American Mineralogist, Volume 70, pages 88-105, 1985

Relationships between properties and structure of aluminosilicate melts

euo FnIernlcH A. SBrPenrrBrdrn O. MvseN, DevIo Vrnco

Geophysical Laboratory, Carnegie Institution of Washington

Washington, D'C. 20008

Abstract

The structure of melts in the composition join Na2Si2O5-NaaAlzOs has been evaluated by

Raman spectroscopy of melts quenched at 1 atm from 1600'C, at 30 kbar quenched from

1600'C, and at 30 kbar quenched from 1400"C. The Al3+ is in tetrahedral coordination.Under all conditions, this tetrahedrally coordinated Al3+ displays strong preference for

three-dimensional network units in the melts. The relative abundance of three-dimensional

network units increases, therefore, with increasing Al/(Al+Si) of the melt even though its

bulk NBO/T is unchanged (NBO/T, nonbridging oxygen per tetrahedrally coordinatedcation). This increase is compensated for by an increase in the relative abundance of

anionic units with NBO/T > 2. The proportion of units with NBO/T = I decreases. These

efects become more pronounced with increasing pressure at the same temperature.For basaltic liquids with similar NBO/T and A1/(Al+Si) the type of charge-balancing

cation for tetrahedrally coordinated Al3+ and the types of network-modiffing metal cations

control the relative importance of feldspar and olivine on the liquidus at 1 atm. Increasing

M+/(M++M2*) results in an increase in the relative abundance of three-dimensionalaluminosilicate units and a decrease in SiOX- units in the liquid. Thus, the liquidus volume

of three-dimensional aluminosilicate minerals (feldspar) will increase as the alkali/alkalineearth increases. This trend is borne out by experimental phase-equilibrium measurementsat I atm on natural rock compositions.

Introduction

Aluminum and silicon are the two most importantnetwork-forming cations in silicate melts. A systematicstudy of the structural role of aluminum in silicate melts istherefore necessary to the understanding of the structureof natural magmatic liquids.

Structure models of aluminosilicate melts represent an

I Present address: Mineralogisch-Petrographisches Institutund Museum, Universitat Kiel, Kiel, Federal Republic of Ger-many.

2 NBO/T; nonbridging oxygens per tetrahedrally-coordinatedcations. As shown, for example, by Mysen et al. (1982a), thisratio is a measure of degree of polymerization of a silicate meltand can be calculated from its bulk chemistry provided that (l)the types and proportions of tetrahedrally-coordinated cationsare known and that (2) "free" oxygens (in the sense ofToop andSamis, 1962) do not exist. Free oxygen (O2-) has been docu-mented only in melts that are less polymerized than that ofpyrosilicate (see data and discussion in Bottinga et al., 198 I ; andMysen et al., 1982a, and references therein). Even if calculatedwith the unrealistic assumption that only Si4* is in tetrahedralcoordination in naturally-occurring magmatic liquids, these liq-uids are more polymerized than that of metasilicate. Thus, freeoxygens are extremely unlikely, and the oxygen budget can bedescribed in terms of only bridging (BO) and nonbridging (NBO)oxygens. In order to assign the tetrahedrally-coordinated cations

0003-004x/85/0102-0088$02.00 EE

improvement over those deduced by extrapolation fromdata in simple binary metal oxide-silica systems. Anexample of the limitation of extrapolations of binarymetal oxide-silicate melt structure data to aluminousmelts is that such models (e.g., Mysen et al., 1982a) donot explain the large olivine liquidus field in many basalticmagmas (Kirkpatrick, 1983). Generally, binary metaloxide-silica melts with bulk melt NBO/T2 = I do not

(T-cations), structural information is required. In addition to

Si4*, the most likely T-cations in silicate melts compositionallyrelevant to natural magmatic liquids are Al3*, P5*, Tia* and

Fe3*. All available structural data show that provided that there

are sufficient M* and M2* metal cations available for charge-

balance, Al3* is in tetrahedral coordination, at least at I atmpressure (e.g., Taylor and Brown, l979a,b; McMillan and Piriou,1983 ; McMillan et al., 1982 ; Navrotsky et al., l9E2; Mysen et al.,

l98la; Seifert et al., 1982a). Both P5* and Tia* are T-cations in

melts as indicated by both structural and liquidus phase equilibri-um information (Kushiro, 1975; Ryerson and Hess, l9E0; Visser

and Koster van Groos, 1979; Mysen et al., 1980b' l98lb' l9E2a).Ferric iron commonly is in tetrahedral coordination with the

same charge-balance constraints as for Al3* (e.g., Fox et al.'

1982; Virgo et al., 19E2, 1983; Mysen and Virgo, 1983; Mysen et

al., 1984) with possible limitations imposed by low Fe3*/IFe(Virgo et al. , 1983; Mysen et al. , 1984).

Detailed structural considerations such as those summarizedabove enter into the calculation of NBO/T. Given these pre-

MYSEN ET AL.: STRUCTURE OF ALUMINOSILICATE MELTS E9

contain significant amounts of orthosilicate units. Basal_tic melts with similar bulk melt NBO/T apparently docontain orthosilicate units, in view of the fact that olivinecommonly precipitates on their liquidus at I atm (Kirk_patrick, 1983). A similar conclusion may be drawn fromthe "quasi-crystalline" model of silicate melts proposedby Burnham (1974, l98l). It may be suggested that theexpanded olivine liquidus field observed in natural basal-tic liquids is related to the role of aluminum in thestructure of these more complex silicate melts.

compositional range (<33 mole Vo metal oxide) wherethree-dimensional network units occur in the melt (Mysenet al., 1982a). The Al3+ = Si4+ substitution in fullypolymerized (NBO/T : 0) aluminosilicate melts results ina rapid decrease in the melt viscosity and activationenergy of viscous flow as the bulk melt AU(Al+Si)increases (Riebling, 1966). The values ofthese two prop-erties are significantly less dependent on AV(AI+Si) thegreater the NaiAl of the melts (see Riebling, 1966; Bock_ris et al., 1955). Moreover, these same two physicalproperties become increasingly sensitive to pressurechanges with increasing AU(Al+Si) in aluminosilicatemelts with three-dimensionally interconnected srructure(Kushiro, 1980). In depolymerized aluminosilicate melts,aluminum may exhibit a preference for specific structuralunits (Mysen et al., l98la) so thar the Al/(Al+Si) in theunits differ from each other. Such a structural configura-tion is likely to affect the rheological behavior of themelts. Furthernore, this suggested distribution of Al3+between the structural units is likely to be pressure,temperature and bulk composition dependent. It is there_fore fundamentally important to evaluate the role of Al3*in silicate melts whose structures resemble those ofmagmatic liquids in order to understand the pressure,temperature and compositional dependence of these meltproperties.

mises, one may calculate bulk melt NBOIT from the expressions(Mysen et al., 1982a, 1984):

N B O / T = ( 4 x T - 2 x O ) n = j n M i * / T ,i - I

where M'* represent network-modifying cations. Their sum isobtained after subtraction of the proportion of metal cationsnecessary for charge-balance of Al3* and Fe3*.

Thus, with these considerations in mind the bulk melt NBO/Tcan be used as a compositional and structural indicator of silicatemelts. In this report, NBO/T denotes bulk melt nonbridgingoxygen per tetrahedral cation. The nbo/t is used to express thenonbridging oxygens per tetrahedral cations in individual struc-tural units.

The present study was undertaken to isolate the influ-ence of AV(Al+Si) on the melt structure in melts withNBO/T >0. Experiments were also conducted at varioustemperatures and pressures in order to study the struc-ture of silicate melts under conditions relevant to mag-matic processes in the earth and terrestrial planets.

Starting compositions

All compositions were in the system Na2O-Al2O3-SiO2(Fig. l) along the composition join sodium disilicate(NS2)-sodium dialuminate (NA2). With Al3+ in tetrahe-dral coordination these melts have bulk NBO/T : l. Thisdegree of polymerization was chosen because of itssimilarity to that of basaltic liquids (see Mysen et al.,l9E2a, for calculations). Furthermore, aluminum-free so-dium disilicate melt contains only a small proportion(<20%; Mysen et al., 1982a; Furukawa et al., l9E1) ofthree-dimensionally interconnected structural units andonly a low relative abundance of units with nbo/si >1(<20%). As a result, possible changes in their relativeabundance with increasing AU(AI+Si) of the melt may beeasily detected. Finally, the phase relations along thecomposition join chosen here (Fig. l) show that liquidusminerals change from sodium disilicate (NBO/Si : l) tosodium metasilicate (NBO/Si = 2) and finally to carne-gieite (NBO/T : 0) with increasing bulk melt Ay(Al+Si)

Fig. l. Compositions of starting materials with simplifiedfiquidus phase relations in the system Na2G-Al2O3-SiO2(wt.%io)(Osborn and Muan, 1960a). Symbols: NS2, Na2Si2O5; N2S,NaaSiOa; Ca, carnegieite (NaAlSioo); Ne, nepheline(NaAlSiOr); NS, NaSiO3; Ab, albite (NaAlSi3Or). The line withopen circles is a portion of the composition join Na2Si2O5-NaaAl2O5. Numbers along the line reflect compositions (mole VoNaaAl2O5 component replacing Na2Si2O5 component) of startingmaterials.

90 MYSEN ET AL,: STRL]CTURE OF ALT]MINOSILICATE MELTS

(see also Osborn and Muan, 1960a). Thus, inasmuch asliquidus phase equilibria reflect the structural properties(and, thus thermodynamic properties) of both the meltsand coexisting mineral(s), the liquidus equilibria in thisportion of the system Na2O-Al2O3-SiO2 may provide aclue as to how the aluminum abundance may affect thestability of depolymerized silicate minerals (e.g., meta-and orthosilicate) on the liquidus of basalt.

