American Mineralogist, Volume 75, pages 120-134, 1990
Role of Al in depolymerized, peralkaline aluminosilicate melts in the systems
Li2O-AlrO3-SiO2o NarO-AlrO3-SiO2, and K2O-AI2O3-SiO2
Bronn O. MvsnN
GeophysicalLaboratory, CarnegieInstitution of Washington, 2801 Upton Street,N.W., Washington, D.C' 20008, U.S.A'
Ansrucr
The structuresof quenched(temperaturequenchedfrom 1400'C) alkali (K, Na, and Li)
aluminosilicate melts compositionally on tetrasilicate (MrSioOr)-tetra-aluminate
[Mr(MAl)oOr] joins (M : Li, Na, and K) have been studied with Raman spectroscopy.In
these melts, AF* is a network-former charge-balancedwith K, Na, and Li, and the bulk
melt ratio of nonbridging oxygensper tetrahedrally coordinated cations (NBO/T) is 0.5.
In the Al-free end-member quenchedmelts, structural units with NBO/Si : 0 (SiOr), I
(SirO?-), and 2 (SiOl-) coexist with the mole fractions X.,o, and Xso3- increasing and
Xsoor'decreasingin the order Li > Na > K. Structural units less polymerized than that
-SiO3
were not detected. Substitution of Al for Si with concomitant alkali changeof
:
balancing enhancesthe abundanceof fully polymerized units (and units with NBO/T
2).
The reaction TrOs (2Q3) * TOr (Q) + TO, (Q') (T : Al * Si and Q'indicates the
degreeof polymerization, where n : 4 - NBO/T) describesthe anionic equilibrium in
these melts. In Al-free end-member composition melts, this equilibrium is shifted to the
right as the ionic radius of the network-modifoing cation decreases.IncreasingAl/(Al +
Si) also shifts this reaction to the right. The extent of this shift depends on the chargebalancing cation for tetrahedrally coordinated Al and increasesin the order Li < Na
<K.
INrnonucrroN
(MAlOr-SiO, and MorAlOr-SiO'); only TO, units occur'
In these compositions, both the Al content and the type
of charge-balancingcation affect melt properties such as
liquidus temperatures,viscosity,compressibility,and heat
of mixing (e.g.,Ryerson, 1985;Urbain etal., 1982;Seifert
etal., 1982;Navrotskyetal., 1985).Thetypeofnetworkmodifier affectsboth melt structure and melt properties.
For example,the heat of mixing of melts on metal oxidesilicajoinsincreases(becomeslessnegative)withincreasing Z/12 of the metal (Eliezer et al., 1978, 1979), and the
melt viscosity increases(Bockris et al., 1955).
Natural magmatic liquids are aluminous and contain
nonbridging oxygen.For example,typical tholeiitic melts
have NBO/T between0.7 and 0.9, the rangefor andesitic
melts is between 0.2 and 0.4, and that for rhyolites is
between 0.05 and 0. 15. Their All(Al + Si) values are
between0.2 and 0.3 (Mysen, 1987).Metal cations in such
melts may act in part as charge-balancingcations and in
part as network modifiers. From spectroscopicstudiesin
simple binary metal oxide-silica systems, it has been
shown that severalstructural units coexist in depolymerized aluminosilicatemelts (Virgo et al., 1980; Mysen et
al., 1980;Matson et al., 1983;Murdoch et al., 1985;Stebbins, 1987). In peralkaline aluminosilicate melts, Al is
distributed between these structural units (e.g.,Mysen et
al., 1981, 1985;Domine and Piriou, 1986;Oestrikeand
Kirkpatrick, 1988). This distribution is not random and,
in addition to pressureand temperature (Mysen et al.,
AI and Si are the principal network-forming cations in
magmatic silicate liquids. Al differs from Si, however, in
thai when in tetrahedral coordination, it requires chargebalancing with metal cations such as alkali metals or
alkaline earths. In felsic magmas (e.g., rhyolite and
rhyodacite), alkali metals u." ih" predominant chargeba'lancingcations,whereasin more mafic compositions
1e.g.,basaltic liquid), some of the alkaline earths (Mg and
Ca, and possibly also Fe2+)are also associatedwith Al3+
in tetrah;dral coordination (Mysen, 1987).
physical and chemical properties of magmatic liquids
vary widely depending on their composition. The compositional variations, in turn, govern the melt structure.
ih" d"gr"e of polymerization (nonbridging oxygen per
tetrahedrally coordinated cations;NBO/T) exertsa major
control on many of these properties. For example, in
compositions ranging from binary metal oxide-silica to
complex natural magmatic liquids, molar volume and
isothermal viscosity increasewith increasing melt polymerization(decreasingNBO/T)(e.g.,Bockrisetal.,1955;
Urbain et a|., 1982;Robinson,1969;Bottingaet al., 1983;
seealso Mysen, 1987, for review of this information).
A major structural unit in natural magmatic liquids is
the fdlt polymerizedunit (TO,; T : Al + Si with appropriate charge-balancingof tetrahedrally coordinated
Al3+). Formeltcompositionsalongaluminate-silicajoins
r20
ooo3-004x/90/01024120$02.00
MYSEN: A1IN DEPOLYMERIZED,PERALKALINE ALUMINOSILICATE MELTS
t2l
1985), probably is also a function of Al/(Al + Si) and TABLE1. Chemicalcompositionof glasses
which metal cationsare in the system(Mysen et al., l98l).
sio,
Alr03
KrO
Na.O
In order to characterizepossible relationships between
72.42
27.58
structure and properties in peralkaline aluminosilicate K54
KS4(KA4)5
67.86
2.96
29.18
melts, these compositional effects on the melt structure KS4(KA4)10 62.24
5.87
31.90
8.54
must be addressed.In this report, melts containing alkali KS4(KA4)15 57.71
33.76
11.22
35.41
metals with Z/12 between0.57 (K) and 1.7 (Li) and hav- KS4(KA4),o 53.36
KS4(KA4),5
49.31
13.48
37.21
ing All(Al + Si) between 0 and 0.4 have been employed N 5 4
80.33
19.68
N54(NA4)5
75.22
3.48
21.31
to determine the effectson melt structure.
ExpnnrvrnNTAl METHoDS
Starting materials
Starting materials were mixtures of LirCOr, AlrOr, and
SiOr; NarCO3, Al2O3,and SiOr; and KrCOr, AlrO., and
SiO, on the composition joins between the tetrasilicate
(MrSi4Or)and tetra-aluminate (Mr(MAt)4O,), where M:
alkali metal. Provided that Al3+ is in tetrahedral coordination, exchangeof Al3* for Si4+in melts on thesejoins
does not affect the bulk melt polymerization (NBO/T :
0.5).
The sampleswere taken from 50-g liquid mixtures also
used for viscosity measurementsin another study (D. B.
Dingwell, in preparation). Part of each mixture was poured
onto a cold surfaceafter the mixture had been stirred for
severalhours in a concentric cylinder viscometer. About
50 mg of the material was then returned to the experimental temperaturein a vertical quench furnaceand then
quenchedin water. Although the quenching rates cannot
be measuredaccuratelywith this method, the cooling rates
from experimental temperatures(1400 "C) to about 700
"C were about 500 "C/s. Chemical compositions of the
glassesare given in Table l.
Spectroscopicanalysis
Structural information was derived from Raman spectra of the quenched melts with an automated Raman
spectrometersystemdescribedby Mysen et al. (1982a).
The 514-nm line of an Ar* ion laser operating at 24 W
was used for sample excitation. Briefly, this system consistsof an LSI I I minicomputer interfacedwith a photon
counter and the slit and wavelengthdrives of the Raman
spectromgter.
The spectra were corrected for temperature- and frequency-dependentscattering intensity (e.g., Long, 1977)
prior to statistical analysis.The intensities in all reported
spectra are normalized to the data point of the greatest
absolute intensity. The background was subtracted from
the uncorrected spectra by least-squaresfitting of a line
(typically an exponential curve) through the data points
at frequenciesgreaterthan those where Raman scattering
was observed.As discussedin detail elsewhere(Seifert et
al., l98la, 1982; Mysen et al., 1982a),the curve fitting
was carried out on a completely statistical basis with the
method of minimization of least squares described by
Davidon (1966),Fletcherand Powell (1963),and Powell
(1964a, 1964b).Upon convergence,
the minimum value
of 12 and maximum randomnessin residual distribution
NS4(NA4),0
N54(NA4),0
NS4(NA4)s
NS4(NA4)40
LSR
LS4(LA4)1o
LS4(LA4),0
70.21
61.37
53.03
44.67
89.20
79.59
70.46
6.45
12.18
1 67 8
21.71
7.36
14.61
Liro
23.34
26.45
30.20
33.63
10.80
13.04
14.93
Note.'Chemicalanalysesfrom D. B. Dingwell(in preparation).Analyses
courtesy of Michel Pichavant by rcp.
are obtained. All line parameters(frequency,half-width,
and intensity) as well as the number of lines are independent variables in the fitting routine.
