,sstcNMENTS

Bands in the peralkaline glasses

The most intense high-frequency Raman band (1090cm-t) in the PrOr-free glass is assigned to the symmetricstretching vibration of nonbridging O of Q3 [generally, Q,notation identifies the number (n) of SiOSi bridging bondsper silicate tetrahedronl and the intense low-frequencyband (490 cm-') is assigned to the delocalized vibrationsof the polymerized aluminosilicate network (Brawer andWhite, 1975; Virgo et al., 1980; Matson et al., 1983; Mc-Millan, 1984). The bands near 704 and l0l4 cm-' gainintensity with increasing PrOr. These bands match the715-cm 'and l0l5-cm-'bands of the Raman spectra ofcrystalline pyrophosphate (Fig. I l), and therefore are as-signed to pyrophosphate-type structural units in the melts.The band near 966 cm-' exists only in PrOr-bearing glass-es but loses intensity with increasing PrOr. The 966-cm-lband is close to the intense 923-cm-t band of crystallineorthophosphate (Fig. I l), although it is about 40 cm-'higher. Nelson and Tallant (1984) observed a 940-cm-rband in their NarO-PrO5-SiO, glasses that has the samewavenumber as the intense high-frequency band in crys-talline NarPOo. This band also loses intensity with in-creasing PrO, content of the glass, as observed in ourglasses. A 957-cm-' band coexists with the l0l0-cm-'

-400 -300 -200 -100 0 100 200 300 400Chemical Shifts (ppm)

Fig. 9. The 3rP static NMR spectra of peraluminous glasses.Samples Pl2 (P,O5 :2o/o),P7 (P,O,: 3.9o/o),P6 (P,O,:3070;.

band in the Raman spectrum of the glass, where bothorthophosphate and pyrophosphate species are identifiedby means of 3'P MAS NMR spectroscopy (Dupree et al.,1988a; see Fig. 12 and analysis in the NMR section be-low). The 964-cm-t band, therefore, is assigned to thesymmetric stretching vibration of nonbridging O (Pn) ororthophosphate tetrahedra.

The 3'P MAS NMR spectra only partially confirm thespecies we have identified in the Raman spectra of per-alkaline glasses. The major peak observed in the 1.7-molo/oand 2.6-molo/o PrO, glasses is close to the chemical shiftof crystalline potassium pyrophosphate (-l.l ppm, seeTable 2). A small bump at - I I ppm coexists with - l.lppm peak in each of the two glasses. Its assignment willbe discussed below. No chemical shifts corresponding toorthophosphate were observed.

The static arP NMR spectra were analyzed using thespectra of crystalline alkali phosphates (Duncan andDouglass, 1984). The 1.7-molo/o and 2.6-molo/o PrO,glasses have an axial symmetry powder pattern (Harris,1983) almost identical to that of crystalline KoPrO, (Dun-

GAN AND HESS: PHOSPHATE SPECIATION IN GLASSES 501

P4 (PrO5

-300 -200 -r00 0 100 2NChemical Shifts (ppm)

Fig. 10. The 3rP static NMR spectra of subaluminous glasses.Samples P5 (P,Os : 4o/o), P4 (PrO, : 7.7o/o).

can and Douglass, 1984) and that is consistent with theinferences from Raman and MAS 3'P NMR. A weak peakat 1.6 ppm occurs in both glasses but has a slightly greaterintensity in the 2.6-molo/o PrO, glass. This peak overlapsthe axial symmetry powder pattern of KoPrO, and seemsrelated to the - I I ppm shoulder in MAS 3'P NMR ofthe same glasses. The MAS 3'P NMR spectra suggest thatthis P site (- I I ppm) is shielded more than that in py-rophosphate and is probably a chainlike phosphate spe-cies (Kirkpatrick, 1988; Dupree et al., 1989). The abun-dance of this species must be low, since its Raman bandis buried under the strong KOSi(Q3) band near 1100-I 150 cm-'. Nevertheless, it is clear that the static NMRprofile of the 1.7-molo/o and 2.6-mol0/o P.O, glass is dom-inated by pyrophosphate species.

It is interesting to note that there is no 3'P chemicalshift indicative ofpotassium orthophosphate in either theMAS or static NMR spectra. Yet the existence of thisspecies in low PrO, glasses is indicated by the 964-cm-'Raman band. Since phosphate species have a strong scat-tering coefficient (Nelson and Tallant, 1984), it is possiblefor Raman spectra to show a much stronger signature of

TABLE 2. MAS 31P NMR chemical shifts of crystals

0 500 1000 1500Raman Shifts (cm-r)

Fig. I l. The Raman spectra of crystalline potassium phos-phates. Among them are orthophosphate (K,PO.), pyrophos-phate (tri.PrOr), and metaphosphate (KPO').

orthophosphate than the 3'P MAS NMR. To check thispossibility, we collected the Raman spectrum of a P-bear-ing high potassium silicate glass from which Dupree etal. (1988a) obtained only a very weak orthophosphateand a very intense pyrophosphate chemical shift (Fig. l2).

