Geochimica et Cosmochimica Ada. Vol. 43. pp.

61 10 75
0 Pergamon Press Lid 1979. Printed in Greal Britain

Structure of mineral glasses-I. The feldspar glasses

NaA1Si,08, KAISi308, CaA12Si,08
MARK TAYLOR*and GORDONE. BROWN,JR.
Department of Geology, Stanford University, Stanford, CA 94305, U.S.A.

(Received 24 April 1978; accepted in revised form 29 August 1978)

Abstract-The short range distribution of interatomic distances in three feldspar glasses has been deter-
mined by X-ray radial distribution analysis. The resulting radial distribution functions (RDF’s) are
interpreted by comparison with RDF’s calculated for various quasi-crystalline models of the glass
structure.
The experimental RDF’s of the alkali feldspar glasses were found to be inconsistent with the four-
membered rings of tetrahedra associated with crystalline feldspars; the structures of these glasses are
probably based on interconnected six-membered rings of the type found in tridymite, nepheline, or
kalsilite: In contrast, the RDF of calcic feldspar glass is consistent with a four-membered ring structure
of the type found in crystalline anorthite. T-O bond lengths (T = Si,Al) increase from 1.60 A in SiOt
glass TJ. H. KONNERT and J. KARLE (1973) Acta Crysr. A29, 702-7101 to 1.63A in the alkali feldspar
&se; to 1.66A in the calcic feldspar glass due to the substitution of Al for Si in the tetrahedral
sites. The T-O-T bond angles inferred from the RDF peak positions are 151” in SiOl glass (see
reference above), 146” in the alkali feldspar glasses, and 143” in the calcic feldspar glass. Detail in
the RDF at distances greater than 5 8, suggests that the alkali feldspar glasses have a higher degree
of long range order than the calcic feldspar glasses.
Assuming that the structural details of our feldspar glasses are similar to those of the melts, the
observed structural differences between the alkali feldspar and calcic feldspar glasses helps explain
the differences in crystallization kinetics of anhydrous feldspar composition melts. Structural interpre-
tations of some thermodynamic and rheologic phenomena associated with feldspar melts are also
presented based on these results.

INTRODUCTION The interference

scattering of X-rays, neutrons, or
electrons by an assemblage of atoms (radial distribu-
A KNOWLEDGE of silicate melt structure and its com-
tion analysis) is perhaps the most structurally sensi-
position-, temperature-, and pressure-dependence is
tive technique presently available for studying glasses.
essential for rationalizing phase equilibria, transport
Scattering experiments on quenched glasses are
and rheological properties, and crystallization pheno-
much easier than similar experiments on melts but
mena in silicate systems. A growing awareness of this
require the assumption that the structure of the
need is evident in a number of recently published
quenched glass reflects the structure of the melt, for
petrologic studies (e.g. K&HIRO, 1975; CAWTHORNE
which there is some evidence in silicate systems (RIE-
and MCCARTHY,1977). Unfortunately, melt structure
BLING, 1968). We believe this to be a good assumption
is far more difficult to characterize than the structure
for systems with high viscosities and slow relaxation
of crystalline silicates, necessitating a combination of
times such as the highly polymerized alkali alumino-
indirect methods. One approach involves a variety
silicates considered in the present study. In this
of physical property measurements on melts and
regard, radial distribution experiments on liquid SiOZ
glasses (e.g. BOCKRISand LOWE, 1954; RIEBLING,1966,
(NUKUI et al., 1972) and a supercooled melt of NaAl-
1968; BURNHAM,1975) which provide constraints for
Si,Os composition above the glass transition tem-
melt structure models. The application of polymer
perature (TAYLOR and BROWN, submitted) indicate
theory to silicate melts has also been useful in this
that the local structural differences between these
regard, permitting rationalization of thermodynamic
melts and their glasses are small.
properties and phase relations of simple silicate sys-
In X-ray scattering studies made before the appli-
tems such as binary alkali silicate melts (TOOP and
cation of the Fourier transform to diffraction data,
SAMIS, 1962; MASSON, 1968; PRETNAR, 1968; HESS,
RANDALLet al. (1930) noticed the similarity between
1971; and others). More direct structural information
the diffraction profiles of amorphous silica and the
concerning melts and glasses has come from a variety
sodium and potassium feldspar glasses and the dis-
of absorption, emission, and scattering experiments,
similarity between the feldspar glasses and their as-
utilizing wavelengths ranging from y-ray to infrared.
sociated polycrystalline diffraction profiles. This led
them to suggest that feldspar crystallites did not
*Present address: Technical Stat% Division, Corning reform on cooling but “remained as cristobalite crys-
Glass Works, Corning, NY 14830, U.S.A. tallites together with crystallites of Na,O, AlzO,, or
61
62 and G. E. BROWN, JR
M. TAYL.OR