Compared with the Raman spectra of crystalline materials, the vibrational modes in amorphous materials can
be expectedto be broader owing to thermal distribution
of local geometriesand vibrational coupling. There is no
a priori test that can be performed to document the line
shape of the broad Raman bands of amorphous materials. As suggestedelsewhere(e.g.,Walrafen, 1967 Hartwig, 1977; Mysen et al., 1982a; Seifert et al., 1982;
McMillan et al., 19821,McKeown et al., 1984), the Raman spectraof amorphous materials suchas silicate glass
are best fitted with bands of Gaussianline shape.In view
of the suggestionfrom both Raman and xun spectroscopy (e.g.,Virgo et al., 1980;Mysen et al., 1982b;Matson
et al., 1983;Murdoch et al., 1985;Stebbins,1987,1988;
Brandrissand Stebbins,1988)that depolymerizedsilicate
melts can be described in terms of several coexisting
structural units (or species),the Raman spectrum of such
melts will be complex, and individual bands in the raw
spectraare cumulative envelopes.Thus, a test of various
line shapes(and perhapswhether or not individual lines
might not be symmetric) for such materials against the
overall topology of a spectrum cannot be easily accomplished. However, the high-frequencyportion of the spectrum of vitreous SiO, is comparatively simple. For this
reason, Mysen and Virgo (1985) undertook a detailed
study ofspectra ofvitreous silica and usedboth Gaussian
and Lorentzian line shapes(and mixtures of the two) in
the analysis of the spectral information. Those authors
also found that Gaussian line shapesprovided the most
satisfactoryresults.
The spectra under consideration here (see Fig. l, for
example,but seealso below for more detailed description
and analysis)are somewhatmore complex than the spectrum of SiOr, but a similar analysisis fruitful to illustrate
how some of the principal variables in the fitting procedure can affect the results. There are three variables of
major interest in the fitting procedure. These are (l) the
MYSEN: AI IN DEPOLYMERIZED, PERALKALINE ALUMINOSILICATE
t22
-l
9l
MELTS
range
Curve-fitted
^
EI
-l
5OO
700
Fig. l.
1100
900
Wavenumber,cm
1300
j
1225
1100
-1
cm
Wavenumber,
97s
1350
Raman spectnrm of composition NS4(NA4)roand curve-fitted high-frequencyportion of spectrum.
line shapes,(2) the number of lines to be fitted, and (3)
whether the minima in y2 arc global or regional.
The high-frequencyenvelope(850-1350 cm-') of the
spectmm of composition NS4(NA4)'. [90 molo/oNarSi,On
component and 10 molo/oNar(NaAl)oO, componentl (Fig.
1) will be used in this example. The fits resulting from
ten different setsofinput variables(frequency,z, full width
at half height, <o,and intensity, 1) are shown in Table 2
together with the resulting values of 12. As can be seen
from the results in that table, widely different input parameters have small or negligible effect on the fitted line
parameters,and the values of 12 from the individual fits
vary by less than 3ol0.
When fitting the spectrum with more than four lines,
statistically insignificant improvements in 1'zresult (Table 2). Moreover, the areasof the additional bands (relative to the total areaofthe high-frequencyenvelope)are
negligibly small (0.250lo+ 0.30/ofor the fifth band in the
fiveline fit and 0.250lo+ 0.50/oand 0.030/o+ 0.30/ofor the
two additional bands in the six-line fit). By reducing the
number of bands below four, the values of 12 increase
rapidly (Table 2).
From the analysisabove, it is concludedthat for Gaussian line shapes,four lines provide the best fit to the highfrequency envelope. Fewer lines greatly diminish the
goodness-of-fi1,and more lines do not improve it. Alternative line shapessuchas Lorentzian have beensuggested
and l0o/o
(e.g.,Barkerand Sievers,1975).The fits with 00/o
Lorentzian component are indistinguishable, but with a
further increasein the fraction of a Lorentzian component, the values of12 (Fig. 2) increasevery rapidly. From
similar analysesof line shapesof Raman bands of spectra
of samples in the systems CaO-SiOr, NaAlOr-SiOr,
CaAlrOo-SiOr,and MgAlrOr-SiO, (Mysen et al., 1982a;
Seifert et al., 1982), the same conclusionswere reached.
An additional test on the suitability of this fitting pro-
cedure and the suggestedsymmetric Gaussianline shape
of the Raman spectra of quenched silicate melts is possible. Becauseabundance data of individual structural
units in simple binary metal oxide-silica melts (previously only available from the Raman data) have recently
been reported from 2eSiNun analyses(Stebbins, 1987,
1988; Schneideret al., 1987; Brandriss and Stebbins,
1988),the abundancedata from Raman spectrafitted with
symmetric Gaussianlines can now be compared with the
abundancedata from 'zeSiNr'rn spectroscopy.The glasses
used in both the Raman and rvr'rnstudies were temperature-quenchedat approximately the samerates.Thus, the
structuresdeducedfrom those spectramost probably were
frozen in at approximately the sametemperature(seealso
detailed discussionbelow on the subject of temperature
and temperature-quenching).These data compare favorably (Fig. 3). In Figure 3, the original Raman spectraused
to calculatethe molar abundanceswere those reported by
Mysen et al. (1982b).It is noted, however,that the mole
fractions shown in Figure 3 were derived by a somewhat
different method than that usedin the original report, and
there are some differencesin the detailed abundances.In
view of these differences,the proceduresand uncertainties associatedwith the two methods need to be assessed
in some detail.
In both methods, relative intensity ratios of Si-O- stretch
bands (the high-frequencyenvelopesbetween about 800
and 1250 cm-') were used. In the original report (Mysen
et al., 1982b;but seealso Mysen eI al., 1982a,and Seifert
et al., l98la, for more detaileddescription),normalized
Raman crosssections, ei, wete related to normalized area
ratios from the Raman spectra,A,, of a Raman band, l,
relative to all the bands in the high-frequencyenvelope.
With the number of nonbridging oxygensin the structural
unit, r, denoted n,, the overall degreeof polymerization
of a melt could be describedwith the mass balance:
MYSEN: AI IN DEPOLYMERZED. PERALKALINE ALUMINOSIUCATE MELTS
123
1.0
0.8
.Y
o
(!
U.O
E v.*
0 102030405060
LorentzianComponent,
ot
o'o
o-
10
si02
u
4u
20
60
60
80 100
80
Lorentzian Component, 7o
Fig. 2. Relationships between percentageLorentzian component for mixed line shapes(betweenGaussianand Lorentzian)
and values of 1, of converged fits with four lines fitted to the
high-frequency envelope of spectrum of composition NS4(NA4),0.
su - *,A,n,:
NBO/T.
40
20
30
mol%
50
*..1
Fig. 3. Mole fractionsof structuralunitsin quenchedmelts
on thejoin NarGSiO, from Ramanspectroscopy
arecompared
with molefractionsof structuralunits of variousquenchedsodium silicatemeltsfrom 2eSimvrnreportedby Stebbins(1987)
(denotedS87)and Schneider
et al. (1987)(denotedSS87).Dark
symbols,from Ramanspectroscopy;
opensymbols,from ruvrn
spectroscopy.
Triangles,SirOl units; squares,SiO, units; circles,SiOl- units.