Experimental methods

Sodium-bearing silicate melts tend to lose Na by volatilizationat high temperatures in open sample containers. Thisexperimental complication becomes increasingly severe withincreasing Na content of the liquid (Kracek, 1930; Seifert et al.,1979a; Big$er, 1981). The problem is compounded by the factthat glasses with high sodium content [in the present study,Na2O increases from about 34 wt.Vo in Na2Si2O5 to about 42wt.Vo in NS2(NA2)30 (70 mole Vo sodium disilicate componentand 30 mole % sodium dialuminate)l are not amenable to electronmicroprobe analysis because of the volatile nature of Na underthe electron beam. In order to document the nominal meltcompositions, Seifert et al. (1979a) used combinations ofcrystallization experiments of NS2 with subsequent optical andX-ray studies, and Raman spectroscopy of quenched silicatemelts on the join NazO-SiOz as a function of intensive andextensive variables, to determine the composition of the startingmaterials. Because melts on the join Na2Si2Oy-NaaAl2O5 do notcrystallize to a phase of the same composition and thespectroscopic data a priori cannot be used to determine whetherthe nominal composition has been afected by the preparationtechnique, the melts studied here cannot be subjected to thesame tests as those on the binary join Na2O-SiO2. In view of thefact that the liquidus temperatures and sodium contents arecomparable to those on the analogous Al-free NazO-SiOz join(Osborn and Muan, 1960a), the same methods of preparationfound to be satisfactory by Seifert et al. (1979a) were employedhere.

The starting glasses were formed by initial decarbonation ofNa2CO3 in finely ground (-l pm) mixtures together with spectro-scopically pure SiO2 and Al2O3 at 800'C for 24 hr. Thetemperature was then raised slowly to 1000'C and kept there forI hr to melt the compositions. Batches of approximately 250 mgof starting material were prepared in this manner.

Experiments at I atm and 1600'C were conducted in a MoSi2-heated, vertical quench furnace. The samples were contained insealed Pt capsules of 3-mm diameter (l cm long) in order to avoidNa loss at these high temperatures (-600'C above the liquidus).Run durations were 10-15 min. The experimental charges werequenched in liquid nitrogen at the rate of about 500'C/sec.

High-pressure experiments were carried out in solid-media,high-pressure apparatus (Boyd and England, 1960) with l/2-in.-diameter furnace assemblies and sealed, 3-mm-o.d. Pt capsules.The experiments were terminated by turning off the power to thefurnace. The quenching rate was about 250"C/sec. The sampleswere held at the same pressure by increasing the hydraulicpressure on the piston during temperature quenching in order tomaintain pressure in the l-2 sec period during which the liquidsremained above their liquidus temperatures after the electricalpower was turned ofr. Otherwise, a pressure drop ofapproximately l\Vo, during which the experimental chargesremained in the superliquidus region, would be experienced.

Raman spectra were obtained with the automated Ramansystem described by Mysen et al. (l9E2b) and Seifert et al.(1982a). Approximately l-mm3 chips of glass were excited with aCoherent CR-18 Ar+-ion laser operating at24 W. Other detailsof the Raman procedure, data acquisition and storage anddeconvolution of the spectra were identical with those reportedby Seifert et al. (19E2a) and Mysen et al. (1982b)' Interestedreaders are referred to those papers for discussion of choice ofline shapes, number of lines, statistical limitations and tests ofthe quality of the fits of the deconvoluted spectra as well asrelevant aspects of the computer programs used.

Effect of quenching

The data reported here as well as most other structural data onsilicate melts have been obtained on quenched melts (glass)' It istherefore necessaxy to assess the extent to which quenching mayhave affected the structure of the material. Data on glasses andequivalent melts obtained at I atm in the system Na2G-Al2O3-SiOz (Sweet and White, 1969; Sharma et al., 1978a:' Seifert et al.'lgEla) indicate that with the quenching rates attainable invertical quench furnaces similar to that used here, there is nospectroscopic evidence, within the sensitivity of the methods,for structural diferences between a glass and its l-atm moltenequivalent. This conclusion is not meant to imply, however, thatdifferences in heat capacity, for example, of materials above andbelow their glass transition points (e.g., Navrotsky et al.' l9E0)are not due to variations in melt structure. It is only suggestedthat the features discernible with the available analytical tools donot reflect changes that may have occurred during coolingthrough the glass transition point.

No comparable data are available for glasses quenched at highpressure. It is likely that any pressure-induced structural changewould involve a decrease in volume and that the dZdP for suchtransformations are probably positive (in analogy with possible

analogous transformations in crystalline aluminosilicatesystems). As a result, isobaric temperature quenching ofpossiblepressure-induced structural changes in the melts is unlikely toresult in a back reaction to a structure that is stable at lowerpressure. In fact, isobaric temperature quenching of possible

transformations with positive dTldP shifts the sample fartheraway from the pressure and temperature conditions under whichthe transformation(s) may have occurred.

Experimental results

Raman spectra of quenched melts on the compositionjoin Na2Si2OrNaaAlzOs as a function of Al/(Al+Si)'temperature and pressure are shown in Figures 2-4. Thespectra of quenched sodium disilicate melts at all pres-sures and temperatures are similar to each other and tothose reported by others for the same composition(Brawer and White, 1975; Verweii, 1979b). The high-frequency envelope (800-1200 cm-r) consists of threepolarized (900-,950- and ll00-cm-r) and two depolarized(1070- and llsO-cm-t) bands. In addition, there is anasymmetric, polarized band near 620 cm-l and a veryweak, broad band near 780 cm-r. Substitution of 5 mole%NAz (NaaAl2O5) component for NS2 [to produce bulk

composition NS2(NA2)51 results in a shift of the l07Gcm-tband to 1050-1060 cm-l coupled with an increase in itsrelative intensity (Figs. 2-4; Table l). This band may be

9 lMYSEN ET AL.: STRUCTURE OF ALUMINOSILICATE MELTS

O

ooo

E

d

N

zNrtz' 6

9 ^

9 , 2

E Eo t ra 6

= N

E qezC o o

3P€ :A Q- O

i ^G ^

o <

9 2> 9

>Eo

! ? t ra x

o xN v

= N

F <E Z9 x oo o= ;= F! t -

txE t rq g

bdE N9 <3 zo 9

. L N

: ZQ AN =

x > r; a

Y O

N ' A

o 0 =t - F

U

t

3

IE

Ea

t

I

E

I

I5

o

N

2

z. B

t/

\e

E q E E E C E 8 B XE F R 3 3 D F S d

rcud'acuqut Fr!u{

MYSEN ET AL.: STRUCTURE OF ALUMINOSILICATE MELTS92

E

P- xB -t t r

t vi q =t E

q)Eaoo

l.

-oN

ob

6 6 ,l o! t ra 6t u )

ti)

93MYSEN ET AL.: STRUCTURE OF ALUMINOSILICATE MELTS

O

r 86 rl et :

TEoI

o

J

EN

oo

' a o

l oI EI 6a u )

+.$

E

IE

I

I

bg 8--. z3Io

94 MYSEN ET AL.: STRUCTURE OF ALUMINOSILICATE MELTS

Table l. Relative intensities (Vo) of Raman bands in the high-frequency envelope between 850 and 1200 cm-'

10so/tArea 1I0o,/A!ea(I) 9so,/Area(l) 900,/Area(I) 860,/Area(I)

frequency envelope. With 20 mole Vo or more NA2component, a new band appears near 550-560 cm-l(Fies. 2-4).

lble tNA2

0

102030

tArea: sm of relat lve intensltes of 860- ' 900-. 950-, 1050-, 1l0O-

and I l50-@ - bands.

Area(l) ! Sm of relat lve lntenslt leE of 860- ' 900-, 950- and

tloo-cr-l bsdE. fbe roso-"rir-r bandt repre&nts that betcen abdt 1oo0

and 10?0 o-I , dtependlng on A1l(Al + st) .

calcuLatei l scatter lng eff lc ieFjf facbrs (Eee text) ! s i2o5z-, I .ot

sior2-, 0.5, s iro16-, o. l : , stor4-, o. l (est i@ted).

deconvoluted into two bands (near 1000 cm-' and 1070cm-r) for NS2(NA2)5 composition, whereas for higher Alcontent the statistical deconvolution procedure (Mysen etal., 1982b) results in a merger of the two bands unless oneor more line parameters (frequency, half-width or intensi-ty) are constrained (held constant) during the fittingprocedure. There is no a priori information with which tojustify such constraints, and deconvolution of this spec-tral region with constraints on the line parameters was notcontinued.

The spectrum of quenched NS2(NA2)10 melt (10 mole7a sodium dialuminate component) at I atm represents afurther evolution of the NS2 and NS2(NA2)5 spectra(Fie. 2). For the spectra quenched from 1600" and 1400'Cat 30 kbar (Figs. 3 and 4), there is a small difference, inthat the I 100-cm- I band is shifted to about 1080 cm-t, thell50-cm-l band can no longer be discerned and both a900- and an 860-cm-1 band are present. Additional alumi-num results in a further slow decrease in the frequency ofthe ll00-cm-l band, a more rapid decrease in the fre-quency ofthe band between 1070 and 1000 cm*r, a rapidincrease in the relative intensity of the band at 1000-1070cm-r and an increase in the intensities of the 860-, 900-and 950-cm-r bands relative to that of the ll00-cm-rband (Table l). The band or group ofbands near 600 cm-rshows an initial increase in the frequency of its center ofgravity and a decrease in intensity relative to the high-

Interpretation of sPectra

The spectra shown in Figures 2-4 may be interpretedon the basis ofthe spectroscopic features ofbinary metaloxide-silica melts and those of silicate melts on alumi-nate-silica joins. For Al3* it must also be establishedwhether or not aluminum is in tetrahedral coordination inthe pressure and temperature range under consideration.If Al3+ is tetrahedrally coordinated, it is important todetermine whether it is substituted randomly for Sia* inthe melt or whether Al3* may exhibit a preference forparticular anionic units.