of some compound of these with silica”. This early atoms in the sample and represents scattering interac-
insight will be shown to be in general accord with tions between electrons on the same atom. The
our findings, although the crystallite hypothesis is no second summation, denoted i,(s), also extends over
longer generally accepted (EVSTROPYE~ and PORAI- all atoms in the sample except when i and j are the
Kos~rrs, 1972). Several more recent studies have same atom, and it is the structurally sensitive part
applied radial distribution techniques to feldspar of I(s). Although these summations extend over all
composition glasses. URNE~ (1969) investigated several possible interatomic distances in the sample, the indi-
glasses in the system Na*O-Al@-Si02 and several vidual contributions become small as rij becomes
which were ion exchanged with Ag to investigate the larger. The third component is the incoherent scatter-
alkali sites (URN=, 1972). None of the glasses studies ing due to inelastic electron-photon interactions. The
are on the NaAiSi,Os composition point, his “feld- variable s is related to the scattering angle, 28, and
spar” glass (URNES, 1972) being closer in composition the X-ray wavelength. i., by s = 4nsin0/~.
to jadeite, NaAlSi,O,. The RDF’s reported in these The experimentally observed intensity, la&) is
papers were calculated as “electronic” distribution related to I(s) by
functions. This leads to a low resolution RDF, but
I(s) = ~~‘,~.~s)/(~(s)~(sl) (2)
the results are not inconsistent with those which we
will report. where A(s) and P(s) are absorption and polarization
KARLE and KONNERT(1974) report and X-ray RDF corrections and K is a constant which scales I,,,, to
for Ab glass and neutron RDF’s for Ab and Or one formula unit (FU) of the scattering material. The
glasses. They demonstrated the similarity of Ab and experimental i*(s) is determined by subtracting
Or glass structures using neutron diffraction, for Ernf!fz(s) and I,,,(s). These two terms are calculated
which Na and I( have similar scattering cross-sec- from published regression coefficients of the coherent
tions. They also pointed out the similarity of Ab and (CROMERand MANN. 1968) and incoherent (BALYUZI,
silica glass structures based on their X-ray RDF’s. 1975) scattering curves. The functions E’“ff(s) and
The objectives of this paper are (I) to briefly review I,,,(s) are illustrated in Fig. 1.
X-ray radial distribution analysis and to discuss our The radial distribution functions, G(r) and D(r) are
method of interpreting experimental radial distribu- calculated by taking the Fourier transform of i,(s):
tion functions (RDF’s) using the quasi-crystaIline
model of TAYLOR (1978): (2) to present RDF’s for
NaAlSi,Os (Ab), KAlSi,Os
(An) composition
(Or), and CaAI&OB
glasses with reference to published
D(r) = 4zr'@fr) - PO) = ” s
r si,(s)
0
-
g(s)
M(s) sin (sr) ds