(l)
equations were solved for the normalized cross sections
mini mization.
by least-squares
An additional mass-balance,
In solving Equation I for a,, four or more equations
were used in the original method for determination of
(2)
) o,,t,: t,
Raman cross sections,and the error in the resulting values of the cross sections is 300/oor more even for the
was also used. In ora".ro solve for the cross section, dr, statistically most accurate spectra. Thus, in using such
a set of equations obtained with a number of spectraequal values to compute the mole fraction of a given structural
to or larger than the number of structural units of samples unit for an unknown sample, the relative error in the
with different bulk degreeof polymerization, NBO/T, was resulting value probably was at least between 300/oand
used(Seiferter al., l98la; Mysen et al., l98la, 1982b). 40o/o.II is suggestedthat the apparent discrepancy beIn view of the errors associatedwith the area determi- tween the abundancedata from Mysen et al. (1982b) and
nations in each spectrum (l0o/o-20o/o,relative), these those from 2eSiwr"rnreported, for example, by Stebbins
t_l
TABLE2. Results of (1) different variable input parameters and (2) different numbers of lines fitted to high-frequencyenvelope for
curve fitting of spectrum of sample NS4(NA4)10
quenchedfrom 1400 "C
Inputparameters*
Band
Outputparameters"
Range
930-980
1040-1090
1060-1110
1100-1150
n1
n2
h
n4
Average(n: 10)
Range
@ (cm 1)
t(7")
2045
40-75
30-50
40-70
s.0-15.0
25-50
20-64
20-s0
/ (cm-1)
950-954
1082-1091
1094-1 101
1119-1134
o (cm ')
I (o/"1
4042
70-76
34J6
57-63
8.6-9.6
31-39
25-31
46-56
/ (cm')
952 +
1 0 8 6+
1 1 0 0+
1 1 2 ' l+
t
No. ot lines
3
4
5
o
1
5
2
6
@ (cm ,)
I (o/")
41+1
71+2
35+1
61+2
9.2+0.3
33+5
28+3
49+4
Bange
(n: 10)
Average
21.3-22.0
21.6+ 0.3
Comment
389.8
29.5
21.3
19.9
19
New lineat 898 cm-', area:0.251o + 0.3% of total area
New lines at 885 cm 1, area : O.25Vo+ 0.5% of total area,
and at 1226 cm 1,area : 0.03% + 0.3% of total area
N€tej Symbols are /, frequency of band; o, full width at half height; ,, intensity of band relative to the maximum intensity in the fitted envelope.
t Input parameters
are the initial values of the line parametersused in the fitting routine.
'- Output parameters
are the fitted values of the line parametersupon @nvergence.
t24
MYSEN: AI IN DEPOLYMERIZED, PERALKALINE ALUMINOSILICATE
MELTS
The procedureoutlined above was usedto calculatethe
abundancesin Figure 3. As can be seen from the comparison with the data on quenched melts from 2eSiNun
Soecies
Quenched
type
melt
1
0
0
0
q
c
1
1
0
0
r
800'C
spectroscopy by Schneider et al. (1987) and Stebbins
(1987), the results from the two methods agreewell withTO, (Q4)
0.05 + 0.01 0.10+ 0.02 0.11+ 0.02 0.16+ 0.02
T,Os(Q3) 0.80 + 0.14 0.80+ 0.13 0.78+ 0.13 0.68+ 0.12
in the statistical uncertainties.
TO3(Q')
0.05 + 0.01 0.10+ 0.02 0.11+ 0.02 0.16+ 0.02
On the basisof the discussionabove and the agreement
Nofe; OriginalRamanspectra reportedby seifert et al. (1981b), but fitted
betweenresults from Raman and Nrranspectroscopy,it is
herefor the presentpurposes. Bulk-meltNBO/T : 1.0, Al/(Al + SD: 0.15
suggested,therefore,that the analysisofthe Raman spectra employed here, using symmetric lines of Gaussian
to deconvolution ofRa(1987), could be reconciled by taking into account the shape,is a reasonableapproach
man spectraof amorphous materials such as silicate glass
large errors in the original data from Raman spectrosand melt. Although this conclusiondoesnot rule out othcopy.
it is consistent with available data and will
It is evident from the analysis above that the use of er solutions,
in this report.
be
used
Raman spectroscopyto deduce concentrations of structural units requiresa more accuratecomputational meth- Effect of ternperature-quenching
od than 16a1srrggested
earlier (Seifertet al., 198I a; Mysen
From infrared and Raman spectroscopicstudies of silet al., 1982a. 1982b), and in this paper a simpler apicate
melts at high temperature (e.g., Sweet and White;
proach has been adopted. This method relies on the
1969;
Sharma et al., 1978; Seifertet al., 1981b), it has
premise that under certain compositional circumstances,
that in general,the overall spectral feabeen
concluded
only one depolymerized structural unit can be detected
the high-temperature spectra of liquids closely
tures
of
in a melt. Such spectracan then be used to calibrate inresemblethose of the quenched materials (glass)formed
tensities(areas)of diagnosticRaman bands.For example,
with quenchingrates on the order ofseveral hundred deon the join NarGSiOr, for melts equal to or more polygreesCelsiusper second.It was concluded,therefore,that
merized than that of NarO.9SiO, (NS9) (see Mysen et
structural featuresof liquids and their quenched
al., 1982b, for spectra),only units with one nonbridging average
(glass)exhibit many similarities. This concluoxygen per Si can be detected (SirO3- or Q3 units). For equivalent
recenthigh-temperaturecalorimetsion
notwithstanding,
the Raman spectrum of a melt of NS9 composition, the
(e.g.,Stebbinset al., 1984;Richet and Bottinga,
ric
data
'
intensity (area) ratio of the 1100 cm antisymmetric
1986; Navrotsky et al., 1989) have shown AC' values
stretch band (SirOl or Q3 units with I nonbridging oxywith the glass transition around 100 J/(mol'
associated
gen per Si) relative to the Si-O0stretch band from bridg(1988)summarizedhigh-temperature2eSimvn
K).
Stebbins
ing oxygenbonds, averagingnear ll50 cm-', can then
data and concluded that equilibria ofthe form
be converted to a measure of the abundance of these
:
(4)
structural units. Composition NS9 has NBO/Si
0.2,
2Q- - Q--'+ Q-*'(4 > m>o),
and the area ratio of the Raman bands p,,ool(l,,oo *
to describeanionic equilibria in silicate melts,
,4,,r0),whereA: arealis 0.23. The arearatio requiresa often used
to the right with increasingtemmay
shift
systematically
correction factor of 0.89, therefore, to be a quantitative
perature. He suggestedthat the in situ, high-temperature
expressionof the abundance of SirOl (Q3) units in the
Raman data in the system NarO-AlrOr-SiO, (Seifert et
melt. With the fitting errors in that spectrum, the error
be interpretedsimilarly.
in the correction factor is about l4o/o.In the KrO-SiO, al., l98lb) could
whether the qualitative interpretation
In
order
to
assess
system (see below for detailed spectral discussion), the
might
be quantitatively accurate,the highof
those
data
KS4 (K,O.4SiOr) spectrum exhibits no evidenceof units
Raman
spectrafrom Seifert et al. (198lb) of
temperature
more depolymerized than SirOl-, and a similar proce: I and Av(Al + SD : 0.15
NBO/T
with
bulk
liquid
dure yieldsa correctionfactor of 0.85 + 0.12.Thus, with(NS2AI5) have been deconvoluted according to the proin the error ofthe analysis,thesecorrection factorsappear
cedure outlined above with the resulting molar abunindependent of composition.
of SiO, (Q4), SirOS- (Q3), and SiOl- (Q'z)units
dances
For melts where the spectraalso indicate the presence
in Table 3. The equilibrium constant for the reshown
of other structural unites, the abundance of the second
depolymerized units can be determined by simple mass actron
balance of the nonbridging oxygens.For example, in sosi,o3 (2Q')* SioS (Q) + sio, (Qo), (s)
dium disilicate melts (bulk NBO/Si : l), SiO,, Si,O3-,
is shown as a function of temperature in Figure 4. Also
and SiOl units can be detected(e.g.,Mysenetal.,1982b;
plotted in this figure is the data point derived from the
Matson et al., 1983; Stebbins,1987).The abundanceof
quenchedmelt, with the temperatureof equilibration asSirOl- units, X"rro?-, is computed with the procedure
equal to that of the glass transition. The latter
above, and that of the SiO?- units (which have two non- sumed
was used becauseit has recently been sugtemperature
bridging oxygensper Si) is
gested (Brandriss and Stebbins, 1988) that in tempera(3) ture-quenched melts, the unit distribution at the fictive
Xsiol- : NBO/T - X",,4-)/2.
TrsLe 3. Mole fraction of structuralunits for NS2NA15comoosition melt
MYSEN: AI IN DEPOLYMERZED, PERALKALINE ALUMINOSILICATE MELTS
125
| 1oOcm-t
a
-4
I
Quenced
melt (glass)
:<
c-5
-6L
7.0
9.0
1.1
cm't
1.3
800
(Ki)
1ff x 10-1
Fig. 4. Equilibriumconstantfor the equilibrium,T,Ol- TO3- + TOr, asa functionof temperature
from thedatashown
in Table3. Also shownis thedatapoint for temperature-quenched
NS2AI5. The datapoint wasplottedat the temperature
of the
glasstransitionpoint with theassumption
thatduringquenching
the unit distributionis quenchedin at a temperature
nearthe
glasstransitiontemperature.
temperature is reflected in the glass.The fictive temperature tends to be slightly higher than that of the glass
transition, but in the absenceof accurateinformation on
its value, this temperaturehas beenapproximated by that
of the glasstransition.