Sodium disilic ate glass

Before the role of Al3+ is addressed, a brief summary ofthe salient points in the interpretation of the spectrum ofquenched NS2 melt is necessary. The 900-, 945- and I100-cm-r bands (Figs. 2-4) are assigned to symmetric Si-O-stretching (O- denotes nonbridging oxygen) in anionicunits with nonbridging oxygens per silicon (nbo/si) about3,2 and l, respectively (e.g., Lazarev, 1972; Verweij 'l979a,b: Furukawa et al., l9E1; McMillan et al., 1982;Domine and Periou, 1983; McMillan and Piriou, 1983;Virgo et al., 1980; Mysen et al., 1982a). In units withnbo/si = 4, the symmetric Si-O- stretch band is near 860cm-r (Verwe[j and Konljnendiik, 1976; Furukawa et al.,1981; Gibbs et al., 1981;McMillan and Piriou, 1983). Theasymmetric band or group of bands near 600 cm-r hasbeen assigned to a mixed bending and stretching vibrationof the Si-O-Si group. The frequency of this band in-creases as the Si-O-Si angle increases (Furukawa et al.,l98l). This interpretation is consistent with its positionnear 580-600 cm-r in Si2O3- groups, near 620-640 cm-lin SiO3- groups and near 700 cm-r in SizO9- groups. Theasymmetry of this band in quenched NS2 melt might thusbe due to the coexistence of more than one anionic unitin the melt. The sodium disilicate quenched melt appearsto consist, therefore, primarily of coexisting Si2O3-and SiO3- units with a minor contribution from SizO9-and SiO2 units, where the relative abundance is SizO3-> SiO3- > SiO2 > Si2O9- (see also Fig. 10, Mysen et al.,1982a).

N e tw or k- mo dify in S alu minum

It is commonly concluded that Al3+ is in tetrahedralcoordination in aluminum silicate melts with sufficientM+ or M2+ cations for charge-balance at least at I atmpressure. Only a brief discussion ofpossible spectrosco-pic consequences of network-modifying AIr* is-thereforenecessary. Coordination transformation of Al'* from anetwork former to a network modifier in a melt with

4 4 . 14 7 , l5 6 . 55 9 , 25 8 . 5

45.74 8 . 14 5 . 762.27 7 . 2

4? .61L.249.76 9 . 57 4 . 2

l . {1 . 42 . 45 , 6

r 0 . 6

3 . 2

3 . 88 . 0

r . 72 . 8

r0 ,5

o.OOI kbar, 16OOoc

7 A . 4 2 0 . 276.A 2L.97 0 . 4 2 6 . 467.5 21,95 3 . 6 2 9 . 8

30 kbar, l6oooc

? 8 . 8 r 1 . 27 5 , 3 2 2 . L7 4 . a 2 0 . 362.3 28.63 6 . 3 3 2 . 0

30 kbar, 1{OOoC

7 9 . 27 4 . 27 2 . 0

3 5 . r

2 . 06 . 0

5 . ru . 5

0

102030

0

102030

1 8 . 020.3

3 4 . 2

MYSEN ET AL.: STRI]CTURE OF ALT]MINOSILICATE MELTS 95

T2O3- G : Al3++ Si4*) stoichiometry may be expressedwith the formalized equation:

sSi2O3- + NaAl)2O3- : lgsio?- + 2Na++ 2AI3+(VD. (l)

Expressions similar to equation I may be written toinclude other anionic units. Regardless of which expres-sion is chosen, a coordination transformation of Al3* inany of the compositions will cause a rapid increase in therelative intensities of Raman bands that result from Si-O- stretch vibrations. The calculated value of NBO/T ofall the starting compositions with Al3+ assumed to be insix-fold coordination IAP+ryD] is shown in Table 2. Formelts on the join Na2O-SiO2, increasing NBO/Si fromthat of sodium disilicare (NBO/Si : l) results in a rapidincrease in the abundance of SiO3- units (see Fig. 6,Furukawa et al., 1981; and Fig. 10, Mysen etal., l9g2a)and a slower increase in relative abundance of Si2Of- andSiOX- units (Mysen et aI., 1982a). The units with nbo/siabout I or 0, respectively, disappear quickly. For a directcomparison with the Raman spectra in Figure 4 in Mysenet al. (1982a), approximately 12 mole VoNA2 componentwith aluminum as AI3*(VI) results in bulk melt NBO/Sisimilar to that of composition Na2O l.2SiO2 NBO/Si :1.7). lt is evident from the spectra in Figures 2-4 and acomparison with the spectra of melts on the join Na2O-SiO2 (see Fig. 4 in Mysen etal.,1982a) that increasing AV(Al+Si) or increasing pressure and decreasing tempera-ture do not result in an evolution of the Raman spectracomparable to that expected if Al3* were a networkmodifier (rapid increase near 950 cm-l and a slowerincrease between 860 and 9fi) cm-r). Instead, there is arapid intensity increase in the region between 10fi) and1050 cm-r with the intensities of all the other stretchbands decreasing relative to this band (Table l). Thus,there is no spectroscopic evidence for octahedrally coor-dinated aluminum in these quenched melts under anv ofthe conditions studied.

Aluminum in tetrahedral coordination

In order to establish the location of tetrahedrally coor-dinated Al3* in more detail, it is necessary to evaluate thevarious possible mechanisms of Al = Si substitution inrelation to the interpretation of the Raman spectra. Thisdiscussion is best conducted with data on aluminate-silicate melts where it is established that Al3* is tetrahe-

Table 2. Hypothetical degree of polymerization (NBO/Si) ofstarting materials with Al3+ as a network modifier

N S 2 N S 2 ( r { A 2 ) 5 N S 2 ( x A 2 ) t ) N S 2 ( X A 2 ) 2 D N S 2 ( N A 2 } l O

drally coordinated in three-dimensionally interconnectedstructures. Two fundamentally different spectroscopicinterpretations of the Raman spectra of such glasses havebeen presented. McMillan et al. (1982) (see also McMillanand Piriou, 1983, and Piriou and Alain, 1979) suggested(p. 2031) that the the Si-O bond in Si-O-Al bridges can bespectroscopically treated as "Si-O stretching motionsmodified by Al3+ as a cation coordinating the oxygen,similar to alkali and alkaline earth cations in simple binarysilicate glasses (e.g., Brawer and White, 1975)." Thus, itwas suggested that in aluminosilicate melts there arediscrete Si-O stretch bands near I140 cm-t for =Si(OAl)groups, near 1000 cm-r for :Si(OAl)2 groups, near 925cm-r for -S(OAI)3 groups and near 890 cm-r forSi(OAl)4 groups (see Fig. 6, McMillan et al., 1982). Thesegroups are presumably spectroscopically analogous to the1100-, 950-, 9fi)- and 850-cm-r stretch bands involvingSi-O- in melts on alkali and alkaline earth silicate melts(see p. 2031), although McMillan er al. (1982) emphasized(p. 2032) that their spectroscopic interpretation differsfrom a geometric interpretation. This interpretation doesnot result, therefore, in a distinction between Si-O-vibrations (with which depolymerized units are identified)and Si-O" vibrations in Si-O-Al bridges (with which theposition of aluminum in the aluminosilicate melts mightbe identified). The alternative spectroscopic interpreta-tion is that the frequency of Si-O" stretch vibrations inthree-dimensionally interconnected aluminosilicate meltsdecrease continuously as a function of AU(Al+Si) of themelt. This decrease is either due to (Si,Al) coupling(Wright et al., 1966; Sharma et al., 1978b; Mysen et al.,1980a) or due to a decrease in the force constant of Si-O"stretching as a result of Si * Al substitution (Seifert etal.,1982a), or both. The relative merits ofthese spectro-scopic models must be determined before the spectra inFigures 2-4 can be interpreted.

The model suggested by McMillan et al. (1982) (see alsoMcMillan and Piriou, 1983) may be evaluated by compar-ing Raman spectra of binary metal oxide-silica glasses (asalso done by Piriou and Alain, 1979) with those of three-dimensionally interconnected aluminosilicate glasses. Itwas initially suggested (Piriou and Alain, 1979; see Mc-Millan et al., 1982, p. 2025) thar the spectrum of CaAl2SiO6(CATS) glass is similar to that of alkali metasilicate.Thus, one might interpret the spectrum of CATS glass onthe basis of a metasilicate structure. This suggestionimplies that because CATS can be considered a metasili-cate only if one half of the A13* is a network former andthe other half is a network modifier, from a spectroscopicpoint of view , 50Vo of the Al3+ must also be considered anetwork former and 50Vo a network modifier. If all Al3+ inthe CATS stoichiometry behaved spectroscopically asalkali metals or alkaline earths, the spectrum would besimilar to that of a pyrosilicate (3 nonbridging oxygensper silicon) rather than that of a metasilicate (2 nonbridg-ing oxygens per silicon) as seems to be the suggested