RDF’s of Si02 glass: and (3) to derive structural
models for these glasses and show how they may be
applied to melt-related phenomena such as viscosities
and crystal growth kinetics. A preliminary
of this study was presented by TAYLOR and BROWN
(1976). Other papers in this series will present radial
distribution results for NaAlSiO,, NaAISi,O,, CaMg-
Si,O,, and hydrous SiO, and NaAlSi,Os glasses and
NaAlSiaOs supercooled melt.
THEORY
Two recent papers (WRIGHT, 1974; WRIGHT and
LEADBETTER,1976) and one text (WARREN, 1969) have
reviewed the theory and practice of radial distribution
analysis, and only an outline will be presented here.
The review papers cited above also contain an exten-
sive bibliography of published radial distribution
studies for numerous glass systems.
The X-ray intensity scattered by an assemblage of
atoms is given by:
where N is the number of formula units in the scatter-
ing volume of the sample, rij is the distance between
atoms i and j and 1;:(s)is the scattering factor for atom
i and is proportional to Z, the number of electrons
on atom i. The first summation extends over all the
(3)
G(r) = 4dp(r) = I + 4xr2po (4)
account where
fU FU
g(s) = ~f#wc zt
and represents a “sharpening function” to correct for
the dependence of $ and ,& on s. M(s) = exp( -as*)
is the “convergence” function used to down-weight
high angle data and reduce ripples due to termination
of the integral in equation (3) as the experimental
s,,,. p. is the number of formula units per cubic
angstrom times the square of the sum of electrons in
one formula unit. The units of G(r) are electrons2/A.
G(r) represents the probability of finding two atoms
separated by a distance r t dr weighted by the
product of the number of electrons on each of the
two atoms.
G(r) may be better visualized by an example. Con-
sider an S&O, dimer joined at the bridging oxygen
with an St0-Si angle of 180’ (Fig. 2). Now draw
vectors rij from each atom to every other atom in
the model and plot the collection of vectors on a
histogram as a function of their lengths. The area
added to each bar in the histogram should corre-
spond to the product of the number of electrons on
atoms i and j. We label the rij vectors with the chemi-
cal symbots for atoms i and j. A number is appended
 Structure of mineral glasses-4 63

TAYLOR (submitt~); however, the approach is as

follows: A sphere of enclosure of radius r, is taken
around each atom in the asymmetric unit of the model
structure, and all the interatomic vectors between the
central atom and those in the sphere are calculated. A
‘disorder” parameter is assigned to each vector based
on a combination of the observed X-ray isotropic
thermal parameters of atoms i and j and the length
of the vector rip The second component is an attempt
to model the effect of loss of correlation or “random-
ness” with increasing distance. The intensity function
is then calculated using a modification of equation
(I) (see Table 1) and used in equation (3) to evaluate
G(r) for the model. Figure 3 shows an example of
Fig. 1. Components of the Ab glass diffraction profile. the total quasi-crystalline model for high albite and
Curve A, the total scattered intensity, i(s): curve B, the its components along with the RDF determined
scattering due to interactions between electrons on the
experimentally on a polycrystalline prism of high
same atom: curve C, the incoherent scattering.
albite.
The function D(r) in equation (3) represents the de-
to the symbol for the jth atom as follows: 1 for j viation of the radial distribution function G(r) from
atoms on the same tetrahedral unit as i; 2 for j atoms a homogeneous distribution of electron density repre-
on the first adjacent tetrahedra; 3 for j atoms on the sented by 4nr2po. D(r) is often quoted in the literature,
second adjacent tetrahedra, etc. The numbers thus but it does not provide better resolution nor does
correspond approximately to the first, second, and it allow for comparison of models with different
third coordination spheres, although ambiguities arise densities, although it may provide somewhat more
beyond the third sphere. The resulting radial distribu- sensitivity to changes in peak positions. Use of the
tion function (RDF) (Fig. 2a) should have maxima pair correlation function instead of 4nr2p(r) is becom-
at approximately 1.6 A (Si-Ol), 2.6A (O-01) 3.2 A ing increasingly popular (WARREN, 1969) because it
(Si-Sil), and at least three more maxima between 3.3
and 4.5 A corresponding to Si-02 and O-02 vectors.
The exact positions will depend on whether an T-O-T=l40”
T-O-T =180”
eclipsed or staggered conformation for the tetrahedra
was chosen. The area under the first maximum should
be 16 x (14 x 8) = 1792 electrons’. As a second
example, consider the same calculation for a dimer
with an Si-0-Si angle of 140” (Fig. 2b). The first three
distances are unchanged, but there are now four
unique Si-01 distances and five unique O-02 dis-
tances. If one now adds more tetrahedra, allows for
different Si-0-Si bond angles, and permits rotation
about the bridging bonds, a reasonable approxima-
tion to the RDF of vitreous silica can be obtained.
This is not unlike the procedure followed by MOZZI
and WARREN(1969) in their analysis of vitreous silica.
Three points can be made from this exercise. First,
it should be clear that G(r) is sensitive to local struc-
ture. Second, the interpretation of G(r) becomes quite
complicated beyond the first few distances. Third, as
the symmetry of the system is reduced, the RDF
becomes more smeared, with fewer sharp maxima.
Much of the secondary structure such as chains,
rings, and the iike will manifest itself between four
and eight Angstroms in G(r), and it becomes necessary
to interpret the observed RDF’s with models based
on known or hypothetical crystal structures. These Fig. 2. (a) Histogram radial distribution function of an
quasi-crystalline models (LEADBETTERand WRIGHT, S&O, dimer in eclipsed conformation with a T-O-Tan&e
1972) cannot be expected to account for all the fea- of 180’. (b) SizO, dimer in eclipsed conformation with a
T-O-T angle of 140”. The notations above the bars refer
tures of the RDF, but they may be used to calculate to the interatomic vector(s) contributing to that bar. Where
trial RDF’s for various polymerization schemes. The several unique vectors fall under one bar, a separate
details of our quasi-crystalline model are given by notation is made for each vector.
 M. TAYLOR and G. E. BROWN. JR