K3PO4 KoPrO, G%O'o KPO3 AIPO. siP2o7 P.O,o

1 1 -1.2, -4, -19- -18 .5 , - 19 - -2s.3t -331 -46+

* Grimmer and Haubenreisser (1983)-" Duncan and Douglass (1984).f Bleam et al. (1989).+ Dupree et al. (1989).

s02

600 800 1000 1200 1400Raman Shifts (cm-')

Fig. 12. The Raman spectrum of 32KrO.5PrOr'63SiO, glass.

The Raman spectrum of the glass, nevertheless, shows astrong orthophosphate band at 957 cmt and a pyro-phosphate band ofslightly higher intensity at 1010 cm '

(The 989-cm-' band tKOS(Q)l probably enhances theapparent intensity of l0lO-cm ' band). It seems that theproportion of orthophosphate species is greatly exagger-ated in the Raman spectrum of the glass. Since the con-centration of KrO and PrO, is much lower in our glassesthan in those of Dupree et al., it is not extraordinary thatthe chemical shift for potassium orthophosphate speciesis absent or is concealed by the neighboring strong py-rophosphate peak. Although orthophosphate is only aminor species in these glasses, it is quite likely the dom-inant phosphate species at lower P concentrations (Du-pree et a1., 1988a).

Bands in the peraluminous glasses

The Raman spectrum of the P-free peraluminous glassis almost identical to those of SiOr-AlrO. (McMillan andPiriou, 1982). The addition of Al,O, to SiO, glass andpresumably the exchange of AlrO. for KrO in a subalu-minous glass result in the development of a weak bandnear I 100 cm-' and a loss of intensity of the sharp 480-cm-' band relative to the 450-cm-' band, which are be-lieved to reflect the vibration of =SiOAl units (McMillanand Piriou. 1982).

PrOr-bearing glasses have two broad polarized bandsat I I l0 and294 cm ' that are not observed in the spectraof PrOr-free peraluminous glass (note that I128-cm-r bandin PrOr-free glass is partially depolarized). The spectrumis compatible with a glass structure containing tetrahe-drally coordinated AIPO. species. The dominant high-frequency band at ll08 cm-' is observed in the spectraof crystalline AIPO. (berlinite) and in AIPO.-SiO, glass(Kosinski et al., 1988). Moreover, the 294-cm-' bandis matched by a band near 300 cm-' in the spectra ofAIPO4-SiO, glasses. A shoulder near 1200 cm-' is ob-served also in the spectra of AIPO.-SiO, glasses and as a

GAN AND HESS: PHOSPHATE SPECIATION IN GLASSES

weak band in berlinite (Kosinski et al., 1988). It is clear,therefore, that AIOP bonds similar to those in crystallineAIPO. characterize the peraluminous glasses.

The MAS and static a'P NMR spectra confirm theseassignments. Both spectra show single peaks from around-27 ppm (MAS) to around -28 ppm (static), consistentwith the MAS of berlinite (-25 ppm) and the static spec-tra of amorphous AIPO. (-28 ppm). A shielding tensorof cubic symmetry is indicated by the static NMR powderpattern, in agreement with the cubic symmetric shieldingtensor observed in crystalline and amorphous AIPO.(Kosinski et al., 1988).

Bands in the subaluminous glasses

The Raman spectrum of the PrOr-free subaluminousglass is similar to that of vitreous silica. The major dif-ferences in the Raman spectra between SiOr-KAlO, andSiO, systems are the appearance of high-frequency bandscaused by the vibration of SiOAI species and loss of in-tensity of the 450-cm-r band relative to the 490-cm-rband. The last two bands reflect the symmetrical stretchof the bridging O atom in SiOSi and SiOtotAl bonds, re-spectively (McMillan et al., 1982). The KOSi(Q3) band isnot observed. Therefore all K* act as charge-balancingcations for the SiOt4rAl bonds. The addition of P increasesthe intensity of the 450-cm-' band (SiOSi) relative to the490-cm-' band (SiOrrtAl) and produces a new band around1090 cm-'. The band near 1090 cm-r has two possibleorigins: AIPO. and KOS(Q) both have major bands atthis frequency. The increasing intensity of the 450-cm-'band indicates the development of a Q'-type SiOSi net-work that argues against the KOSi(Q3) assignment. Al-though there is no general agreement in the assignmentof low-frequency Raman bands because of the delocal-ized nature ofthese vibrations, the consistent change ofthe low-frequency vibration envelope with chemicalcomposition indicates that a gain in intensity of the 450-cm ' band reflects the increasing degree of polymeriza-tion of the aluminosilicate network (Matson et al., 1983).This leaves AIPO, as the species responsible for the high-frequency band at 1088 cm-'. Since alumina are incor-porated in the aluminosilicate network together with K,the formation of AIOP bonds (AlPOo type) requires thebreakdown of KOtotAl bonds. This reaction is supportedby the decreasing intensity of the 490-cm-' band that isinterpreted to be due to symmetrical stretching vibrationof SiOtotAl bonds in the silica-rich network (McMillan etal., 1982). The broad envelope from I 100 to 1250 cm '