N B l . / S i 1 . 0 1 , 2 6 t , 5 3 2 . 0 t 2 . 1 3

MYSEN ET AL.: STRUCTT]RE OF ALUMINOSILICATE MELTS

basis for the spectroscopic interpretation of aluminosili-cate glasses by McMillan et al. (1982). Thus, the suggest-ed metasilicate basis for spectroscopic interpretation ofCATS glass (see McMillan et al., 1982, p. 2025) isinconsistent with their a priori assumption (p. 2031) inwhich the oxygen in Si-O-Al is assumed to have thespectroscopic characteristics of a nonbridging oxygen.Instead, glasses of NaAlSizOo (Jd) and Cao.5AlSi2O6(CA2S4) composition should be interpreted spectroscopi-cally on the basis of a metasilicate structure. A compari-son of spectra of CATS, CA2S4 and Jd glasses (at I atm)with those of metasilicate compositions is shown inFigure 5. The metasilicate compositions are BazSizOo(BS), Ca2Si2Oe (Wo), CaMgSi2O6 (Di) and Mg2Si2O6 (En)(from Mysen et al., 1982a). Alkali metasilicate glasses arenot included in this comparison because of the extremedifficulty in retaining alkali metals in the melts duringpreparation of sodium metasilicate glass (Kracek, 1930;Seifert et al.,1979a; Bigger, l98l) and the difficulties indocumenting the glass composition after preparation. Thespectra of alkali metasilicate glasses of Brawer and White(1975) do, however, closely resemble that of bariummetasilicate (BS) composition in Figure 5. According tothe spectroscopic model of McMillan et al. (1982), thespectra of Jd and CA2S4 (and perhaps CATS?) glass willresemble those of En, Di, Wo and BS composition at leastin Raman band frequencies. Perhaps the most character-istic band of metasilicate glasses is the mixed bending andstretching band between 620 and 640 cm-r (e.g., Etche-pare, 1972; Brawer and White, 1975; Furukawa et al.,l9E1; McMillan and Piriou, 1983). The intensity of thisband is 50Vo or more of that of the high-frequencyenvelope of the metasilicate glass spectra but is totallyabsent in the spectra of CATS, Jd and CA2S4 or anyother three-dimensionally interconnected aluminosilicateglass spectrum (Fie. 5; see also Fig. 3 in Seifert et al.,1982a; and Figs. 3 and 7 in McMillan et al., 1982).Moreover, even without any deconvolution of the high-frequency envelopes of these spectra, it is evident thatthose of the metasilicate glasses show most of theirintensity below 10fi) cm-r (primarily due to the stretchvibration from Si-O- bonds in SiOS- units), whereas thealuminosilicate spectra that should be interpreted on thebasis of metasilicate (whether CA2S4, Jd or, apparently,CATS) have most of their high-frequency envelope atlfi)O cm-r or higher. Thus, on the basis of the spectra ofaluminosilicate and metasilicate glasses on which those ofaluminosilicate glasses should be interpreted (e.g., Mc-Millan et al.,1982, p. 2031) there is no resemblance. Theinference is, therefore, that the Si-O bridging bonds in Si-GAI bridges do not have spectroscopic characters re-sembling that of the Si-O bond in Si-GM (where M is anetwork-modifying cation and the oxygen is nonbridg-ing).

Results from molecular orbital calculations (e.9., deJong and Brown, 1980) indicate that with an idealized T-O-T angle of 180', the proportion ofbonding electrons in

ffi-.rr,--T

Fig. 5. Unpolarized spectra of metasilicate compositionsBa2Si2O6 (BS), CazSizOo (Wo), CaMgSizOu (Di), MgzSizOo @n)(from Mysen et al., 1982a) and the aluminosilicate compositionsCATS (CaAlrSiO6), Jd (NaAlSizOJ and CA2S4 (CaosAlSizOe)(from Seifert et al., l9E2a).

the Si-O bridgine bond of Si-O-Si and Si-O-Al is unaf-

fected because of the charge-balance of Al3* with metal

cations. This conclusion is inconsistent with that of

McMillan et al. (1982), who suggested that a formal

charge of - I could be assigned to the oxygen in Si-O-Al

bridges in =Si(OAt), -2 in :Si(OAl)z and so on (p.

MYSEN ET AL.: STRUCTURE OF ALUMINOSILICATE MELTS 97

2035). The assignment of formal negative charge to thebridging oxygen in charge-balanced aluminosilicate sys-tems does not have any foundation in the crystal chemis-try of aluminosilicates either.

The T-O-T angle does, however, decrease with in-creasing charge-balanced Al3+ substitution (an average of7" in crystalline aluminosilicates; Gibbs et al., lggl),resulting in a decrease in the proportion of bondingelectrons in the Si-O bond (de Jong and Brown, l9g0).The proportional decrease in bonding electrons results ina reduction in the Si-O bond strength and thus the forceconstant for stretching. This anticipated lowered bondstrength is consistent with the approximately 20% de_crease in force constants calculated by Seifert et al.(1982a) for Si-O stretching in quenched melts along thejoin SiO2-NaAlSiOa. This decrease in force consranrsresults in a systematic and continuous decrease in stretchfrequencies as a function of SilAl of the tetrahedra inquestion. The conclusion does not, however, imply thatthe Si-O bond spectroscopically is like a nonbridgingbond but simply that it will weaken as the neighboringtetrahedral silicon is replaced by aluminum. Whether(Si,Al) coupling of the vibrations from T-O bridgingoxygen bonds may also affect the frequencies of thestretch vibrations cannot be determined. For the presentdiscussion it is not important to decide which of the twomechanisms operates or whether both may be involved.Rather, it merely has to be established that the frequen-cies of relevant stretch vibrations depend on the AVSi ofthe appropriate tetrahedron. Furthermore, the vibrationsthat result from nonbridging oxygens must be spectro-scopically distinct from Si-O vibrations involving Si-O-Al bridges. In view of the above discussion and otherpublished information (e.g., Brawer and White , 1975,1977; Sharma et al., 1978b; Mysen et al., l9g0a, l9g2a;Furukawa et al., 1981; Seifert et al., 1982a), this distinc-tion can be made.

The structural position of tetrahedral Al3+ inglasses on the join NS2-NA2

The Raman spectra of the quenched melts on the joinNa2Si2O5-NaaAlzOs Gigs. 2-4) show that increasing Al/(Al+SD results in rapid disappearance of the ll50-cm-tband (Si-O" stretch from three-dimensional SiO2 units)and a decrease in the frequency of the 1070-cm-t bandcoupled with an increase in its relative intensity (Table l).This increase is associated with growth of a band near 560cm-r, which most likely is due to the presence of Al-O-Al bridges in these quenched melts (Seifert et al., l9g2a:McMillan et al., 1982). The spectroscopic changes near1000-1070 cm-r probably result from increasing Al/(Al+Si) in the three-dimensional network units. Thisintensity increase may result partly from increasing rela-tive abundance of aluminous, three-dimensional networkunits and partly from increasing Ay(Al+Si) of the three-dimensional network units. The frequencies of the g60-,900- and 945-cm-r bands are not afected by increasing

AV(Al+Si) of the melt. One may conclude, therefore, thatthe structural units with nbo/si : 4. 3 and 2 do not containdiscernible amounts of Al3*. Substitution of Al3* for Sia+in these units would result in a lowering of these Ramanfrequencies. The SizO3- unit, on the other hand, probablyhas some Al3+ at all pressures and temperatures forcompositions NS2(NA2)20 and NS2(NA2)30 because theSi-O- stretch band near llfi) cm-r shows a slight de-crease in frequency relative to that of the Si-O* bandfrom SizO3- units in melts with lower Al content. At 30kbar, the frequency shift commences in spectra ofquenched melts at NS2(NA2)10, a result indicating that athigh pressures aluminum enters Si2O!- units at lower Al/(Al+Si) of the melt.

The relative intensities of the Si-O- stretch bandsfrom Siof- (860 cm-t), SirO9- (9fi) cm-') and Sio3- (945cm-l) units appear to increase relative to the total intensi-ty of all the Si-O- stretch bands (Table 1). It seems,therefore, that the relative abundance ofthe most depoly-merized units in the melts increases together with theincrease in TO2 (T : Al + Si) with increasing Ay(Al+Si)of the melt.

Solubility mechanism of aluminum in silicate melts

Aluminum occurs in tetrahedral coordination and pri-marily occupies structural units with a three-dimensional-ly interconnected network. This conclusion is in accordwith that of Mysen et al. (l98la) for melts on thecomposition joins Na2Si2O5-NaAlO2, Na2Si2O5-CaAl2Oaand CaSi2O5-CaAl2Oa at I atm and that of Dowty (1983)for aluminosilicate melts in general on theoreticalgrounds.

The preference of Al3* for three-dimensional networkunits may be rationalized by comparing these inferenceswith the crystal chemical behavior of tetrahedrally coor-dinated A13* in aluminosilicate minerals. For example,Brown and Gibbs (1970) observed that in such minerals,Sia+ would occur in tetrahedral sites associated with T-O-T bridges with the widest possible range in T-O-Tangle, whereas Al3+ shows a preference for sites associat-ed with the smallest possible range in T-O-T angle and infact the smallest value of this angle. Brown and Gibbs(1970) noted that this crystal-chemical behavior of Al3+ isseen in low albite, for example. In its crystal structure,Al3+ occurs principally in the T1(O) site, a site associatedwith a T-O-T angle 3-10' smaller than the other T-O-Tangles (Brown et al., 1969; Stewart and Ribbe, 1969). Inanalogy with crystal chemistry, the T-O-T angle in thevarious types of silicate units decreases in the orderSi2O9- > SiO3- > Si2O3- > SiO2. One may suggesr thateven though details in the geometry of coexisting Si2O9-,SiO3-, Si2O3- and SiO2 units are not known in melts, asimilar order exists (as also suggested by Furukawa et al.,l98l). In that case, the preference of A13* for the mostpolymerized unit (SiO) may be understood.

9E MYSEN ET AL.: STRUCTTJRE OF AL(TMINOSILICATE MELTS

Relative abundance of structural units

The relative abundance of individual structural units inthe melts, Xi, can be obtained from the relative intensitydata of Raman bands as given in Table l. The relativeintensity of each Raman band depends not only on therelative abundance of the structural unit containing thespecific type of bonding that gives rise to the vibration,but also on the scattering properties of this bond (e.9.,Long, 1977). Scattering properties cannot be determinedroutinely. Raman cross sections, which include geometryand polarizability of relevant bonds, may be determinedempirically from a set of Raman spectral band intensitiessuch as given in Table 1.