A straight line through the four data points (including
that of the glassas noted in Figure 4), indicates that (l)
there is a distinctive temperature effect on the unit distribution at l-bar pressure and (2) in quenched melts
(glass),the unit distribution is frozen in near the glass
transition. From the fit ofthe four data points, enthalpy
and entropy of 34 + 6 kJ/mol and 5 + 6 J/(mol.K),
respectively,for Reaction 5 was obtained. Although these
values are somewhatuncertain, they accord with the values reported by Brandriss and Stebbins(1988) for a similar reaction obtained from high-temperature 2eSir.rt\rn
spectroscopy.Consequently, the structural data in the
present report should be consideredto be quantitatively
applicable only at temperatures near those of the glass
transition.
,l
a
o
o
Wwsrnbs,m{
Fig. 5. Unpolarized Raman spectra,corrected for temperature- and frequency-dependent scattering intensity for quenched
melts on the join Li,Si4O, (LS4FLir(LiAl)"O, (LA4) for (A) 0
mol0/0,(B) l0 mol0/0,(C) 20 molo/0,and (D) 30 mol9o LA4 component substituted for LS4.
Li,SioOe glass (Fig. 5A), there is only a shoulder near this
frequency for the NarSioO, glass (Fig. 6,4'). The Raman
spectrum of K,SioOn glass (Fig. 7A) shows little or not
intensity in this frequency region.
l1@cm''
I
500cm-r
ll'*lli"'
Rnsur,rs
The Raman spectra of the quenched melts, corrected
for temperature- and frequency-dependentscatteringintensity, are shown in Figures 5-7. The spectraof Al-free
samples (Figs. 5A-7A) exhibit a characteristically intense,asymmetric envelope with a maximum near I100
cm-', together with weaker and broader envelopes or
shoulders near 800 cm-', 600 cm ', and 500 cm-'
(marked with arrows in the figures).The spectraare similar to other equivalently polymerized alkali and alkalineearth silicate glasses(e.9., Brawer and White, 1975; Furukawa et al., l98l; Matson et al., 1983; McMillan and
Piriou, 1983,Mysen et al., 1982b).
The spectra of LirSioOn (LS4), Na,SioOn(NS4), and
KrSi4Oe(KS4) glassesdifer in detail in that with decreasing ionic radius of the metal cation (Z/r2 increases),the
envelope near 800 cm-' becomesmore intense and the
intensity of the 500 cm-'envelope decreases
relative to
that ofthe 600 cm-'envelope. Furthermore,whereasthere
is a distinct band near 950 cm-' in the spectrum of
i
*o
o
Wamurte,
cnil
Fig. 6. Unpolarized Raman spectra, corrected for temperature- and frequency-dependent scattering intensity for quenched
melts on the join NarSioO, (NS4lNar(NaAl).O, (NA4) for (A)
0 mo106,@) 5 molVo,(C) l0 mol0/0,(D) 20 mo19o,(E) 30 mol9o,
and (F) 40 molo/oNA4 component substitutedfor NS4.
r26
MYSEN: AI IN DEPOLYMERIZED, PERALKALINE ALUMINOSILICATE
1lOocnt
*#,'*T""t'
13m
Fig. 7. UnpolarizedRamanspectra,correctedfor temperascattering
intensityfor quenched
ture-andfrequency-dependent
meltson the join IlSi4O, (KS4FK,(KAI)4O,(KA4) for (A) 0
(B) 5 mol0/0,
(C) l0 mol0/0,
(D) 15 mol0/0,
(E) 20 molo/0,
mol0/0,
and (F) 25 molohKA4 componentsubstitutedfor KS4.
Substitution ofcharge-balancedAl for Si in theseglasses results in a shift in frequency of the maximum of the
I 100 cm ' envelopeto lower values (Figs. 5 and 6). Furthermore, whereasthe spectrum of Al-free KrSioOnglass
does not exhibit any spectral signaturesnear 950 cm-r,
a shoulder grows in this frequency region as (KAl)a* substitutes for Si4* (Fig. 7, B-F). Intensity increasesnear
950 cm ' with increasing Al/(Al + Si) can also be discerned visually in the spectra of the glasseson the join
Li,SioOe-Lir(LiAl)4O, (LS4-LA4) and NarSioOnNa,(NaAl)oO, (NS,I-NA4) (Figs. 5-6) although this change
in spectraltopology is partially maskedby the downward
frequencyshift of the high-frequencyenvelopeas the melts
become Al-rich. (see,in particular, Figs. 5B-5D, 6E, and
6n.
MELTS
(full width at half height,FWHH, <40 cm-') near I100
cm ' (2.)dominates the high-frequencyenvelope.Its area
relative to the overall areaofthe high-frequencyenvelope
is strongest for the KS4 composition (Fig. l0A) and
weakestfor the LS4 composition (Fig. 8A). The 950 cm '
band (2,) is strongestin the spectrum of LS4 glass and
weakestin KS4 glass.
The frequenciesof the individual bands in the highfrequency envelopesare systematicfunctions of the bulk
melt Al/(Al + Si) (Fig. I l) although the details of these
relationships depend on the alkali metal. In the system
LirSioOr-Lir(LiAl),O, (Fig. I lA), all the bands shift to
lower frequency with increasing Al/(Al + Si). In the
equivalent Na- and K-bearing systems(Fig. I lB and C),
the v., uo,and z, band frequenciesdecreasesystematically
with increasing Al/(Al + Si), bul significant Al substitution for Si is required [All(Al + Si) > 0.1] before an
effect on the frequencies of the v, ?r1dv, bands can be
observed. In the system NarSioOn-Nar(NaAl)oOn(Fig.
I lB), the frequencyofthe z, band (l 100 cm ') doesnot
changeuntil the bulk metal All(Al + Si) is between 0.1
and 0.2, in contrast to the equivalent Li-bearing system
where the z. band shifts by about l0 cm-'per 0.1 unit
changein AV(AI + Si) over the entire compositional range
(Fig. I lA). Similar relationships between Raman frequency and All(Al + Si) have been observedin the spectra of glassesin the systemK,Si.Or-Kr(KAl).O, (Fig. I lC).
It appears,though, the bulk melt All(Al + Si) > 0.2 is
required before there are significant effectson the z, and
z. band frequencies.Furthermore, the rate of frequency
decreaseof these bands with increasing All(Al + Si) is
smaller in the K-bearing than in the Na-bearing system.
The principal variations in intensity (area)among the
bands in the high-frequencyenvelope are seenfor z' and
z. (Fig. l2). In Figure 12,the equivalent areasare denoted
A , ( - 9 5 0 c m - t ) ,A , ( - 1 0 5 0c D - ' ) , A , ( - 1 1 0 0 c m ' ) , , 4 0
(- I150 cD-'), andA, (- 1200cm-'). The generaltrends
are similar for all compositions (different alkali metals)
with the intensity, lr, decreasingrelative to Aj + A4 +
A, and,4, increasing relative to At + Ir as the melts
become more aluminous. Theseintensities are, however,
more sensitive to Ay(Al + Si) in the K-bearing system
(Fig. l2C) than in either the Na- or the Li-bearing system.
Those in the Na-bearing system(Fig. l2B) are more sensitive than those in the Li-bearing system(Fig. l2A). For
example, for melts in the Al/(Al + Si) range between 0
from 0.6
and 0.2, the Ar/(A, + 44 + Ar) ratio decreases
to 0.3, from 0.55 to 0. 15, and from 0.6 to 0.05, and the
A,/(A, + Ar) ratio increasesfrom 0.2 to 0.4, from 0.1 to
0.4 and from 0 to 0.7 for the systemsLS4-LA4 (Fig. l2A),
NS4-NA4 (Fig. l2B), and KS4-KA4 (Fig. l2C), respectively.
Thus, the Raman spectra of glasseson the three joins
visually exhibit systematic changesas a function of Al/
(Al + Si). In order to addressthese spectral changesin
more detail, the high-frequencyenvelopes(betweenabout
850 and 1300 cm ' as marked in Figs. 5-7) have been
curve-fitted (as described above) into their individual
bands (Figs. 8-10). For convenience,with increasingfrequency (in the Al-free systems),thesebands are denoted
z y( - 9 5 0 c D - ' ) , v z ( - 1 0 5 0 c m - r ) , z , ( - 1 1 0 0 c t r - r ) , z o
( - I 1 5 0 c m - ' ) , a n d z , ( - 1 2 0 0c m - ' ) .