is difficult to interpret. Berlinite (AIPOJ has a weak bandat 1220 cm ', and the AlPOo-bearing peraluminous glass-es have a broad band near this frequency (Kosinski et al.,1988). Thus, some of the intensity of the 1250-cm-'bandmay be caused by an AIPO. species. Crystalline KPO,has a major band near I 153 cm-' and may also contributeenergy in this range. The formation of KOP bonds is, infact, required by mass balance to accommodate the Kreleased from broken KOt4lAl bonds (see the discussion

GAN AND HESS: PHOSPHATE SPECIATION IN GLASSES 503

section). It is concluded, therefore, that both AIPO. andpotassium phosphate chains coexist in P-bearing subalu-minous glasses. Additional evidence in support of thisinterpretation is given below.

The MAS IP NMR isotropic chemical shifts (-20 to-22 ppm) exclude the existence of SiPrOr-type and PoO,o-type species that have chemical shifts near -35 and -46ppm, respectively (Table 2). No 3'P MAS NMR data areavailable for tatSiOP-type species in glasses. However thelower coordination number of the Si atom tends to in-crease its bond order, which will in turn cause moreshielding to 3'P nuclei (more negative in this case). There-fore, it is very unlikely that -20 ppm chemical shifts arecaused by t4rsioP-type structural units. This implies thatKOt4lAl bonds have to be broken to form new P-bearingspecies. The -20 to -22 ppm peaks are within the rangeof the chemical shift of KPO. (-19 ppm, Duncan andDouglass, 1984) and AIPO4 (-25 ppm, Bleam et al.,1989). The FWHH of the -20 to -22 ppm peak is aboutl6 ppm, which could cover both the - l9 and -25 ppmpeaks. Thus, it is possible that the isotropic chemical shiftsfrom KPO.-type and AIPOo-type structural units arecombined into a single broad peak.

Molecules with similar isotropic chemical shifts couldhave a distinct symmetry of shielding tensors that appeardifferent in the static a'P NMR powder patterns. The stat-ic 3'P NMR spectra of the 4 and 8o/o PrO, glasses have aprominent low-symmetry powder pattern that is verysimilar to that of (NaPOr)n (- 148, 16, and 90 ppm; Dun-can and Douglass, 1984). A strong shoulder around -29ppm falls within the range of the AIPO. chemical shift.The static rlP NMR of AIPO. in peraluminous glasses(Fig. 9) shows a single, isotropic, symmetrical peak ataboul -27 to -30 ppm. Therefore, the static 3'P NMRspectra can be modeled by combining the powder pat-terns of the shielding tensors of cubic symmetry (AIPO.)and nonaxial symmetry (KPO3), as deduced from MAS3'P NMR.

DrscussroN

Solution rnodel

The structure ofthe P-free peralkaline glass is relativelywell understood. The role of K is that of a charge-bal-ancing cation for AlOo tetrahedra and as a network-mod-ifying cation for KOSi(Q3) tetrahedra. The silicate net-work is a solution of copolymerizing Q3 and Qa silicatetetrahedra and charge-balanced AlOn tetrahedra (Engel-hardt et al., 1985; Kirkpatrick et al., 1986; Domine andPiriou, 1986; McMillan et al., 1982; Mysen et al., 1981a).The addition of P causes significant modifications to theglass structure. P enters as variously polymerized phos-phate tetrahedra using K as a network-modifying cation.The speciation ofthe potassium phosphates is controlledby the ratio ofexcess K (i.e., K in excess ofthat neededto charge balance the AlOo tetrahedra) over P; the lowerthe ratio, the more polymerized the phosphate species.