The procedure involves the solution of a system oflinear equations with one unknown per structural unit inthe melt. The relative intensities of the Raman bands, 41,are related to the relative abundance (mole fraction), Xi,by;

X;: Aiai, Q)

where a; is the normalized Raman cross section. In themelts under discussion, there are N structural units, ofwhich one has a three-dimensional network structure.Another equation:

N - l N - l

> 4= ) 4o.i, (3)g : l j = l

describes the mass balance of those N- I structural unitsthat contain nonbridging oxygen. The respective relativeabundances (mole fractions) and relative intensities ofRaman bands from Si-O- bonds in structural units withnonbridging oxygens are X; and A.;. For each sample,there is an additional conservation equation of the form:

N - l

2 A,": (nbo/t) = NBO/T, (4)j = l

where (nbo/t) is the number of nonbridging oxygen pertetrahedral cations in the individual structural unit andNBOIT is the bulk quantity. The compositional andstructural use of the NBO/T is discussed further in thefootnote in the introduction of this report.

The ar factors are calculated for Si-O- symmetricstretch vibrations. The three-dimensional network unitsare not included in equations 3 and 4 because its scatter-ing efrciency is most likely a function of the Al content ofthis unit and because the scattering efficiency ofbridgingin Si-O bonds is so much smaller than those of thenonbridging bonds (Phillips, 1982) that the calculatedvalue of a3p is likely to be uncertain. The X3p is deter-mined instead from the difference:

N - l

X z o : l - > X j .j : I

so that:

N N - I

)X=&D+ > x j . ( 6 )i = l j = l

With these considerations in mind, the c3 factors arecalculated from the relative intensities of the 900-, 945-and 1100-cm-l bands normalized to their total intensityfor melt compositions where there is little indication thatAl3+ is in the SizO3- units (Table l). Although it isdifficult to assess the uncertainty in the calculated valuesof a;, the results by Seifert et al. (l98lb) and Mysen et al'(l9S2b) indicate that bulk melt NBO/T-values of binaryCaO-SiOz melts calculated with this method are within5Vo ftelativd of those calculated from the CalSi. Thestatistical quality ofthe present spectra is comparable tothat of Seifert et al. (1981b) and Mysen et al. (1982b).Thus, a 5Vo relative uncertainty in the a;values obtainedhere most probably is a maximum.

The scattering efficiency of Si-O- bonds from Si2O3-units with some (unknown amount ofl Al'* substitutedfor Sia+ is not known. Its value probably would besmaller than in the absence of Sia*, but for the presentcalculations this scattering efficiency is considered con-stant. Finally, there is a small proportion of SiOl- units inthe most aluminous melts. The proportion of this unitappears to increase with increasing pressure (Table l;Figs. 2-4). The value of 451e4- cannot be calculatedaccurately because the 860-cm-r band (Si-O- stretchingin SiOX- units) can be resolved only in spectra where thefrequency of the llfi)-cm-r band (Si-O- stretching inSizO3- units) is slightly AV(Al+Si) dependent. Its valueis, however, unlikely to be greater than that of the SizO9-unit (0.33). It is assumed to be 0.3 for the presentcalculation. Although this assumption results in an uncer-tainty in the absolute value of the abundance of SiOX-units, the small value of the a; factor itself and the lowrelative intensity of the 860-cm-r band result in only asmall uncertainty in the relative abundance of the otherunits in the melts.

Results of such calculations (Fig. 6) indicate that underall conditions studied, with increasing bulk melt Al/(Al+Si) the relative abundance of units with NBO/T = Idecreases significantly. The relative abundance of three-dimensional network units increases. Mass-balance cal-culations of AV(AI+Si) and the data in Figure 6 indicatethat even wirh bulk melt AU(Al+Si) : 0.1, the AV(Al+Si)of the TO2 unit exceeds 0.6. The changes in relativeabundance of coexisting structural units agree, in princi-ple, with the suggestion of Dowty (1983) and data ofMysen et al. (19Ela). The latter data are not directlycomparable with the present results, however, because inthe experiments of Mysen et al. (l98la), aluminum wasadded in the form of NaAlO2 and CaAl2Oa' As a result,increasing AV(Al+Si) was coupled with a decrease inbulk melt NBO/T. Those data did indicate, however, that

(5)

MYSEN ET AL: STRUCTURE OF ALUMINOSILICATE MELTS

t6oo"c, Lrm

rr&u-

ro!-

rol-- rro9-

o.20

Ar/(Ar+si)

14&)"c- 3' kb.r

r.oi

r.09-

0.10 0.2oAr/(At+tiil

Fig. 6. Relative abundance of anionic units in melts on the composition join NS2-NA2 as a function of NA2 content, temperatureand pressure. Symbols on curyes: TO2, three-dimensional network units; T2O3-, structural units with average nboit : l; TO3-,structural units with average nbo/t = 2;T2Ol-, relative abundance of structural units with nbo/t = 3; TOX-, relative abundance ofun i tsw i thnbo/ t=4 .As ind ica ted in the tex t , in th ree-d imens iona lne tworkun i ts ,T=Al +S i .A t30kbarand l600 'C,T=Si+Al inmelts with AV(AI + St > 0.1. Most probably, in all other units T = Si.

99

E 8 0nto

F 6 0!

f

fiooc,

o.10

et

co

oE

00

p

to

E!

cE

Al3+ shows a distinct preference for the most polymer-ized units in the melt although this preference becomesless pronounced with increasing CalNa.

The increase in the relative abundance of three-dimen-sional network units and the decrease in abundance ofunits with nbo/si = I are compensated for by an increasein the relative abundance of units with nbo/si > I (Fig. 6).This compensation is necessary in order to maintain thesame NBO/T of the melt. In its simplest form, the relativechange in abundance may be expressed with the equation

si2o3-(r) + (NaAl)'zoflt3ioa-tzl + 2NaAlo2(0 ), (7)

where (0), (l) and (2) represent the nbo/t values of theanionic units expressed with that stoichiometry. Analo-gous expressions may be written for structural changes

involving units with nbo/t > 2 and also to account forvariable AV(AI+Si) of the structural units (principally thethree-dimensional network units).

For melts quenched from 1600"C and 1 atm the com-pensation for increased abundance of the three-dimen-sional network units is accomplished predominantly withincreasing relative abundance of SiOl- units, with only asmall increase in the proportion of more depolymerizedstructural units. At 30 kbar the adjustment of the relativeabundance of the structural units with increasing bulkmelt Al/(Al+Si) is more pronounced. For example, at 30kbar and 1400"C, only 30Vo of the melt structure iscomprised of SizO3- units in quenched NS2(NA2)30melt as compared with 40Vo at 30 kbar and 16fi)"C andabott 60Vo at 1600"C and I atm. The relative abundanceof the highly depolymerized units (SiOf- and

Al/(At+si)

100 MYSEN ET AL.: STRUCTURE OF ALUMINOSILICATE MELTS

Si2O9-) is also greater in melts quenched at 30 kbar. Thisdifference in relative stability as a function of pressureand temperature stems from (l) diminished stability ofSi2Ol- units with increasing pressure or decreasing tem-perature, or (2) an increase in AV(Al+Si) of the three-dimensional network units with increasing pressure ordecreasing temperature, or (3) both. In any case, theadditional shift to the right of an expression similar toequation 5 with increasing pressure and decreasing tem-perature is anticipated.

Aluminum distribution between immiscibletitanium- and iron-bearing melts

The strong preference of Al3+ for the most polymerizedstructural unit in the melts reported here (Fig. 6) andelsewhere (Mysen et al., 1981a) is in apparent conflictwith the suggestion of Hess and Wood (1982). Theydetermined the distribution of Al3* between coexistingimmiscible melts in the system CaO-MgO-FeG-TiO2-Al2O3-SiO2 and found that Al3" was preferred by theleast polymerized of the two melts. Hess and Wood(1982) did not consider the Al3+-distribution data ofVisser and Koster van Groos (1979) for the system K2O-FeO-Al2O3-TiO-SiO2 @. 972). In that system Al3*showed a preference for the most polymerized of twocoexisting immiscible melts (the weight ratio of Al2O3between the coexisting melts ranged from 1.1 to 5.3).

The present data set results in similar apparent discrep-ancies with the results of Hess and Wood (1982). Itappears, therefore, that rather than resulting from misin-terpretation of data, the different behavior of Alr+ in thevarious systems may be due to the very large differencesin chemical compositions of the liquids in the two studieson AI3+ partitioning and the two Raman studies. Thereare several important diferences in bulk chemistry inthese systems. The charge-balancing cations for Al3* inthe data set of Hess and Wood (1982) were Ca2*, Mg2*and possibly Fe2*, a difference that may affect the extentof AI3+ preference. It has been shown (Navrotsky et al.,1982) that the relative stability of three-dimensional alu-minosilicate complexes decreases in the order K > Na >Ca > Mg. The preference of Al3* for three-dimensionalnetwork units in the system Na2O-CaO-AlzOrSiOz be-comes less pronounced with increasing CalNa (Mysen etal., 1981a), in agreement with the conclusion of Nav-rotsky et al. (1982). The charge-balancing cation of Al3+in the study of Visser and Koster van Groos (1979) wasK+, probably resulting in more stable aluminosilicatecomplexes than in the chemical compositions studied byHess and Wood (1982). This difference in the charge-balancing cation may at least partly account for thequalitative differences between the present results, thoseof Mysen et al. (l98la) and Visser and Koster van Groos(1979) on the one hand, and those of Hess and Wood(1982) on the other.