Srnucrun-lr, TNTERPRETATToN
The intensities and frequenciesof the Raman bands in
the high-frequency envelope in spectra from glasseson
The structural interpretation of Raman spectraof alkali
each of the three joins changeas systematicfunctions of silicatemelts and glasses(e.9.,Brawer and White, 1975;
bulk melt Al/(Al + Si) (Figs. ll and 12). In the spectra Verweij, 1979a, 1979b; Virgo et al., 1980; Furukawa et
of Al-free samples(Figs. 8A-l0A), a narrow, strong peak al., 198l; McMillan et al., 1982; Matson et al., 1983;
MYSEN: AI IN DEPOLYMERIZED. PERALKALINE ALUMINOSILICATE
q)
c
c)
0)
o'-
c
a
c)
o
o
MELTS
t27
e
g
N
N
6
E
o--
z
z
9625
1075
11875
Wavenumber,
cm'1
100
#
g
o
E
E
o
a
c)
E
o
c
o
o
N
N
(6
d
E
E
z
o
z
6
6
l
o
o
c)
9625
1075
11875
'!300
Wavenumber,crnl
420
s40
1060
1r80
1300
Wavenumber,crnl
Fig. 8. Curve-fitted high-frequencyregion of the spectrain Fig. 5
Mysen et al., 1982b;Domine and Piriou, 1986)generally
accord with information from 2eSiand ,?Al Nun (e.g.,
Murdoch er a1.,1985;Kirkpatrick et al., 1986;Stebbins,
1987)and X-ray data (e.g.,DeJonget al., 1981;Imaoka
et al., 1983).As originallysuggested
by Morey and Bowen
(1924) on the basis of liquidus phase relations in alkali
silicate systems,the melt structure inferred from the spectroscopic data can be considered in terms of a limited
number of coexisting structural units (Virgo et al., 1980).
These units are characterizedby their number of nonbridging oxygensper Si. The units that have been identified are fully depolymerized (NBO/Si : 0), and units
wirh NBo/si : 0, l, 2,3, and 4 (or Q", Q" Q" Q' and
Qo, respectively).
In the Raman spectraof silicate melts and glasses,the
Si-O (O : nonbridging oxygen) stretch bands in the
high-frequency envelope (Figs. 8-10) show systematic
frequency increaseswith decreasingNBO/Si of a particular structural unit. For units with NBO/Si : 4, this band
is near 850 cm-'. More polymerized structural units have
their equivalent Si-O- stretch bands near 900 cm-'
(NBO/Si : 3), 950 cm ' (NBO/Si : 2), and near 1100
cm ' (NBO/Si : l) (e.g.,Brawerand White, 1975;Virgo
et al., 1980;Furukawaet al., l98l; Mysenet al., 1982b;
McMillan, 1984).For structuralunits with NBO/Si < 4,
there is also a mixed bending and stretching mode resultingin a Raman band between600 and 700 cm-r (Furukawa et al., l98l), the frequencyof which dependson
the Si-O-Si angle in the unit. This angle, in turn, decreasesas the structural unit of interest becomes more
128
MYSEN: A1 IN DEPOLYMERIZED. PERALKALINE ALUMINOSILICATE
E 100
o
g
dt
d75
b
100
o
o
q)
o
tc
*
E
o
c
o
MELTS
X2= 40.9
!,
g
€50
o
!
.N
o
.N 25
E
(u
E
o
o
z
z0
t
:t
p
U'
o
:E
t
!
o
o
1100 1225 1350
850 975
tr
975
cm-1
Wavenumber,
cm-1
Wavenumber,
c {, nv nv
a
o
o
o-j/c
1100 1225 1350
E 1oo
(I)
o
o
tzs
.=
o
'6
o
E50
eso
'o
ro
o
o
N
t25
E
+2s
E
Z^
I
(g
z0
I
1
'@0
€o
o
cc
o
I
850 975
1100 1225 1350
8-1
850
Wavenumber,
cm-1
E 100
o
o
0)
975 1100 1225 13s0
Wavenumber,cm-1
100
5
o
x2=21.6
o
*75
'6
c
qEn
c-"
37s
.=
o
;(l)
!
c
o
E50
N^-
o
N
t25
=zr
E
E
z0
0
1
1
16
6
iU'o
6
'6v
o-1
o
(r
8 0 0 9 3 7 . 5 1 0 7 51 2 1 2 . 51 3 5 0
cm-1
Wavenumber,
8 0 0 9 3 7 . 5 1 0 7 5 1 2 1 2 . 51 3 5 0
Wavenumber,cm-l
Fig. 9. Curve-fitted high-frequencyregion ofthe spectrain Fig. 6.
MYSEN: AI IN DEPOLYMERIZED, PERALKALINE ALUMINOSILICATE
c
c
o
o
o
()
o
o-
100
E- zs
*
a
c
o
c
!
MELTS
a
c
.9 50
s
o
N
o
+zs
E
t
E
o
z
o
z0
rd
1
t
f
E
at
o
(E
!
t
0
-1
ah
o
E,
850
Wavenumber,cm-1
-c 1 0 0
6q) 1 0 0
I
qt
g
8- 7s
*
o
t7s
o
6
C
c
9c 5 0
€50
;(l,
T'
o
.N 25
(d
E
+zs
E
2o
o
z0
1
Itt
1
(d
€g, o
EU
o
o
EI
&-1
850
Wavenumber,cm-1
E 100
o
E 100
o
co
co
1zs
3tu
'6
c
o
E50
al,
c
€50
;o,
E
o
N
N
t2s
tzs
zo
zo o
E
E
o
-6
1
t
3
E
o
o)
cc
!
f
o
o
E.
-1
975
1100 1225 1350
yenumber.cm-1
Wavenumber.
975
1100 1225
Wavenumber,cm-1
Fig. 10. Curve-fittedhigh-frequency
regionofthe spectrain Fig. 7.
r29
130
MYSEN: AI IN DEPOLYMERZED. PERALKALINE ALUMINOSILICATE
MELTS
0.6
€ o.s
f o.+
s 0.3
&.o.2
0.1
0.0
0.2
0.1
Al/(Al + Si)
0.7
0.6
o 0.5
(6 0.4
E
(! 0.3
o 0.2
0.1
0.0
0.0
0.2
Al/(Al+Si)
0.1
B lNS4+rAa
A''/(A''+A.)
+A5)
0.1
0.2 0.3
Al/(Al+ Si)
1.0
0.8
o
# 0.6
3 0.4
4 o.z
0.0
0.0
E
()
1100
0.1
0.2
Al/(Al+ Si)
0.3
Fig. 12. RelativeintensitiesofSi-O stretchbands(fromFigs.
8-10)asa functionof bulk-meltAV(AI + Si).
o
-o
E
:t
o
(!
1000
=
0.3
a.4
0.2
Al/(Al+Si)
0.1
0.5
c
KS4.KA4
V/+V
E
()
o)
o
E
(l)
1000
=
%v1
0.0
0.1
0.2
Al/(Al+Si)
0.3
Fig. I l. Frequencyshifts ofSi-O stretch bands as a function
ofbulk-meltAY(AI + Sil.
depolymerizedso that the band is near 600 cm ' for units
with NBO/Si : I and near 700 cm I for units with NBO/
Si : 3 (seealso Lazarev,1972).
In addition to Raman bands associatedwith nonbridging oxygen, Raman bands due to Si-Oo (O0 : bridging
oxygen) bonds are evident. For any structural unit that
contains bridging oxygen,a stretch band near 1050 cm-'
appears (see also lasaga, 1982). Moreover, whenever
three-dimensionally interconnected network units exist
in a melt, the Si-Oobonds from such units are manifested
in one or two antisymmetricstretchbandsbetween- I 150
(v) and - 1200 cm-' (zr). These latter two bands are not
alwaysresolved and in the spectraof quenchedmetal oxide-alumina-silica melts such as discussedin this report,
the two bands often merge into one (referred to as ro *
zr).Thesebands tend to becomedifficult to resolve statistically in many of the spectra of Al-bearing quenched
melts in part becauseof their apparently different frequency dependenceon Al/(Al + Si) and in part because
their relative intensities might be a function of the All(Al
+ Si) of the melts. In fully polymerized aluminate-silica
melts (e.g.,Seifertet al., 1982),such variations have been
used to infer that perhaps two (or more) three-dimensionally interconnectedstructural units coexist.Their relative abundanceand Al/(Al + Si) contents are functions
of the bulk melt Al/(Al + Si) and the electronicproperties
of the charge-balancingcations. Although similar structural features might also exist in the three-dimensional
structural units in the depolymerized aluminosilicate melts
MYSEN: A1 IN DEPOLYMERIZED, PERALKALINE ALUMINOSILICATE
MELTS
131
discussedhere, the spectra statistically generally do not
0.8
permit such a detailed analysis.