Thus, phosphates exist first a monomers, then dimers,and then probably small chains as the P concentration isincreased. In the process, P strips network-modifying Kfrom KOSi(Q3) species, resulting in a higher state of po-lymerization of the aluminosilicate network. The in-creased degree of polymerization with increasing PrO' inperalkaline silicate glasses was also observed by other re-searchers. Nelson and Tallant (1984) reported that ad-dition of PrO, to metasilicate (Q'?) glasses gives rise to anew Raman band characteristic of the symmetricalstretching vibration of NaOSi (Q3). Dupree et al. cameto the same conclusion by means of "Si MAS NMR spec-troscopy (Dupree et al., 1988a): the resonance of Qo typestructural units increases in intensity as PrO, increases inperalkaline glasses. The occurrence of cristobalite in 7.6mol0/o PrO, peralkaline glass prepared for, but not ana-lyzed, in this study is consistent with the above conclu-sions. Therefore, the homogeneous equilibria can be ex-pressed as

2KOSi + POP: 2KOP + SiOSi (l)(a) G) (P',P) (a')

where P' refers to POo tetrahedra containing n POPbridging bonds. The phosphate species must largely re-side outside the aluminosilicate network, as there is noevidence for the existence of Si and P bonding in theMAS 3'P NMR spectrum (see also Nelson and Tallant,1984;Dupree et a l . , 1988a).

The structural role of excess Al in peraluminous meltshas been a topic ofconsiderable debate (e.g., Lacey, I 968;Mysen et al., 1980, l98la; McMillan and Piriou, 1982;Sato et al., I 99 I ). It has been concluded that tetrahedrallycoordinated Al is the primary structural unit for Al insubaluminous to moderately peraluminous melts. Five-fold coordinated Al is observed only as a minor speciescoexisting with the fourfold coordinated Al inl5CaO35AlrOr50SiO, glass (AlrOr/CaO : 2.3; Sato et al.,l99l). For glasses with an even higher AlrOr/CaO ratio(7 .6 to 12.2), tl;'e data from 'z?Al MAS NMR suggest thata portion of the Al is in highly distorted sites in additionto the four-, five-, and sixfold coordinated sites (Sato etal., l99l). The AlrOr/CaO ratios of the peraluminousglasses of this study are less than or equal to 2.3, andthelow excess AlrO3/SiO, ratios ( < 0. I 3) of the glasses wouldalso stabilize Al in tetrahedral sites (Risbud et al., 1987).It is concluded, therefore, that excess Al is primanly lo-cated in tetrahedral sites in the peraluminous melts andit forms triclusters with or without SiOo units (noted asAIOSi and AIOAI for "with" and "without," respectively,in the following).

PrO, reacts with excess Al in peraluminous melts toform AIPO. species. If the excess Al exists as triclusterscontaining silicate tetrahedra, then the homogeneousequilibrium is

2AlOSi + POP: 2AIOP + SiOSi (2)

where the formation of AIOP units results in the poly-merization of the silicate network. Alternativelv. if the

504 GAN AND HESS: PHOSPHATE SPECIATION IN GLASSES

excess Al exists outside the silicate network, then the ho-mogeneous equilibrium is

AloAl + PoP:2AIOP. (3)

Reaction 2 may be indicated by the increase in inten-sity of the 475-cm ' band, which is believed to reflect thevibrations of bridging O in the aluminosilicate network(McMillan and Piriou, 1982).

One question raised about the above two reactions iswhether AIOP bonds reside within the aluminosilicatenetwork, forming Si-O-AI-O-P-O type species, or outsidethe network, as a separate species. It has been argued thatthere is no independent MAS NMR data for the isotropicchemical shift of the species Si-O-AI-O-P-O in crystal orglasses, thereby, rendering our previous 3'P MAS NMRdiscussion less convincing. There is not yet a generalagreement on this. The analysis of vibrational spectrayields different interpretations. Mysen et al. (198 lb) andTallant and Nelson (1986) argued against the existenceofsuch a species because ofthe absence ofthe SiOP Ra-man band, which is highly localized in the high-frequencyregion. However, Mysen's conclusion is based on some-what subjective deconvolution analysis, and the glassesof Tallant and Nelson are far too low in SiO, concentra-tion to be compared to this study. In contrast, Kosinskiet al. (1988) argued for the Si-O-AI-O-P-O type structuralunits in glass of 50SiO,-25AlrO3-25PrO5 based on thesimilarity of Raman spectra between the glass and vit-reous silica. Considering the ambiguity involved in theinterpretation of Raman spectra, it is possible that 3'P

static NMR spectra may offer clues to the correct model.If Si-O-AI-O-P-O type species exist, the cubic symmetryof the shielding tensor of AIPOo-type species will be de-moted to a lower symmetry, a result which should beresolved in the static spectra. This is not observed; thestatic 3'P NMR spectra has the expected cubic symmetry(Fig. 9). For this reason, it is thus concluded that AIPO.species reside mostly outside the aluminosilicate net-work. This does not exclude the existence of a small num-ber of P-O-Si species, which certainly are predicted fromthermodynamic considerations.