The silicate melts in the experiments of Hess and Wood(1982) contained large amounts of TiO2 Q.65-24.68 wt.%

TiO2 in the least polymerized melt and 2.98-7.39 wtVo inthe most polymerized melt). In the experiments of Visserand Koster van Groos (1979) the TiO2 in coexistingimmiscible melts in the system K2O-FeO-AI2O3-SiO2(and inverse partitioning behavior of Al3*; was typicallyabout 50Vo of that in the experiments of Hess and Wood(1982). Thus, it appears that Alr* distribution in morecomplex melts depends on the Tia+ content as well as theavailable metal cations for Al3* charge-balance. As al-ready mentioned, neither the present data nor those ofMysen et al. (1981) were obtained with Ti4* in thesystem. The relationship between Tia* content and Ali*distribution may be at least partly explained by theobservation (Mysen et al., 1980b) that Tia+ in aluminosili-cate melts forms aluminotitanate complexes (even in thesystem NaAlSirOrTiOz). Dickenson and Hess (1983)suggested, on the basis of redox ratio determinations ofiron in the system K2O-AI2O3-SiO2-Fe-O-TiO2, thatTia* may form complexes with K+ rather than with Al3+.These redox data can, however, be explained eq-ually wellby aluminum titanate complexing because Fer+DFe ofsilicate melts is also positively correlated with AU(AI+Si)of the silicate network (Neumann et al., 1982; Seifert etal., 1982b; Mysen and Virgo, 1983; Mysen et al., 1984).Formation of aluminotitanate complexes will be accom-panied by a decrease in the Al/(Al+Si) of the aluminosili-cate network. Thus, the observations by Dickenson andHess (1983) are consistent with aluminum titanate com-plexing in the melts. The relative stabilities of aluminosili-cate and aluminotitanate complexes with different cationsfor Al3+ charge-balance is not known but is likely todepend strongly on the type of charge-balancing cationspresent (Mysen et a1., 1981a; see also Navrotsky et al.,1982). Consequently, suggestions as to the relationshipsbetween element distribution of this kind and the chemi-cal complexities introduced by Hess and Wood (1982) arepremature at this stage and should await evaluations ofthe structural roles of all components involved.

Properties of aluminosilicate melts

Aluminum is in tetrahedral coordination in aluminosili-cate melts with M+ = A13+ or M2* = 2Al3* (see alsoSeifert et al., 1982a: McMillan et al., 1982; Navrotsky etal.,1982; Mysen et al., 1982a). The Al3+ = Si4+ substitu-tion in melts where this compositional requirement is metdoes not. therefore. affect the NBO/T of the melt. Substi-tution of Al3+ for Sia* results in decreasing activity ofsilicate components in the melts simply because of thedilution of Sia*-components with possible Al3+-analogs.One might expect, therefore, that the liquidus fields ofaluminosilicate minerals (e.g., feldspars) and those of AI-free, depolymerized minerals (e.g., meta- and orthosili-cates) expand as the Al/(Al+Si) increases. This principaleffect is illustrated in Figure 7, where the liquidus bound-aries of minerals with diferent polymerization shiftto less polymerized compositions (negative change ofNBO/T) with increasing Al/(Al+Si). In this figure, the

MYSEN ET AL.: STRUCTURE OF ALUMINOSILICATE MELTS l0t0.1

OisilicateMebdli@tc- Na2O.At2O3-SiO2-- : NarO.Fe203.Si02

Metasi l icate..'-

t,,\

0.0

r+

=z+o+

o.20

0.18

0.16

0.06

-s00 -400 -300 -20.0

Percent changs, NBO/T o{ liquid

0 2 0

0.04

o.o2

-10.0

\\

Moi!tilicrra/or6orilic.tc- : CaO-Al2O3.SiO2-- : MgO.Al203-SiO2- . - : N a r O - A l 2 O 3 - S i O 2

ttt'

\ \metasi l icate \ \ orthGi l icate\ \

\

\

\

-60.0 -50.0 -40.0 -30.0 -20.0 -10.0 0.0

Percent change, NBO/T of liquid

0 .18

0 1 6

9)

- - 0.14

(!'

f, otz=f o.roo

< 0.08

0.06

=d 0 .14

=S 0 .12+

? o.ro

< 0.08

0-700 -60.0 -500 -40.0 -30.0 -20.0 -10.0 0.0

Percent change, NBO/T of liquid

0 L-70.0

Fig. 7. Relative change (percent relative to value in Al-free systems) of non-bridging oxygens per tetrahedrally-coordinated cations(NBO/T) of liquids along the tectosilicate-disilicate, tectosilicate-metasilicate and metasilicate-orthosilicate liquidus boundaries in thesystems NazO-AlzOgSiO2, Na2O-Fe2O3-SiOr, CaO-Al2O3-SiOz and MgO-Al2O3-SiO2 (derived from phase equilibrium data ofOsborn and Muan, l960a,b,c and Bowen et al., 1930). Both ferric iron and aluminum are considered in tetrahedral coordination in allmelts (for structural data relevant to the melt systems, see also Brown et al., l97E; Seifert et al.,1979b; Mysen and Virgo, l9E3;Taylor and Brown, l979a,b; Seifert et al., 1982a).

Tccto.ilicat€/mct.riliG.ts bound.rier- : CaO.Al2O3.SrO2_- : M9O-A|2O3.S|O2

\\

\

metasi l icate

102 MYSEN ET AL,: STRUCTURE OF ALUMINOSILICATE MELTS

lines are terminated at the intersection of the relevantliquidus boundaries with the liquidus boundary involvingan aluminosilicate mineral. It is noted, however, that themagnitude of the shifts depends on the type of metalcation (Mg, Ca or Na), thus perhaps suggesting that theproportions of the anionic units and the distribution ofAl3* between the units in the melts affect the position ofthe liquidus boundaries. The relative stabilities of theseminerals depend on both the thermodynamic propertiesof the minerals and the melt only partly. The properties ofthe melts will reflect the distribution of Al3* between theindividual structural units. For a melt composition ofgiven NBO/T and AV(Al+Si), the proportions of theindividual units depend, for example on the extent of Al3*preference for three-dimensional network units. For disil-icate compositions, for example (NBO/T : l), the pro-portions of aluminous, three-dimensional network unitsand the proportions of Al-free structural units with nbo/si) I increase with increasing AV(Al+Si) of the system(see also Fig. 6).

Extension of the present structural data to magmaticliquids requires consideration of Ca2* and Mg2* forcharge-balance of AI3+. As also indicated above, therelative stabilities of Ca2*- and Mg2+-charge-balancedaluminosilicate complexes are less than those of Na+-stabilized aluminosilicates (Navrotsky et al., 1982). Thisdifference is probably reflected in the less pronounceddistribution of Ca2*-charge-balanced Al3* between coex-isting structural units compared with that of Na+-charge-balanced Al3* in silicate melts (Mysen et al., 1981a). Inaddition, the type of network-modifying cation affects therelative abundance of anionic units in the melt. Forexample, in binary metal oxide-silica melts of disilicatestoichiometry, the siro3-/(sio3- + SioX- + Sioz) de-creases rapidly with increasing Zlt of the metal cationMysen et al., 1982a). In an MgSi2O5 melt, for example, itis unlikely that Si2O!- units are stable. Kirkpatrick (1983)and Kirkpatrick et al. (1981) pointed out that typicalbasaltic melts have NBO/T between 0.7 and 0.9. In mostbinary metal oxide-silica melts, compositions with thisNBO/T have little or no abundance of SiOl- units. Infact, on both the joins MgO-SiO2 and CaO-SiOz (Greig,1927; Phillips and Muan, 1959) compositions in the rangeof NBO/T comparable to basaltic liquid show liquidimmiscibility. Basaltic liquids apparently, however, dohave SiOX- units in the melt (see also Burnham, l98l),and liquid immiscibility is not particularly common.These suggested diferences in liquid structure of basalticliquids as indicated by their liquidus phase equilibria andthose of melts on binary metal oxide-silica joins may beexplained by the Al content of the basalt compositionsIAV(AI+Si) typically is about 0.25; Chayes, 19751. It canbe seen from the phase equilibria in the systems CaO-Al2O3-SiO2 and MgO-AlzOr-SiOz (Osborn and Muan,l960b,c) that this level of Ay(Al+Si) results in completedisappearance of the liquid immiscibility field.

The aluminum content of basaltic magmatic liquidsmay also help explain the apparent expansion of theolivine liquidus field compared with the orthosilicateliquidus fields of binary silicate systems and the apparentabsence of SiO

- units in the relevant melt compositions.

Regardless of whether Na*, Ca2+ or Mg2* charge-bal-ances Al3+ in tetrahedral coordination, this Al3+ probablywill show a strong preference for three-dimensional net-work units in the melt. For Ca2* or Mg2+ charge-balanced Al3+, this unit probably approximates Al2Si2O3-stoichiometry (Seifert et al., 1982a; see also McMillan etal., 1982, and Navrotsky et al., 1982, for discussion ofordering in alkaline earth aluminate-silica melts). Theexcess silica over Si/Al = I probably occurs as SiO2units. In alkaline earth systems the nonbridging oxygensin the mel ts ex is t predominant ly as SiOl- ,SizO9- and SiO3- units. Thus, increasing AV(Al+Si) ofaluminosilicate melts results in an increase in the abun-dance of these units. For Na+ charge-balanced Al3*, thepreference of aluminum for the three-dimensional net-work units is stronger than in the Ca2+ system (Mysen etal., l98la). Thus, the concentration of three-dimensionalaluminosilicate units in alkali aluminosilicate melts in-creases more rapidly with increasing AV(AI+Si) than inalkaline earth aluminosilicate melts. In this case, theabundance ofthe depolymerized units also increases. It issuggested, therefore, that the large olivine liquidus fieldin basaltic magma at least partly is related to the partition-ing of Al3* in three-dimensional network units with aconcomitant increase in the relative abundance ofdepoly-merized units such as orthosilicate.

It has also been observed. however, that some basalticliquids have plagioclase on their liquidus, whereas othersdo not (Fig. 8). In the two compositions in Figure 8(Kushiro and Thompson, 1972) both liquid compositionshave similar bulk melt NBO/T and AV(Al+Si). The meltwith plagioclase on the liquidus has, however, M*/(M*+M2*) about 60Vo higher. It is suggested that this differ-ence in liquidus mineralogy results at least partly from thestronger partitioning of Al3+ into the three-dimensionalnetwork units in the alkali-rich basalt. This preferencecauses a higher activity of aluminosilicate component inthe melt, thus resulting in plagioclase, rather than olivineon the liquidus.