Table 1. Quasi-crystalline model intensity function

-
AU
iq(s) = cc Nij[exp(-bijs2)fi(s)f (s)sin(srij)/srijl+ ISA(s)
j

First summation extends over the asymmetric unit of the unit

cell of the model structure.

2) Second summation extends over all the atoms within a distancr

rc of atom i .

3) Nij is a multiplicity factor for ri. given by the multiplicity

atom i in the unit cell divided by she number of formula units
per unit cell.

4) bij is the disorder term given by:

bij = (Bi+Bj)/G3~2)+ (rijlkrc)2

and B. are the isotropic thermal parameters of atoms i and

5
J ; k is'an adjustable constant = 4 for most glasses. The second
term is an arbitrary function to model the loss of correlation be-
tween atoms as the distance between them increases.

5) fi(s) and fj(s) are the X-ray scattering factors.

6) 'ij is the mean distance between atoms i and j calculated from

the model structures.

7) ISA(s) is the small angle-scattering correction for the finite

size of the model:

FUFU
ISA(S) = c ~fi~s~fj~S~471Po~Sr~COS~Sr~~ -sin(srcH/s3
i j

eliminates small shifts in the peak positions, but we MATERIALS AND EXPERIMENTAL METHODS
have continued with the traditional form because of Samples of Ab and Or glasses were kindly provided by
its more direct physical interpretation. We have car- Dr. D. B. Stewart of the U.S. Geological Survey and are
ried out our calculation of G(r) to an rmlx of 7.5 A the same materials used by KARLE and KONNERT (1974).
The original samples were prepared at Corning Glass
because most of our interpretations involve distances Works. Corning, New York. by melting the component
shorter than this. Both experimental and calculated oxides in Pt/Rh crucibles for 6 hr at 18oo’C, crushing, and
RDF’s show features beyond 7.5& however. remelting for 6 hr at 18oo’C. The resulting glass was trans-

Fig. 3. Experimental (---) and quasi-crystalline model (---_) RDF’s of the high albite structure. The
curves under the solid curve are contributions to the total RDF from five different types of interatomic
vectors.
 Structure of mineral glasses-l 65