P-free subaluminous melts are composed of fully poly-merized aluminate-silicate networks in which all K cat-ions are incorporated into KAIO. complexes as charge-balancing cations, noted here as KOtarAl (Sharma et a1.,1978; Virgo et al., 19791,Mysen et al., 1980, l98la, 1982;McMillan et al., 1982;- Sato et al., l99l). In analyzing thestructures of peralkaline and peraluminous melts, it isclear that KOI4lAl bonds are very strong compared withKOS(Q) and AIOSi or AIOAI bonds. Once KOSi, AlOSi,or AIOAI bonds are eliminated, PrO. must obtain K andAl from the strong KOt4lAl species. The major differencebetween subaluminous melts and PrOr-rich peralkalineor peraluminous melts is that the charge-balanced sub-aluminous melts contain only KOt4tAl species, even atthe lowest PrO, contents, and potassium phosphates oraluminophosphate are not present when PrO, is requiredto break the KOtatAl bonds. Therefore it is energetically

favorable and stoichiometrically feasible for the firstphosphate species to have a low K,/P (or AllP) ratio. Theproduct is therefore metaphosphate (K/P : l) in subalu-minous melts instead of orthophosphate (K/P : 4), as inlow PrO, peralkaline melts. The homogeneous equilibri-um describing this process is

KAIO, + P2Os: KPO. + AIPO4 (4)

where AIPO. species rather than the less stable AIOSispecies are formed. The intensity of the 450-cm-r Ramanband increases relative to the 480-cm ' band because partof the KOtorAl bonds are eliminated. KOt4rAl bonds de-press the 450-cm ' bands of SiOSi vibration in vitreousSiO, (McMillan and Piriou, 1982). Since PrO' is evenlyshared by potassium metaphosphate and AIPOo-type alu-minophosphate (see Reaction 4), the abundance of eachspecies is roughly half of those similar species in othermelts with the same amount of PrOr. This explains therelatively low intensities of the KOP and AIOP bonds inthe Raman spectra of subaluminous melts.

Applications

The solution behavior of P in silicate melts is rich incomplexity. The activity coefrcient of P is smaller in sil-icate melts that contain network-modifying cations ap-propriate to form stable phosphate complexes. Excess al-kalis perform this function in peralkaline melts, excessAl in peraluminous melts, and mainly divalent cations inmetaluminous melts. This model explains a number ofinteresting experimental observations.

l. The addition of PrO, to multisaturated liquids (e.g.,enstatite-forsterite, cristobalite-enstatite) shifts theboundary curves to lower SiO, contents, thereby increas-ing the 7sio, in the liquids (Kushiro, 1975). This is con-sistent with the homogeneous equilibrium

2MgOSi + POP: 2MgOP + SiOSi (5)

in which the production of MgOP complexes causes thesilicate network to polymerize (Hess, 1980; Ryerson andHess, 1980). The role of P, therefore, is to increase theactivity coefficient of SiO, but to lower the activity coef-ficient of selected network-modifying cations. Converse-ly, the activity coefficient of P is lowest in silicate meltsthat contain the highest concentration of network-modi-fying cation. This factor explains why the solubilities ofapatite and whitlockite are greatest in mafic to ultramaflcmelts rich in network-modifuing cations and least in nearlysubaluminous granitic melts under identical tempera-tures and pressures (Watson, 1979; Hess et al., 1989;Dickinson and Hess, 1983).

2. The solubility of LaPOo (monazite structure) exhib-its both a peralkaline and peraluminous effect in anhy-drous potassium aluminosilicate melts (Ellison and Hess,1988). The solubility is at a minimum in the subalumi-nous melts because the free energy change ofthe reaction

KAIO, + P2O5 : KPO. + AIPO4 (6)

is not as favorable as the free energy changes of the re-

actions in peraluminous and peralkaline melts (see Eqs.2 and l). The reason for this observation, ofcourse, isthat P must attack the very stable KAIO, complex insubaluminous melts but has easy access to the relativelyless stable excess Al and excess K complexes in othermelts. It is noteworthy that the monazite peralkaline ef-fect is seen also in hydrous melts (Montel, 1986), dem-onstrating that the homogeneous equilibrium

2SiOH + POP: 2POH + SiOSi (7'l

does not have a significant role in stabilizing PrOr. Thisconclusion is consistent with the relatively constant sol-ubility of apatite in anhydrous and HrO-saturated gra-nitic melts (Green and Watson, 1982).