An increase in the compressibility and thermal expan-sion of silicate melts on binary metal oxide-silica joinshas been related to increases in the abundance of three-dimensional network units in silicate melts (Bockris andKojonen, 1960; MacKenzie, 1960; Scarfe et al., 1979).The present data indicate that silicate melts with the sameNBO/T are likely to become more compressible as theirAl content is increased.

The pressure dependence of melt viscosity becomesmore pronounced with decreasing NBO/Si (Scarfe et al.,1979) at least on binary metal oxide-silicajoins. One mayspeculate, therefore, that as Al3* is exchanged for Sia+ in

MYSEN ET AL.: STRUCTURE OF ALUMINOSILICATE MELTS

From: Kushiro and Thompson (1972)

Comoosi t ional Features

103

()o

o

q)

F

sio2Tio2At203FerOaFeOMnOMsoCaONaroKzoM+/M++M2+Ar/(Al+si)NBO/T

A

49.681 .55

15 .49o.928.2s0 .189 . 1 7

10 .612.880 . 1 10 .1 510.2690.739

B

49.891 .35

16 .701 .557 .750 . 1 97.44

10.903.010.23o.2570.2830.690

Pressure, Kbar

Fig. 8. Portions of temperature-pressure equilibria of two basaltic liquids with different values of M*/(M* + M2*) (after Kushiroand Thompson, 1972).

1 1001300

a silicate melt (without altering b'ulk melt NBO/T) thepressure dependence ofthe melt viscosity increases. Thisincrease is due not only to the increasing bulk melt AV(Al+Si) but also to the increased relative abundance ofthree-dimensional network units as the bulk melt Al/(Al+Si) or pressure is increased. In analogy with theviscosity data of three-dimensional melts on the joinsNaAlO2-SiO2 and CaAl2Oa-SiO2 (Kushiro, 1978, 1980,l98l), the effect of substituted Al3+ is much greater forNa+-balanced than Ca2+-balanced melts. As a result, onewould expect that the pressure dependence ofthe viscosi-ty of aluminosilicate melts is more pronounced the highertheir alkali content.

Acknowledgments

Critical reviews by R. M. Hazen, P. McMillan and H. S.Yoder, Jr. are appreciated.

References

Bigger, G. M. (l9El) Na2O mobility in one-atmosphere quenchfurnaces. Progress in Experimental Petrology, Fifth ProgressReport ofResearch Supported by NERC (1978-1980), 89-90.

Bockris, J. O. and Kojonen, F. (1960) Compressibility of certainmolten alkali silicates and borates. Journal of the AmericanCeramic Society, 82, 44934497.

Bockris, J. O., MacKenzie, J. D. and Kitchner, J. A. (1955)

Viscous flow in silica and binary liquid silicates. Transactionsof the Faraday Society, 5l, 1734-174E.

Bottinga, Y., Weill, D. F. and Richet, P. (1981) Thermodynamicmodeling of silicate melts. In R. C. Newton, A. Navrotsky andB. J. Wood, Eds., Thermodynamics of Minerals and Melts, p.207-245. Springer Verlag, New York, Heidelberg, Berlin.

Bowen, N. L., Schairer, J. F. and Willems, H. W. V. (1930) Theternary system Na2SiO3-Fe2O3-SiO2. American Journal ofScience. 20. 405-455.

Boyd, F. R. and England, J. L. (1960) Apparatus for phaseequilibrium measurements at pressur€s up to 50 kilobars andtemperatures up to 1750'C. Journal ofGeophysical Research,65,741-:t48.

Brawer, S. A. and White, W. B. (1975) Raman spectroscopicinvestigation of the structure of silicate glasses. I. The binarysilicate glasses. Journal of Chemical Physics, 63,2421J432.

Brawer, S. A. and White, W. B. (1977) Raman spectroscopicinvestigation of the structure of silicate glasses. II. Soda-alkaline earth-alumina ternary and quaternary glasses. Journalof Non-Crystalline Solids, 23, 261-27E.

Brown, G. E. and Gibbs, G. V. (1970) Stereochemistry andordering in the tetrahedral position of silicates. AmericanMineralogist, 55, 1587-1607.

Brown, G. E., Gibbs, G. V. and Ribbe, P. H. (1969) The natureand variation in length of the Si-O and Al-O bonds inframework silicates. American Mineralogist, 54, 1044-1061.

Brown, G. E., Keefer, K. D. and Fenn, P. M. (1978) ExtendedX-ray fine structure (EXAFS) study of iron-bearing silicate

104 MYSEN ET AL.: STRUCTURE OF ALUMINOSILICATE MELTS

glass (abstr.) Annual Meeting, Geological Society of America,373.

Burnham, C. W. (1974) NaAlSi3OrH2O solutions. A thermody-namic model for hydrous magmas. Bulletin de Min€ralogie, 97,223-230.

Burnham, C. W. (1981) The nature of multicomponent alumino-silicate melts. Physics and Chemistry of the Earth, 13-14,197-227.

Chayes, F. (1975) Average composition of commoner Cenozoicvolcanic rocks. Carnegie Institution of Washington YearBook,74. 547-549.

de Jong, B. H. W. S. and Brown, G. E. (1980) Polymerization ofsilicate and aluminate tetrahedra in glasses and aqueous solu-tions.-Il The network-modifying effects of Mg2+, K*, Na*,Li*, H+, OH-, F-, Cl-, H2o, COz and H3O+ on silicatepofymers. Geochimica et Cosmochimica Acta, 44, 1627-193.

Dickenson, M. P. and Hess, P. C. (1983) Fe redox equilibria andthe structural role of Fr*, Tia* and Fe3+. (abstr.) EOS, 64,350.

Domine, F. and Piriou, B. (19E3) Study of sodium silicate meltand glass by infrared spectroscopy. Journal ofNon-CrystallineSolids,55, 125-131.

Dowty, E. (1983) Aluminum tetrahedra in silicate glasses: Nor-mal-coordinate vibrational analysis. (abstr.) EOS, 64, 3,14.

Etchepare, J . (1972) Study by Raman spectroscopy of crystallineand glassy diopside. In R. W. Douglas and B. Ellis, Eds.,Amorphous Materials, p. 337-3tt6. Wiley, New York.

Fox, K. E., Furukawa, T. and White, W. B. (1982) Transitionmetal ions in silicate melts. Part 2. Iron in sodium silicateglasses. Physics and Chemistry of Glasses, 23, 169-17E.

Furukawa, T., Fox, K. E. and White, W. B. (l9El) Ramanspectroscopic investigation of the structure of silicate glasses.III. Raman intensities and structural units in sodium silicategfasses. Journal of Chemical Physics, 75, 322G3237.

Gaskell, D. M. (1982) The densities and structures of silicatemelts. ln S. K. Saxena, Ed., Advances in Physical Geochemis-try, Vof . 2, p. 153-157. Springer Verlag, New York.

Gibbs, G. V., Meagher, E. P., Newton, M. D. and Swanson, D.K. (lgEl) A comparison ofexperimental and theoretical bond-length and angle-variations for minerals, inorganic solids andmolecules. In M. O'Keeffe and A. Navrotsky, Eds., Structureand Bonding in Crystals, p. 195-225. Academic Press, NewYork.

Greig, J. W. (1927) Immiscibility in silicate melts. AmericanJournal ofScience, 13, 133-155.

Hess, P. C. and Wood, M. I. (19E2) Aluminum coordination inmetaaluminous and peralkaline silicate melts. Contributions toMineralogy and Petrology, 81, 103-112.

Kirkpatrick, R. J. (1983) Theory of nucleation in silicate melts.American Mineralogist, 6E, 6-:lE.

Kirkpatrick, R. J., Kuo, L.-C. and Melchior, J. (1981) Crystalgrowth in incongruently-melting compositions: Programmedcooling experiments with diopside. American Mineralogist,66,223-242.

Kracek, F. C. (1930) The system sodium oxide-silica. Journal ofPhysical Chemistry, 34, 1583-1598.

Kushiro, I. (1975) On the nature of silicate melt and its signifi-cance in magma genesis: Regularities in the shift of liquidusboundaries involving olivine, pyroxene, and silica minerals.American Journal of Science, 275,411431.

Kushiro, I. (197E) Viscosity and structural changes of albite(NaAlSi3OJ melt at high pressures. Earth and PlanetaryScience Letters, 41, 87-91.

Kushiro, I. (1980) Viscosity, density, and structure of silicatemelts at high pressures, and their petrological applications. InR. B. Hargraves, Ed., Physics of Magmatic Processes, p. 93-l2l. Princeton University Press, Princeton.

Kushiro, I. (1981) Viscosity change with pressure of melts in thesystem CaO-Al2O3-SiO2. Carnegie Institution of WashingtonYear Book, 80, 339-341.

Kushiro, I. and Thompson, R. N. (1972) Origin of some abyssaltholeiites from Mid-Atlantic Ridge. Carnegie Institution ofWashington Year Book, 71,403-4M.

Lazarev, A. N. (1972) Vibrational Spectra and Structure ofSilicates. Consultants Bureau, New York.

Long, D. A. (1977) Raman Spectroscopy. McGraw-Hill BookCompany, New York.

MacKenzie, J. D. (l%0) Structure of some inorganic glassesfrom high temperatures. In J. D. MacKenzie, Ed,, ModernAspects of the Vitreous State, p. 188-218. Butterworths,Washington, D.C.

McMillan, P. and Piriou, B. (1983) Raman spectroscopic studiesof silicate and related glass structure: A review. Bulletin deMindralogie, 106, 57J7 .