ferred to an annealing furnace (700°C) and allowed to cool
slowly to remove residual stress. The An glass was pre-
pared in our laboratory by melting 20g batches of the
component oxides in Pt crucibles at 155OY for 6 hr, crush-
ing, and remelting for 6 hr. No annealing step was made
and the glass shattered spontaneously after about 3 hr.
Samples were checked for homogeneity by microprobe
analysis and optical microscopy. All microprobe analyses
were carried out at 20 keV accelerating voltage, 0.01 PA
sample current, and a 1OOpm spot size; the use of smaller
spot sizes caused appreciable sodium loss. Refractive index
was measured by immersion in calibrated media. No
refractive index inhomogeneities (cord) could be observed
in either the bulk glass samples or microscopic pieces in
a matching refractive index medium. The microprobe and
microscopic checks on homogeneity are not sensitive to
compositional gradients within the bulk melt (such as
might be formed by melt stratification) or minor composit-
ional differences on the order of a few microns in size.
They indicate, however, that all the batch materials were
well dissolved and distributed in the melt. Densities were
measured on bubble-free pieces at 22°C by flotation in
CCI,/CHBr, mixtures which were in turn calibrated with
a Westphal balance. The chemical compositions, refractive
indices,‘and densities are presented in Table 2. While the
compositions are not exactly on the nominal compositions,
the differences should not have major effects on the struc-
ture. The observed densities and refractive indices are in
good agreement with published values (WINCHELLand
WINCHELL,1964).
Cylinders 1.4mm in dia of Ab and Or glasses and a
1 x 1 x 1Omm prism of An glass were cut from bubble-
free regions of the bulk glass for use in the diffraction
experiment. Diffraction data were collected on a Picker
FAG-1 diffractometer equipped with a graphite crystal in-
cident beam monochromator set for the MoK, line
(i. = 0.71069). A relatively wide PHA window was used
so that all the incoherent intensity would be included in
the total count. The diffraction profile was sampled at in-
tervals of 0.05 in s, using a narrow receiving slit from
%l,” = 0.95 to s = 5.00 A - ’ and a wide slit from s = 3.00
to smar= 15.65.&-‘. s,,,corresponds to 124.5”, the maxi-
mum 20 of our goniometer. Intensities were measured
using the fixed count method (lo4 counts). Five sets of
low and high angle data were collected and averaged for
each sample to improve counting statistics and compensate
for short-term instabilities in the diffractometer system.
The diffraction profile was extrapolated from s.,~,, to
s = 0.0 such that f,,,,(O) = 0.0. After correcting for absorp-
tion and polarization, the scale factor was evaluated by
matching the high angle data (s = 13.5c15.65 A-‘) to the
calculated background. Z’ “ff(s) and I,,,(s) were subtracted,
and the resulting iO(s)was converted to the radial distribu-
tion function using equation (3). The convergence term was
set such that M(s,,,) = 0.1 (WRIGHT and LEADBETTER,
1976). Additional details and comments on the data collec-
tion and reduction procedure are given in TAYLOR(submit-
ted).
RESULTS
The corrected and scaled diffraction profiles and
radial distribution functions are presented in Figs. 66
for Ab, Or and An glasses, respectively, together with
the radial distribution functions calculated for the as-
sociated crystalline phases. Figures 4b, Sb and 6b also
show the quasi-crystalline model distributions of
several types of interatomic vectors (T-O, O-O, and
T-T where T = Al, Si) and AC(r) the difference
between the RDF obtained for the glass and the qua-
si-crystalline model. The T-01 and T-T1 distances
and T-O-T angles estimated from the glass RDF’s
are given in Table 3. These values have been taken
from the total pair correlation functions, 4nrp(r)
rather than G(r).
Ab glass
The radial distribution functions for Ab composi-
tion glass and that calculated for the crystalline high
albite structure (PREW~TT et al., 1976) are shown in
Fig. 4. This RDF agrees qualitatively with that of
KARLE and KONNERT (1974) though they cannot be
compared directly because of the different forms of
the radial distributions functions used. There is a
close correspondence between the crystal and glass
curves to about 4A suggesting a primary structure

Table 2. Physical properties and compositions of glasses used in this study

Designation Ab' Or2 An

Formula unit NaAlSisOe KAlSi3Oa caAlzsi2os

Glass High Albite Glass Sanidine Glass Anorthite

Density3 2.382 2.615 2.366 2.575 2.691 2.763
Refrackve
index4 1.482 1.53-1.54 1.488 1.52-1.53 1.570 1.58-1.59
Composition5 Observed Nominal Observed Nominal Observed Nominal
sioz 70.34 68.74 65.30 64.76 45.87 43.19
Al203 17.75 19.44 19.48 18.32 34.74 36.65
NazO 12.01 11.82 0.00 - 0.02 -
KzO 0.04 - 15.89 16.92 0.03 -
cao 0.03 - 0.03 - 20.81 20.16
Total 100.17 100.70 101.47

' Corning glass X955QB”.