3. The strong peraluminous effect on P and the strongassociation of P and rare earth elements (Ryerson andHess, 1978) suggest that P has a strong affinity for M3*.Indeed, the addition of P increases the Fe3*,/Fe2* redoxratios in peraluminous melts and implies the formationof Fe3*POo complexes (Dickenson and Hess, 1983; Gwinnand Hess, 1992). BOP complexes in peraluminous meltsare also anticipated (Gan and Hess, unpublished work).The high P2O5 contents of peraluminous pegmatites rel-ative to high silica rhyolites are likely to be a consequenceof the formation of BPO, complexes (London, 1987).Crystal-liquid partition coefficients of other trivalent cat-ions such as Sc3*, Ga3*, and Y3+, among others, shouldbe sensitive to the PrO, contents of silicate meits.

The solubilities of high field-strength cations (Zrat,Tia*,etc.) in high silica melts strongly depend on the avail-ability of network-modifying cations (e.g., K*, Ca2+). Pi+joins in the competition for network-modifying cations,and therefore reduces their availability to other high field-strength cations, thus changing their solubilities. How PrO,competes with other high field-strength cations for a lim-ited number of network-modifying cations is a questionthat cannot yet be addressed. The establishment ofa hi-erarchy of such interactions is an important goal in thecharacterization of the solution properties of highlycharged trace elements.

Acxuowr-nocMENTS

We wish to acknowledge the generous use of the spectrometers of BillRisen in the Department of Chemistry and of Phil Bray in the Departmentof Physics at Brown University. Thanks also to Jim Kirkparrick for hisassistance in getting additional NMR spectra. H.G. thanks George Tsa-garopoulos for his help in using the Raman spectrometer. We also thankJim Dickinson ofCorning Glass for supplying us with most ofthe glassesand R.K. Brow and two anonymous referees for reviews of the manu-script. Finally, we express our gratitude to the National Science Foun-dation for their generous financial support (granrs EAR-8719358, EAR-90t7790).

RrpnnnNcos crrEDBleam, W.F., Pfeffer, P.E., and Frye, J.S. (1989) 3'P solid state nuclear

magnetic resonance spectroscopy of aluminum phosphate minerals.Physics and Chemistry of Minerals, 16,455-464.

Brawer, S.A., and White, W.B. (1975) Raman spectroscopic investigationofthe structure ofsilicate glasses I. The binary alkali silicates. Joumalof Chemical Physics, 63, 2421-2432.

505

- (1977) Raman spectroscopic study ofhexavalent chromrum rn somesilicate and borate glasses. Materials Research Bulletin, 12,281-288.

Dickenson, M.P., and Hess, P.C. (1981) Redox equilibria and the struc-tural role of iron in aluminosilicate melts. Contributions to Mineralogyand Petrology, 78, 352-357.

-(1983) Fe redox equilibria and the structural role of P5*, Ti'* andFert. Eos. 64. 350.

-(1986) The structural role and homogeneous redox equilibria ofiron in peralurninous, metaluminous and peralkaline silicate melts.Contributions to Mineralogy and Petrology, 92,207-217.

Dickinson, J.8., and Hess, P.C. (1983) Role of whitlockite and apatite inlunar felsite (abs.). Lunar and Planetary Science, 14, 158.

-(1985) Rutile solubility and titanium coordination in silicate melts.Geochimica et Cosrnochimica Acta. 49. 2289-2296.

Domine, F., and Piriou, B. (1986) Raman spectroscopic study of the SiOr-AlrO3-KrO vitreous system: Distribution of silicon and second neigh-bors. American Mineralogist, 71, 38-50.

Dowty, E. (1987) Vibrational interactions oftetrahedra in silicate glassesand crystals: II. Calculations on melilites, pyroxenes, silica polymorphsand feldspars. Physics and Chemistry ofMinerals, 14,122-138.

Duncan, T.M., and Douglass, D C. (1984) On the I'P chemical shift an-isotropy in condensed phosphates. Chemical Physics, 87, 339-349.

Dupree, R., Holland, D., and Mortuza, M.G. ('1988a) The role of smallamounts ofPrO, in the structure ofalkali disilicate glasses. Physics andChemistry of Glasses, 29, 18-21.

Dupree, R., Holland, D., Mortuza, M.G., Collins, J.A., and Lockyer,M.W.G. (1988b) An MAS NMR study of network-cation coordinationin phosphosilicate elasses. Joumal of Non-Crystalline Solids, 106, 403-407.

-(1989) Magic angle spinning NMR of alkali phospho-alumino-silicate glasses. Joumal of Non-Crystalline Solids, I12, I I l-l19.

Ellison, A., and Hess, P.C. (1988) Peraluminous and peralkaline eflectsupon "monazite" solubility in high-silica liquids Eos, 69, 498.

-(1989) Solution properties of rare earth elements in silicate melts:Inference from immiscible liquids. Geochimica et Cosmochimica Acta,53,1965-1974.