5 0.6
In the spectraof aluminum-free lithium, sodium, and
Eoo
potassium silicate melts, Raman bands assigned to
SiO3- (Q'), Si,Oi (Q3),and SiO, (Q') units can be iden3 o.z
tified (Figs. 5-10). The band near ll00 cm '(zr) is as0.0
signedto Si-O- stretchingin a structural unit with NBO/
Si: t 1q' or SirO!- units).The bandsnear l150-1200
0.8
cm-t (voand zr) result from the presenceof three-dimensional network units (Qo or SiO, units). The band or
5 0.6
shouldernear 950 cm ' (2,)is assignedto Si-O- stretchEoo
ing in a structural unit with NBO/Si : 2 (Q' or SiOl
3 o.z
units). The broader band near 600 cm ' is assignedto
the mixed stretch and bending modes in mixtures of
0.0
0.0 0.1 o.2 0.3 0.4 0.5
SirO3 and SiOl-. The anionic structure of Al-free tetraAl/(Al+Si)
silicate melts is, therefore, viewed as mixture of threedimensionally interconnected structural units, together
0.8
with units wirh NBO/Si: I and units with NBO/Si: 2.
5 o.o
For Al-free tetrasilicatemelts, Reaction 5 adequatelydeo
g 0.4
scribes,therefore, the anionic equilibrium. This conclusion accordswith those from other Raman and Nrrrndata
3 o.z
for melts and glassesin the same polymerization range
0.0
(e.g.,Virgo et al., 1980;Mysenet al., 1980,1982b;Mat0.0
0.1
0.2
0.3
Al/(Al+Si)
son et al., 1983;Murdoch et al., 1985;Kirkpatrick et al.,
1986;Stebbins,1987,1988).
Fig. 13. Relative abundanceof structural units as a function
Substitution of Al3+ for Sia+with concomitant Li, Na, of bulk-melt Al/(Al + Si) for the systemsindicated.
or K charge-balancinghas two principal spectral effects.
The intensity relations among the bands change,and the
frequenciesofsome ofthe diagnostic bands are affected. Kirkpatrick et al., 1986) as well as other Raman data
There is no evidencefor additional bands in the spectra (Domine and Piriou, 1986)indicating, therefore,that the
as the Al content of the melts is increased.Thus, a reac- Qo unit (NBO/Si : 0) is a principal location for Al3* in
tion analogousto Reaction 5 can be written for the an- the melts relative to Q3 (NBO/Si : l) and Q'?(NBO/Si :
ionic equilibrium in the aluminosilicate melts by replac- 2) units.
ing Si in Reaction 5 with T [where T : Sia+ + (MAl)a*;
that the preference
Mysen et al. (1981, 1985)suggested
M : alkali metall.
of Al for fully polymerized structural units could be raThe Al-dependent frequencies of the high-frequency tionalized by analogy with crystal-chemical data from
Raman bands (Figs. 8-10) may be related to Al substi- aluminosilicate minerals. In such silicates, Al shows a
tution for Si in the structural units. The zo and z, band distinct preferencefor the tetrahedral site neighboringthe
frequenciesare lowered with increasingbulk-melt All(Al
oxygen bridge with the smallest intertetrahedral angle (e.g.,
* Si), whereasthose ofthe z, and z, bands are less sen- Brown et al., 1969).Among the three coexistingstructural
sitive to Al contents (Fig. I l). The frequency decreases units dominating the structure of the aluminum-bearing
either result from (Si,Al) coupling or a decreasein the alkali tetrasilicate melts, from geometric considerations
force constant resulting from the weakening of the T-O it is most likely that the fully polymerized unit in a given
bonds as Al is substituted for Si, or both. In either case, system has the smallest average intertetrahedral angle
the frequency reductions reflect substitution of Al3* for (Furukawa et al., l98l). Thus, a preferenceby Al3* for
Si4* in tetrahedralcoordination. With lessthan l0o/o-20o/o such units would be expectedin accord with the strucAl in substitution for Si in the Na- and K-bearing sys- tural interpretation of the Raman spectra.
tems, the frequenciesof thesetwo bands do not vary with
CorvrposrrroN-ABUNDANcE RET.ATIoNS
All(Al + Si), but they do show a decreaseas the Al contents are increasedfurther. The greater sensitivity of the
The intensities of Si-O- stretch bands (2, and z.) from
SiOostretch-band(zo* zr) frequenciesas compared with the deconvoluted, high-frequencyenvelopesofthe specthose of the Si-O- stretch bands (2, and z.) might reflect tra can be usedto estimate the relative abundancesof the
a preferenceof the Al3* for the most polymerized struc- structural units in the melts as discussedabove. The retural unit (TO, or Qa) relative to structural units that sults ofthese calculations are shown in Figure 13. In all
contain nonbridging oxygen (TrO3- and TO3-, or Q3 and three systems, the abundance of TrO' (Q') units deQ'z;seealso Mysen et al., I 98 I , I 985). This conclusionis creases,and the abundancesof TO, (Q') and TO, (Q')
consistent with 'z?Aland 2eSiNMR spectroscopyof melts units increase systematically as a function of increasing
of similar compositions (e.g., Engelhardtet al., 1985; All(AI + Si) of the melt. As no other structural units can
132
MYSEN: AI IN DEPOLYMERIZED. PERALKALINE ALUMINOSILICATE
TABLE4.
MELTS
Bulk compositionaland structural informationfor tholeiite and nepheline-normativemelt composition
o
e
o
Parameter
C
o
'e
Ar/(Ar
+ sD
8.
AU(Al+Si)
E
o
NBO/T
TO3 (Q')
T,Ou(Q3)
TO, (Q.)
M*(TO,).
M'*OO,)'
Ar/(Ar
+ sD(To,)
Tholeiite
(1092 analyses)
0.267+
0.688+
0.215+
0 012+
0.687+
0.586+
0.413+
0.389+
0.028
0.257
0.057
0.002
0.094
0.081
0.081
0.048
Neohelinenormative
basalt (233 analyses)
0.274 +
0.867 +
0.252 +
0 . 0 1 5+
0.589 +
0 . 6 1 9+
0 . 3 7 9+
0.465 +
O.O32
0.272
0.061
0.009
0.095
0.126
0.115
0.057
Note.'Data base: RockfileRKNFSyS
for Cenozoicexlrusive igneousrocks
is unpublished).
compiledby Chayes(1975a,1975b; 1985 versionNrRM2
- Proportionof AF* in TO2units charge-balancedwith alkali metals (M.)
or alkalineearths (M,*).
e
o
e
ering or distortion of the structural units with increasing
Z/r2 of the metal cation may result in smaller average
*
6
differencesin intertetrahedral angle between the coexisting structural units. Becausethe Al distribution between
the structural units appearsrelated to this angle, by diminishing the angle difference betweenthe structural units,
Ali(Al+Si)
it is likely that the Al preferencefor a given structural
unit is lesspronounced.This rationale may be employed,
therefore, to explain the decreasingdegreeof dispersion
800
of Al with increasingZ/ r' of the metal cation. For a given
bulk-melt All(Al + Si), the Al probably is partitioned
ooo
$
t
more strongly into the fully polymerized network units
5 400
(TO, or Qo)as Z/r'of the metal cation decreases,as also
6
indicated by the systematicvariations in the shifts in the
F zoo
Raman frequencies as a function of Al/(Al + Si) dis0
cussedabove (Fig. I l). Becauseenhancedabundanceof
-200
TO, units also requires an increasein TO, abundance(in
order to maintain mass balance),it is suggestedthat this
AU(Al+Si)
behavior of Al is a principal driving force behind the
Fig. 14. Dispersionof structural-unitabundance
as a func- enhanced dispersion (Fig. la) as the Z/r2 of the metal
tion of bulk-meltAl/(Al + Si) for the systemsand units indi- cation is lowered.
cated.
Appr,rcarroNs
Transport properties of silicate melts (e.g., viscosity,
be identified in the melts, this observation implies that diffusivity, and conductivity) can be correlated with the
substitution of Al for Si with alkali charge-balancingshifts abundance of fully polymerized structural units (e.9.,
Reaction 5 to the right.