* Corning glass X955QA’.
3 Density of high albite, sanidine, and anorthite are calculated from the cell
parameters.
4 Refractive indices of crystals from WINCHELLand WINCHELL(1964).
5 Compositions determined by microprobe analysis (wt. per cent).
G.T.A.
43,I-E
 M. TAYLORand G. E. BROWN.JR

4 8 12

60000-

8
”

r. ,’
20000 -

I I I I I i I _
I 2 3 4 5 6 7
f (a)

Fig. 4. (a) Corrected and scaled diffraction profile I(s), for Ab glass. (b) Ab glass RDF (solid line):
quasi-crystalline model for the high albite structure (dashed line); contributions from T-On. O-On.
and T-Tn types of vectors (dotdash, dotted, and small dashed Iines, respectively); AG(r) (lower curve)
is the difference between the quasi-crystalline model RDF and the observed glass RDF. The area
under AG(r) to 7.5 A is 2.41 x lo4 electrons2.

Table 3. Interatomic distances and T-O-T angles

Ab Or AII Silka Silica'

T-01 1.63 1.63 1.66 1.59 1.62

T-T1 3.12 3.12 3.15 3.08 3.12

a 146' 146" 143" 151* 144"

’ KONNERT and KARLE (19733.

’ Mozzt and WARREN(1969).
The interatomic distances are taken from the positions of maxima
in the total pair correlation function G&)/r. The T-O-T angle, a, is in-
ferred from the T-01 and T-T1 distances as a = 2 arcsin[(T-
T1)/2(T_Ol)]. The T-OI distance should be considered more reliable
than the T-T1 distance because of the possibility of other interactions
contributing to the maximum at 3.1 A.
 Structure of mineral glasses--l 67

I I I I I I ’ I

60000-

Fig. 5. (a) Corrected- and scaled diffraction profile, I(s), for Or glass. (b) Or glass RDF (solid line);
quasi-crystalline model RDF for the sanidine structure (dashed line); contributions from T-On, O-On,
and T-Tn types of vectors (dot-dash, dotted, and small dashed lines, respectively); AC(r) (lower curve)
is the difference between the sanidine and Or glass RDF’s. The area under AC(r) to 7.5 A is 2.24 x lo4
electrons2.

of tetrahedrally coordinated silicon and aluminum
atoms sharing all four corners with other tetrahedra.
Beyond 3.4 A, there is progressively poorer agreement
with respect to both the area under the curve and
the positions of the peaks. The most important of
these differences occur at 4.0A where the high albite
model rises considerably higher than the glass and
at 5.1 A where the glass RDF reaches a maximum
and the high albite RDF goes through a minimum.
The peak between 4.0 and 4.5A in the high albite
RDF is due mainly to the distribution of T-02 and
T-T2 vectors, as can be seen from the T-O and T-T
components (Fig. 4b). The glass curve follows the
T-02 component of the high albite structure quite
closely (allowing for small contributions from Na and
0), but the smaller area under the glass curve between
4 and 4.5 A may be attributed to a difference in the
T-T2 distribution between Ab glass and the high
albite structure. This range of distances corresponds
to the vectors between tetrahedral cations directly
across the characteristic feldspar four-membered ring.
The T-T2 distances in Ab glass are apparently shifted
to c 5.1 A where the RDF shows a maximum. There
are no interatomic vectors in the high albite structure
that could account for a maximum at this distance.
Beyond this point, the high albite RDF rises steadily
above the glass RDF, as must be expected from the
difference in density. It therefore appears that while
Ab glass and crystalline high albite have similar struc-
tures in the first coordination sphere, there are funda-
68 M. TAYLOR and G. E. BROWN. JK

60000-

I I I I I I I I
I 2 3 4 5 6 7
r (ii)

Fig. 6. (a) Corrected and scaled diffraction profile, I(s), for An glass. (b) An glass RDF (solid line):
quasi-crystalline model RDF for the anorthite structure (dashed line); contributions from ~-on. o-on,
and T-Tn types of vectors (dot-dash. dotted. and small dashed lines. respectively): AC(r) (lower curve)
is the difference between the anorthite and An glass RDF’s. The area under AC(r) to 7.5 A is 1.67 x IO4
electrons’.