Englehardt, G., Nofz, M., Forkel, K., Wihsmann, F.G., Magi, M., Sa-moson, A., and Lippman, E. (1985) Structural studies of calcium alu-minosilicate glasses by high resolution solid static '?eSi and '??Al magicangle spinning nuclear magnetic resonance. Physics and Chemistry ofGlasses.26. 157-165.

Furukawa, T., Fox, K.E., and White, W.B. (1981) Raman spectroscopicinvestigation of the structure of silicate glasses. III. Raman intensitiesand structural units in sodium silicate glasses. Journal of chemicalPhysics, I 5, 3226-3237.

Green, T.H., and Watson, E.B. (1982) Crystallization of apatite in naturalmagmas under high pressure hydrous condition, with particular refer-ence to orogenic rock series. Contributions to Mineralogy and Petrol-ogy,79,96-105.

Grimmer, A.R., and Haubenreisser, U. (1983) High-field static and MASrrP NMR: Chemical shift tensors of polycrystalline potassium phos-phates PrO, xKrO (0 < x < 3). Chemical Physics Irtters, 99, (5-6),487-490.

Gwinn, R., and Hess, P.C. (1989) Iron and titanium solution propertiesin peraluminous and peralkaline rhyolitic liquids. Contributions toMineralogy and Petrology, l0l, 326-338.

-(1992) The role of phosphorus in rhyolitic liquids as determinedfrom the homogeneous iron redox equilibria. Contributions to Miner-alogy and Petrology, in press.

Harris, R.K. (1983) Nuclear magnetic resonance spectroscopy. Pitman,Marshfi eld, Massachusetts.

Hess, P.C. (1980) Polyrnerization model for silicate melts. In R.B. Har-graves, Ed., Physics of magmatic processes, p. l-48. Princeton Uni-versity Press, Princeton, New Jersey.

Hess, P.C., Horzempa, P., and Rutherford, M.J. (1989) Fraclionation ofApollo I 5 KREEP basalrs (abs.). Lunar and Planetary Science, 20, 408-409.

Kirkpatrick, R.J. (l9EE) MAS NMR spectroscopy of minerals and glasses.In Mineralogical Society of America Reviews in Mineralogy, 18, 341-403.

Kirkparrick, R.J., Dunn, T., Schramm, S., Smith, K.A., Oestrike, R., and

GAN AND HESS: PHOSPHATE SPECIATION IN GLASSES

506 GAN AND HESS: PHOSPHATE SPECIATION IN GLASSES

Turner, F. (1986) Magic-angle sample-spinning nuclear magnetic res-onance spectroscopy ofsilicate glasses: A review. In G.E. Walrafen andA.G. Revesz, Eds., Structure and bonding in noncrystalline solids, p.

303-327. Plenum Press, New York.Kosinski, S.G., Krol, D.M., Duncan, T.M., Douglass, D.C., MacChesney,

J.B., and Simpson, J.R. (1988) Raman and NMR spectroscopy of SiO,glasses co-doped with Al,O, and P,Or. Journal of Non-Crystalline Sol-ids. 105.45-52.

Kushiro, L (1975) On the nature ofsilicate melts and its significance in

magma genesis: Regularities in the shift ofliquid boundaries involving

olivine, p1'roxene, and silica minerals. American Journal of Science,275,4t t -431.

Lacey, E.D. (1968) Configurational change in silicates with particular ref-

erence to network structures. Acta Crystallographica, 1E, 14l-l 50.London, D. (1987) Internal diferentiation ofrare element pegmatites:

EFects of boron, phosphorus and fluorine. Geochimica et Cosmochimi-ca Acta. 51. 4o3-42o.

Matson, D.W., Sharma, S.K., and Philpotts, J.A. (1983) The structure of

high-silica alkali-silicate glasses-A Raman spectroscopic study. Jour-nal of Non-Crystalline Solids, 58, 323-352.

McMillan, P. (1984) Structural studies of silicate glasses and melts-Ap-plications and limitations of Raman spectroscopy. American Miner-

alogist,69, 622-644.McMillan, P., and Piriou, B. (1982) The structures and vibrational spectra

of crystals and gJasses in the silica-alumina system. Journal of Non-

Crystalline Solids, 53, 279-298.McMillan, P., Piriou, 8., and Navrotsky, A (1982) A Raman spectro-

scopic study of glasses along the joins silica-calcium aluminate, silica-

sodium aluminate, and silica-potassium aluminate' Geochimica et

Cosmochimica Acta. 46. 2O2l -2037

Montel, J.-C. (1986) Experimental determination of the solubility of Ce-

monazite in SiO'-AlrO3-KrO-NarO melts at 800 "C, 2 Kb, under H'O-

saturated conditions. Geology, 14, 659-662.Mysen,8.O., Virgo, D, and Scarf, C.M. (1980) Relations between the

anionic structure and viscosity ofsilicate melts, a Raman spectroscopicstudy. American Mineralogist, 65, 690-710.