Dingwell, 1986)and also with the strengthof the bridging
The rate of changeof unit abundancewith Al substi- T-O bond. On the one hand, substitution of Al for Si in
tution, relative to the Al-free end-member, will be re- tetrahedral coordination will weaken the bond. This
ferred to as dispersion. The dispersion is more pro- weakening is more pronounced the smaller the chargenounced for metal cations of Z/r2 (Fig. la). Thus, the balancing cation (as suggested,for example, by the heateffect of Al substitution on the relative abundanceof the of-solution data for melts on SiOr-MAlO, joins; seeNastructural units is greatestfor the K-bearing system and vrotsky et al., 1985). In melts where Al3* substitution
least evident for the Li-bearing system.This observation, does not affect the proportions of other structural units,
in turn, may be related to the extent to which the inter- the weakeningof the T-O bonds is likely to enhancethe
tetrahedral anglesin the coexisting units depend on the diffusivity, conductivity, and fluidity of the melt. The Al
network-modifying metal cation. From crystal chemistry substitution in depolymerized silicate melts also results,
(e.g.,Liebau, l98l), sheetsand chainsin crystallinesili- however, in enhancementof the proportion of three-dicatesbecomeincreasinglypuckeredas the network-mod- mensionalnetwork units becauseof the preferenceof Al3+
ifying cation becomessmaller or more highly charged,or for the three-dimensional network unit. This preference
both. It is suggestedhere that in silicate melts, this puck- is more pronounced the larger the charge-balancingcate
o
MYSEN: AI IN DEPOLYMERIZED, PERALKALINE ALUMINOSILICATE
ion. Thus, the abundanceincreaseof three-dimensional
network units is greater the larger the charge-balancing
alkali metal. Increasedabundance of three-dimensional
network units will result in decreasingvalues of fluidity
and diffusivity. Although Al substitution for Si in fully
polymerized silicate melts results in a lowering of the
viscosity and an increasein diffusivity and conductivity,
it is, thus, possiblethat the concomitant increasein abundance of TO, units, resulting from this substitution in
depolymerized melts, could reverse these trends. The
likelihood that this reversal occurs for a given bulk-melt
degree of polymerization is greater the larger the alkali
metal.
One might speculatethat properties of silicate melts
that are sensitive to both the abundanceof three-dimensional network units in the melts and to the strength of
bridging T-O bonds (T : Si,AD are significantly dependent on the type of charge-balancingand network-modifying cation even at the sameAll(Al + Si) and M/(Al +
Si) (M : metal cation). For example, the compressibility
and thermal expansionof silicate melts are primarily governed by the abundanceof TO, units (seeTomlinson et
al., 1958, and Bockris and Kojonen, 1960, for relevant
experimental data). For aluminosilicate melts with the
same bulk-melt degreeof polymerization and the same
All(Al + Si), the values of these parameters are, therefore, likely to be greater for potassium aluminosilicate
melts than for sodium aluminosilicate melts and greater
for sodium aluminosilicate melts than for lithium aluminosilicate melts. This trend is expected becausethe
abundanceof TO, units in the depolymerized aluminosilicate melts increasesthe smaller rhe Z/12 of the metal
cation. This suggestionmay explain, for example, why
nepheline-normative basalt melts are more compressible
than tholeiite melts even though the overall Al/(Al + Si)
ratios of thesegroups of rocks are similar and the average
NBO/T of nepheline-normative basalts is slightly higher
than that of tholeiite (Table 4). In alkali basalts,a greater
fraction of the tetrahedrally coordinated Al3+ is chargebalancedby alkali metals (Na and K) than in olivine tholeiite. Furthermore, the AI/(AI + Si) ratio of the threedimensionally interconnected units (TOr) in melts of
nepheline-normativebasaltis also significantlyhigherthan
in tholeitte (Table 4). Thus, both the abundanceof TO,
units and the compressibility of these units are likely to
be greater in alkali basalt than in olivine tholeiite.
AcrNowr,oocMENTS
The manuscript was prepared while I was in residenceat the BayerischesGeoinstitut, Universitet Bayreuth,during the summer of 1988.The
support and hospitality enjoyed during this stay are appreciated.Critical
reviews by F. A. Seifert and C. T. Prewitt are greatly appreciated.
RnnnnnNcnscrrED
Barker, A.S , and Sievers,A.S. (1975) Optical studies of the vibrational
propertiesofdisordered solids. Reviewsin Modern Physics,47,1-169.
Bockris, J.O'M., and Kojonen, F. (1960) The compressibility of certain
molten alkali silicatesand borates.Journal of the American Ceramic
Society, 82, 44934497.
MELTS
I33
Bockris, J.O'M., Mackenzie,J.O., and Kitchner, J.A- (1955) Viscous flow
in silica and binary liquid silicates.Transactionson the Faraday SocieIy,5l,1734-1748.
Bottinga, Y., Richet, P., and Weill, D.F. (1983) Calculationof the density
and thermal expansioncoefficientof silicate liquids. Bulletin Min6ralogique,106,129-138.
Brandriss, M.E., and StebbinsJ F. (1988) Effectsoftemperature on the
structures of silicate liquids, '"Si NMR results. Geochimica et Cosmochimica Acta, 52, 2659-2669.
Brawer. S. A.. and White, W.B. (1975)Raman spectroscopicinvestigation
ofthe structureofsilicate glasses.I. The binary silicateglasses.Journal
of ChemicalPhysics,63, 2421-2432.
Brown, G.E., Gibbs, G.V, and Ribbe, P.H. (1969) The structure and
variation in the length of Si-O and Al-O bonds in framework silicates
American Mineralogist, 54, 1044-106l.
Chayes,F. (1975a)Averagecomposition ofthe commoner Cenozoicvolcanic rocks. Camegie Institution of Washinglon Year Book 74' 547549.
(1975b) A world data basefor igneouspetrology. CarnegieInstitution of WashinS,tonYear Book 74, 549-550.
Davidon, W.C. (1966) Variable metric method for minimization (3rd
revision). Lawrence Livermore, Argonne National Laboratory ANL
5990.
DeJong,B.H.W.S.,Keefer,K.D., and Brown,G'E. (1981)Polymerization
of silicate and aluminate tetrahedra in glasses,melts and aqueoussolutions. III-Local silicon environments and internal nucleation in silicate glasses.Geochimica et Cosmochimica Ac12',44, 1627-1643.
Dingwell, D.B. (1986) Viscosity-temperaturerelationshipsin the system
NarSirOyNaoAlrO, Geochimicaet CosmochimicaActa, 50, 1261-1265.
Domine, F., and Piriou, B. (1986)Raman spectroscopicstudy of the SiOr
AlrOr-K,O vitreous system: Distribution of silicon and secondneighbors American Mineralogist, 71, 38-50.
Eliezer,N., Howald, R.A., Marinkovic, M., and Eliezet,l' (1978) Vapor
pressuremeasurements,thermodynamic parameters,phasediagram in
the systempotassiumoxide-silicon oxide at high temperature.Journal
of ChemicalPhysics,82, 1021-1028.
Eliezer, N., Eliezer, I., Howald, R.A., Verwoll M.C., and Viswanadam,
P. (1979', The enthalpy of sodium silicate glassesand liquids. CALPHAD. 3. l-8.
Engelhardt, G., Nofz, M., Forkel, K., Wishmann, F.G., Magi, M.' Samoson, A., and Lippmaa, E. (1985) Structural studiesof calcium aluminosilicate glassesby high resolution solid state "Si and':7Al magic
angle spinning nuclear magnetic resonance.Physicsand Chemistry of
Glasses.26.157-165.
Fletcher, R., and Powell, M.J.D. (1963) A rapidly converying descent
method for minimization. Computer Journal, 6, 163-168'
Furukawa, T., Fox, K.L., and White, W.B. (1981) Raman spectroscopic
investigation of the structure of silicate glasses.III. Raman intensities
and structural units in sodium silicate glasses.Journal of Physical
Chemistry,15, 3226-3237.
Hartwig, C.M. (1977) The radiation-induced formation of hydrogen and
deuterium compoundsin silica as observedby Raman scattering.Journal ofChemical Physics,66, 227-239.
Imaoka, M., Hasegawa,H., and Yasui, I. (1983) X-ray diffraction study
ofthe structureofsilicate glasses.Part 2. Alkali disilicate glasses'Physics and Chemistry of Glasses,24, 72-78'
Kirkpatrick, R.J., Oestrike, R., Weiss, C.A., Smith, K.A', and Oldfield'
E. (1986) High-resolution':TAl and -Si NMR sp€ctroscopyof glasses
and crystalsalong the join CaMgSirOu-CaAlSiOu.American Mineralo g i s t ,7 1 , 7 0 5 - 7 1 1 .