McMillan, P., Piriou, B. and Navrotsky, A. (1982) A Ramanspectroscopic study of glasses along the joins silica-calciumaluminate, silica-sodium aluminate and silica-potassium alumi-nate. Geochimica et Cosmochimica Acta, 46,2021-2037.

Mysen, B. O. and Virgo, D. (1983) Redox equilibria, structureand melt properties in the system NazO-AlzOrSiOrFe-O.Carnegie Institution of Washington Year Book, 82,313-317.

Mysen, B. O., Virgo, D. and Scarfe, C. M. (1980a) Relationsbetween anionic structure and viscosity of silicate melts-aRaman spectroscopic study. American Mineralogist, 65, 690-7lo.

Mysen, B. O., Ryerson, F. J. and Virgo, D. (l9E0b) Theinfluence of TiOz on the structure and derivative properties ofsilicate melts. American Mineralogist, 65, l150-1165.

Mysen, B. O., Virgo, D. and Kushiro, I. (l9Ela) The structuralrole of aluminum in silicate melts-a Raman spectroscopicstudy at I atmosphere. American Mineralogist, 66,678-:101.

Mysen, B. O., Ryerson, F. J. and Virgo, D. (l98lb) Thestructural role of phosphorous in silicate melts. The AmericanMineralogist, 6, lFll7.

Mysen, B. O., Virgo, D. and Seifert, F. A. (1982a) The structureof silicate melts: Implications for chemical and physical prop-erties of natural magma. Reviews of Geophysics and SpacePhysics, 20, 353-383.

Mysen, B. O., Finger, L. W., Seifert, F. A. and Virgo, D.(1982b) Curve-fitting of Raman spectra of amorphous materi-als. American Mineralogist, 67, 6E6-696.

Mysen, B. O., Virgo, D. and Seifert, F. A. (1984) Redoxequilibria of iron in alkaline earth silicate melts: Relationshipbetween melt structure, oxygen fugacity, temperature andproperties of iron-bearing silicate liquids. American Mineral-osist,69, E34-847.

Navrotsky, A., Hon, R., Weill, D. F. and Henry, D. J. (1980)Thermochemistry of glasses and liquids in the systems CaMg-Si2O6-CaAl2Si2OrNaAlSi3Oe, SiOrCaAl2Si2OrNaAlSi3Orand SiO2-Al2O3{aO-Na2O. Geochimica et CosmochimicaActa, 44, 1409-1425.

Navrotsky, A., Peraudeau, P., McMillan, P. and Coutoures, J.-P. (1982) A thermochemical study of glasses and crystals alongthe joins silica-calcium aluminate and silica-sodium aluminate.Geochimica et Cosmochimica Acta, 46, 2039-2049.

Neumann, E.-R., Mysen, B. O., Virgo, D. and Seifert, F. A.

MYSEN ET AL.: STRUCTURE OF ALUMINOSILICATE MELTS r05

(1982) Redox equilibria of iron in melts in the system CaO-Al2O3-SiO2-Fe-O. Carnegie Institution of Washington yearBook. 81 .353-355.

Osborn, E. F. and Muan, A. (1960a) Phase Equilibrium Dia-grams of Oxide Systems. Plate 4. The System Na2O-Al2Oj-SiO2. American Ceramic Society, Columbus, Ohio.

Osborn, E. F. and Muan, A. (1960b) Phase Equilibrium Dia-grams of Oxide Systems. Plate 2. The System CaO-Al2O3-SiO2. American Ceramic Society, Columbus, Ohio.

Osborn, E. F. and Muan, A. (1960c) Phase Equilibrium Dia-grams of Oxide Systems. Plate 3. The System MgO-AlrO.-SiOr. American Ceramic Society, Columbus, Ohio.

Phillips, B. and Muan, A. (1959) Phase equilibria in the systemCaO-iron oxide-SiO2 in air. Journal of the American CeramicSociety, 42,413-423.

Phillips, J. C. (1982) Spectroscopic and morphological structureof tetrahedral glasses. Solid State Physics, i7,93-171.

Piriou, B. and Alain, P. (1979) Density of states and structuralforms related to physical properties of amorphous solids. HighTemperatures-High Pressures, ll, 407-414.

Riebling, E. F. (1966) The strucrure of sodium aluminosilicatemelts containing at least 50 mole Vo SiO2 at 1500.C. Journal ofChemical Physics, 44, 2857-2865.

Rontgen, H., Wintergaher, P. and Kammel, R. (1960) Strukturund Eigenschaften von Schlacken der Metallhuttenprozesse.II. Viskositatsmessungen an Schmelzen der System Eisenoxy-dul-Kalk-Kieselsaure und Eisenoxydul-Kalk-Tonerde-Kiesel-saure. Zeitschrift fiir Erzbergbau und Metallhuettenwesen, 8,363-373.

Ryerson, F. J. and Hess, P. C. (1980) The role of P2O5 in silicatemelts. Geochimica et Cosmochimica Acta. M.6ll-625.

Scarfe, C. M., Mysen, B. O. and Virgo, D. (1979) Changes inviscosity and density of melts of sodium disilicate, sodiummetasilicate and diopside composition with pressure. CarnegieInstitution of Washington Year Book, 78, 547-551.

Seifert, F., Virgo, D. and Mysen, B. O. (1979a) Sodium loss fromsodium metasilicate melts in CO2 and CO atmospheres. Carne-gie Institution of Washington Year Book, 78,679.

Seifert, F. A., Virgo, D. and Mysen, B. O. (1979b) Meltstructures and redox equilibria in the system Na2O-FeO-Fe2O3-Al2O3-SiO2. Carnegie Institution of Washington yearBook,78 , 511-519.

Seifert, F. A., Mysen, B. O. and Virgo, D. (l98la) Structuralsimilarity ofglasses and melts relevant to petrological process-es. Geochimica et Cosmochimica Acta. 45. 1879-1884.

Seifert, F. A., Mysen, B. O. and Virgo, D. (l98lb) euantitativedetermination of proportions of anionic units in silicate melts.Carnegie Institution of Washington Year Book, 80, 301-302.

Seifert, F. A., Mysen, B. O. and Virgo, D. (1982a) Three-dimensional network melt structure in the systems SiO2-NaAlOz, SiOr-CaAl2Oa and SiO2-MgAl2Oa. American Miner-alogist,67, 696-718.

Seifert, F. A., Mysen, B. O., Virgo, D. and Neumann, E.-R.(1982b) Ferric-ferrous equilibria in melts in the system MgO-Al2O3-SiO2-Fe-O. Carnegie Institution of Washington YearBook. 8l. 355-357 .

Sharma, S. K., Virgo, D. and Mysen, B. O. (1978a) Structure ofglasses and melts of Na2O xSiOz (x : l, 2, 3) composition fromRaman spectroscopy. Carnegie Institution of WashingtonYear Book. 77. U9-652.

Sharma, S. K., Virgo, D. and Mysen, B. O. (1978b) Structure ofmelts along the join SiO2-NaAlSiOa. Carnegie Institution ofWashington Year Book, 77,652-658.

Shartsis, L., Spinner, S. and Capps, W. (1952) Density, expan-sivity and viscosity of molten alkali silicates. Journal of theAmerican Ceramic Society, 35, 155-160.

Stewart, D. B. and Ribbe, P. H. (1969) Structural explanation forvariations in all parameters of alkali feldspar with Al/Siordering. American Journal of Science, 267 A, 444-462.

Sweet, J. R. and White, W. B. (1969) Study of sodium silicateglasses and liquids by infrared spectroscopy. Physics andChemistry of Glasses, 10,246-251.

Taylor, M. and Brown, G. E. (1979a) Structure of mineralglasses. I. The feldspar glasses NaAlSi3Os, KAlSi3Or,CaAl2i2Os. Geochimica et Cosmochimica Acta, 43, 61J7.

Taylor, M. and Brown, G. E. (1979b) Structure of mineralglasses. II. The SiOr-NaAlSiOa join. Geochimica et Cosmo-chimica Acta. 43. 1467-1475.

Toop, G. W. and Samis, C. S. (1962) Activities of ions in silicatemelts. Transactions of the Metallurgical Society AIME,224,878-887.

Verweij, H. (1979a) Raman study of the structure of alkaligermanosilicate glasses. I. Sodium and potassium metager-manosilicate glasses. Journal of Non-Crystalline Solids, 33,4t-53.

Verweij, H. (1979b) Raman study of the structure of alkaligermanosilicate glasses. II. Lithium, sodium and potassiumdigermanosilicate glasses. Journal of Non-Crystalline Solids,33, 55-69.

Verweij, H. and Konirjnendijk, W. L. (1976) Structural units inK2O-PbO-SiO2 glasses by Raman spectroscopy. Journal ofthe American Ceramic Society, 59, 517-521.

Virgo, D., Mysen, B. O. and Kushiro, I. (1980) Anionic constitu-tion of silicate melts quenched at I atm from Raman spectros-copy: Implications for the structure of igneous melts. Science,208, 137l-1373.

Virgo, D., Mysen, B. O., Danckwerth, P. and Seifert, F. A.(1982) Speciation of Fe3* in l-atm Na2O-SiO2-Fe-O melts.Carnegie Institution of Washington Year Book, 81,349-352.

Virgo, D., Mysen, B. O. and Danckwerth, P. (1983) Redoxequilibria and the anionic structure of Na2O .xSiOz-Fe-O

melts: Effect of oxygen fugacity. Carnegie Institution of Wash-ington Year Book, 82, 305-309.

Visser, W. and Koster van Groos, A. F. (1979) Effect of P2O5and TiO2 on liquid-liquid equilibria in the system KzO-FeO-Al2O3-SiO2. American Journal of Science, 279,970-988.

Wright, A. C., Rupert, J. P. and Granquist, W. T. (1968). High-and low-faujasites: a substantial series. American Mineral-ogist,53, 1293-1303.

Manuscript received, October 21 , 1983;acceptedfor publication, August 21, 1984.