Mysen, B.O., Virgo, D., and Kushiro, I. (l98la) The structural role of

aluminum in silicate rnelts-a Raman spectroscopic study at I atmo-

sphere. American Mineralogist, 66, 67 8-701.Mysen, B.O., Ryerson, F.J., and Virgo, D. (l98lb) The structural role of

phosphorus in silicate melts. American Mineralogist, 66, 106-117 '

Mysen, B.O., Virgo, D., and Seifen, F.A. (1982) The structure of silicatemelts: Implications for chemical and physical properties of natural

magma. Reviews ofGeophysics, 20, 353-383.-(1985) Relationship between properties and structure of alumi-

nosilicate melts. American Mineralogist, 69, 834-847 .Naski, G., and Hess, P.C. (1985) SnO' solubility: Experimental results in

peraluminous and peralkaline high silica glasses. Eos, 66, 412.

Nelson, C., and Tallant, D.R. (1984) Raman studies of sodium silicate

glasses with low phosphate contents' Physics and Chemistry ofGlasses,

25,3r-38.Nelson, C., Furukawa, T., and White, W.B. (1983) Transition metal ions

in glasses: Network modifiers or quasi-molecular complexes. Materials

Research Bulletin, 18, 959-966.Risbud, S.H., Kirkpatrick, R.J., Taglialavore, A.P., and Montez, B. (19E7)

Solid state NMR evidence for 4-, 5-, and 6-fold aluminum sites in

roller-quenched SiOz-AlrO. glasses. Journal of American Ceramic So-

ciety, 70, Cl0-C12.Ryerson, F.J., and Hess, P.C. (1978) Implications of liquid-liquid distri-

bution coefficients to mineral-liquid partitioning. Geochimica et Cos-

mochimica Actz, 42, 921-932.-(1980) The role of PrO, in silicate melts. Geochimica et Cosmo-

chimica Acta, 44, 6ll-624.Sato. R.K., McMillan, P.F., Dennison, P., and Dupree, R. (1991) A struc-

tural investigation of high alumina glasses in the CaO-AlrOr-SiO, sys-

tem via Raman and magic angle spinning nuclear magnetic resonance

spectroscopy. Physics and Chemistry of Glasses, 33, 149-1 56.

Sharma, S.K., Virgo, D., and Mysen, B.O. (1978) Structure of glasses and

melts of NarO'xSiO' (x : l, 2, 3) composition from Raman spectros-

copy. Carnegie Institution of Washington Year Book, 77 , 649-652'

Tallant. D.R., and Nelson, C. (1986) Raman investigation of glass struc-

ture in the Na,O-SiOr-P,Or-Al,O3 system. Physics and Chemistry of

Glasses,2T (2) ,75-79.Verweij, H. (1981) Raman study of arsenic-containing potassium silicate

glasses. Journal ofthe American Ceramic Society, 64, 493-498'

Virgo, D., Seifert, F., and Mysen, B.O. (19?9) Three dimensional network

structures of glasses in the systems CaAlrOo-SiOr, NaAlOtSiOr' Na-

FeOr-SiOr, and NaGaOlSiO, at I atm. Carnegie Institution of Wash-

ington Year Book, 78, 506-5 l l.Virgo, D., Mysen, 8.O., and Kushiro, I (1980) Anionic constitution of

silicate melts quenched at I atm from Raman spectroscopy: Implica-

tions for the structures ofigneous melts. Science, 2O8, l37l-1373.

Visser. W.. and Koster van Groos, A.F. (1979) Effects of PrO5 and TiO,

on the liquidJiquid equilibria in the system KrO-FeO-Al,O.-SiO'.

American Journal of Science, 27 9, 97 O'989.Watson, E.B. (1976) Two liquid partition coemcients: Experimental data

and geochemical implications. Contributions to Mineralogy and Pe-

trofogy, 56, ll9-134.-(19791Zircon saturation in felsic liquids: Experimental data and

applications to trace element geochemistry. Contributions to Mineral-

ogy and Petrology, 70, 407'419.Wyllie, P.J., and Tuttle, D.F. (1964) Experimental investigation of sili-

cates systems containing two volatile components. III. The effects of

SOr,P2O5, HCI and LirO in addition to H,O on the melting tempera-

tures of albite and granite. American Journal of Science, 262, 9 30-9 39'

M,c.NuscnrFr RECEIVED ApnIr- 18, l99lMexuscnrrr ACCEPTED Jem;enY 6, 1992