I-asaga,A.C. (1982) Optimization of CNDO for molecular orbital calculation on silicates.Physicsand Chemistry of Minerals, 8,3646.
l-azarev, A.N (1972) Vibrational spectraand structure ofsilicates. New
York, ConsultantsBureau.
Liebau, F. (l 98 l) The influence of cation propertieson the conformation
of silicate and phosphateanions. In (M. O'Keefe and A. Navrotsky,
Eds., Structure and bonding in crystals, p.197-232. Academic Press,
New York.
Long, D.A. (1977) Raman sp€ctroscopy.New York, McGraw-Hill.
Matson,D.W., Sharma,S.K, and Philpotts,J.A. (1983)The structureof
134
MYSEN: AI IN DEPOLYMERIZED. PERALKALINE ALUMINOSILICATE MELTS
high-silica alkali-silicate glasses-A Raman spectroscopicinvestigation. Journal of Non-Crystalline Solids, 58, 323-3 52.
McKeown, D.A, Galeener,F.L., and Brown, G.E. (1984)Raman studies
of Al-coordination in silica-rich sodium aluminosilicate glassesand
somerelatedminerals.Journal of Non-Crystalline Solids,68, 361-378.
McMillan, P. (1984)A Raman specrroscopicstudy of glassesin the system
CaO-MgO-SiO,. American Mineralogist, 69, 645-659.
McMillan, P., and Piriou, B. (1983) Raman sflectroscopicstudiesof silicate and relatedglassstructure:A review. Bulletin Min6ralogique, 106,
57-77.
McMillan, P, Piriou, B., and Navrotsky,A. (1982) A Raman spectroscopicstudy ofglassesalong the joins silica-calcium aluminate, silicasodium aluminate and silica-potassiumaluminate. Geochimicaet Cosmochimica Acta. 46. 2021-2037.
Morey, G.W., and Bowen, N.L. (1924) The binary systemsodium metasilicate-silica.Journal of Physical Chemistry, 28, 1 167-l l7 9.
Murdoch, J.B., Stebbins,J.F., and Carmichael,I.S.E.(1985)High-resolution z'qSi
NMR study of silicateand aluminosilicateglasses:The effect
of network-modifying cations. American Mineralogist, 70, 332-343.
Mysen, B.O. (1987)Magmatic silicatemelts: Relationsbetweenbulk composition, structureand prop€rties.In B.O. Mysen, Ed., Magmatic processes:Physiocochemicalprinciples, p. 375-400. GeochemicalSociety
SpecialPublication No. l.
Mysen,8.O., and Virgo, D. (1985) Raman spectraand structure of fluorine- and water-bearingsilicate glassesand melts. In J. Snyder, R L.
Condrate, and R.R. Johnson, Eds., Advances in materials characterization, p. 43-55. New York, Plenum Press.
Mysen, 8.O., Virgo, D., and Scarfe,C.M. (1980) Relations betweenrhe
anionic structure and viscosity of silicate melts-A Raman spectroscopic study. American Mineralogist, 65, 690-710.
Mysen, B.O., Virgo, D., and Kushiro, I. (1981) The structuralrole of
aluminum in silicate melts-A Raman spectroscopicstudy at I atmosphere.American Mineralogist, 66, 678-701.
Mysen, B.O., Finger, L.W., Seifert, F.A., and Virgo, D. (1982a) Curvefitting of Raman spectraof amorphous materials. American Mineralogist,67, 686-696.
Mysen, B.O., Virgo, D., and Seifert,F.A. (1982b)The structureof silicare
melts: Implications for chemical and physical propenies of natural
magma.ReviewsofGeophysics,20, 353-383.
Mysen, 8.O., Virgo, D., and Seifert, F.A. (1985) Relationshipsbetween
structure and properties of aluminosilicate melts. American Mineralogist, 69, 83+847.
Navrotsky,A., Geisinger,K.L., McMillan, P., and Gibbs, G.V. (1985)
The tetrahedral framework in glassesand melts-Influences from molecular orbital calculations and irnplications for structure, thermodynamics and physical properties. Physics and Chemistry of Minerals,
lr,284-298.
Navrotsky, A.,Ziegler, D., Oestrike, R., and Maniar, P. (1989) Calorime1ry of silicate melts at 1773K, measurementof enthalpiesof fusion
and ofmixing in the systemsdiopside-anorthiteand anorthite-forsterite. Contributions to Mineralogy and Petrology, l0l,122-130.
Oestrike,R., and Kirkpatrick, R.J. (1988) r?Al and,,Si MASS NMR spectroscopy of glassesin the systemanorthite-diopside-forsterite.Arnerican Mineralogist, 73, 534-546.
Powell, M.J.D. (1964a) A method for minimizing a sum of nonlinear
functions without calculating derivatives. Computer Journal, 7, 303307.
(1964b) The efficient method for finding a minimum of a function
of severalvariableswithout calculatingderivatives.Computer Journal,
1,155-162.
Richet, P., and Bottinga, Y. (1986) Thermochemicalpropertiesofsilicate
glassesand liquids.-A review. Reviews of Geophysics,24, l-26.
Robinson, H.A (1969) Physical properties of alkali silicate glasses.I.
Additive relations in alkali binary glasses.Journal of the American
Ceramic Society, 53, 392-399
Ryerson,F.J. (1985)Oxide solution mechanismsin silicatemelts: Systematic variations in the activity coefficientsof silica. Geochimica et CosmochimicaActa, 49, 637-651.
Schneider,E, Stebbins,J.F., and Pines, A. (1987) Speciationand local
structurein alkali and alkaline earth silicateglasses.-Constraintsfrom
ISi NMR spectroscopy.Journal of Non-Crystalline Solids, 89, 371383
Seifert, F.A., Mysen, B.O., and Virgo, D. (l98la) Quantitative determination ofproportions ofanionic units in silicate melts. CarnegieInstitution ofWashingtonYear Book 80, 301-302.
(l 98 lb) Structural similarity betweenglassesand melts relevant to
petrological processes.Geochimica et Cosmochimica Acta, 45, 1879I 884
-(1982)
Three-dimensionalnetwork melt structure in the systems
SiO,-NaAlO,, SiO,-CaAl,Oo and SiOrMgAlrOo. American Mineralogst, 67, 696-7 18.
Sharma,S.K., Virgo, D., and Mysen, B.O. (1978) Structureof glassesand
melts of Na,O.xSiO, 1:r: l, 2, 3) composition from Raman spectroscopy. CarnegieInstitution of Washington Year Book 77, 649-652.
Stebbins,J.F. (l 987) Identification ofmultiple structural speciesin silicate
glassesby asi NMR. Nature, 330, 465467.
-(1988)
Effects of temperature and composition on silicate glass
structureand dynamics,Si-29 NMR results.Journal of Non-Crystalline
Solids,106,359-369.
Stebbins,J.F., Carmichael,I.S.E.,and Moret, L.K. (1984) Heat capacities
and entropies of silicate liquids and glasses.Contributions to Mineralogyand Petrology,86, 131-148.
Sweet,J.R., and White, W.B. (1969) Study of sodium silicateglassesand
liquids by infrared spectroscopy.Physicsand Chemistry ofGlasses, I 0,
246-251.
Tomlinson,J.W., Heynes,M S.R.,and Bockris,J.O,M. (1958)The structure ofliquid silicates.Part 2. Molar volume and expansivities.Transactions ofthe Faraday Society, 54,1822-1834.
Urbain, G., Bottinga, Y, and Richet, P (1982) Viscosity ofliquid silica,
silicatesand aluminosilicates.Geochimica et Cosmochimica Acta, 46,
to6l-t072.
Verweij, H. (1979a) Raman study of the structure of alkali germanosilicates.I Sodium and potassium metagermanosilicateglasses.Joumal
of Non-Crystalline Solids, 33, 4l-53.
-(1979b)
Raman study of the structure of alkali germanosilicate
glasses.II Lithium, sodium and potassium digermanosilicateglass.
Journal of Non-Crystalline Solids, 33, 55-69.
Virgo, D., Mysen, 8.O., and Kushiro, I. (1980) Anionic constitution of
silicate melts quenchedat I atm from Raman spectroscopy:Implications for the structure ofigneous melts. Science,208, l37l-1373.
Walrafen,G.E. (1967)Raman spectralstudiesofthe effectsoftemperature
on water structure.Journal ofChemical Phvsics.44. 114-126.
MaNuscnrpr RECETvED
Noverdsen 7, 1988
MeNuscnrp'rAccEprEDSevrrrvrnrn25. 1989