mental differences in the way in which the tetrahedra
group to form rings.
Or glass
The observed radial distribution functions for Or
glass is compared with that calculated for the sanidine
structure (BROWN et al., 1974) in Fig. 5. There is no
a priori reason to expect a sanidine structure for the
melt since sanidine is not a stable liquidus phase for
this composition at atmospheric pressure, but our
purpose here is to compare feldspar glasses with the
bonding topology of their associated crystalline struc-
tures. As with albite, the short range structure in the
glass and crystal is quite similar, but the correspon-
dence breaks down beyond about 3.5.A The K-01
vectors are still not resolved despite the larger
number of electrons on potassium. This is not too
surprising because, even in the sanidine structure, a
large thermal parameter for K and a wide distribution
of K-O distances cause a broad maximum in the
RDF. Beyond 3.5 A the Or curve is seen to follow
the sanidine T-02 component, but drops off at about
4.2 A while the sanidine curve continues to rise. This
rise is due to contributions from K-02 and T-T2
vectors. At 5.1 A, the sanidine RDF goes through a
minimum while the Or glass RDF is at a maximum.
The situation is again reversed at 5.8 A, where the
sanidine curve is at a maximum due to contributions
from T-T2 and T-T3 vectors while Or glass is at
a minimum. Beyond 6A, the sanidine RDF rises
above the glass RDF in keeping with the differences
in density between the glass and crystal. As with Ab
 Structure of mineral glasses--l
glass, it can be concluded that Or glass has a dis-
tinctly different secondary structure than the ogrre-
sponding feldspar. The differences around 4.6A are
very likely due to a lack of four-membered rings in
the glass.
A second quasi-crystalline model based on the leu-
cite structure (WYART, 1940; PEACOR, 1968) was also
compared to the Or glass RDF following the idea
that the highest temperature crystalline phase is in-
dicative of the melt structure (e.g. KUSHIRO, 1975).
The leucite RDF shows a clear minimum at 4.5 A,
presumably due to a smaller number of four-mem-
bered rings, but it does not accout for the maximum
at 5.1 A. Beyond this point, the Or glass RDF and
leucite RDF are completely uncorrelated.
An glass
The RDF for An glass is shown in Fig. 6 along
with the calculated crystalline RDF based on the
anorthite structure reported by WAINWRIG~ and
STARKEY (1971). As with Ab and Or glass, the An
glass RDF shows a close correspondence to its crys-
talline counterpart to at least 3.5 8. The rise in the
RDF between 2.3 and 2.5 A can be attributed to a
closely grouped set of Ca-O vectors suggesting that
Ca occupies a seven-fold coordination site in the glass
structure, which is very similar to that of crystalline
anorthite. Unlike the Ab and Or glasses, however,
An glass shows a close correspondence with its crystal
structure between 3.5 and 4.5 A. The glass RDF does
not match the crystalline curve exactly, but the maxi-
mum at 4.48, and the lack of a distinct minimum
between 4.5 and 5.OA suggest a larger number of
T-T2 distances around 4.7 A and hence a larger
number of four-membered feldspar rings. The RDF
is essentially featureless beyond 4.5 .& suggesting that
there is very little long-range order,
The A@) function plotted in Figs. 4b, 5b and 6b
is a measure of relative differences in structure and,
as a consequence, bulk density between each feldspar
composition glass and its quasi-crystalline
based on the structure of the respective crystalline
feldspar. The fact that the area under AC(r) for An
glass-anorthite pair (1.67 x 104e2) is smaller than the
corresponding values for the Ab-high
(2.41 x 104eZ) and Or-sanidine (2.24 x 104e2) pairs is
consistent with the difference in heat capacities, AC,,
between crystalline and amorphous feldspars recently
observed by ROBIE et al. (1978). Their data show that
AC, values for the glass-crystal pairs increase in the
order An < Ab < Or, suggesting that the structural
arrangements and bonding energies in the tetrahedral
portion of the Or or Ab glass-crystal pairs are signifi-
cantly different relative to the An glass-crystal
DISCUSSION
Comparison offeldspar and silica glass RDF’s
The silica (SiOJ glass RDF is an obvious point
of comparison for the feldspar RDF’s. There have
69
been many radial distribution studies of silica glass
(WRIGHT, 1974; WRIGHT and LEADBETTER,1976), but
two are of special interest. MOZZI and WARREN (1969)
performed a very careful determination of the RDF
and interpreted it in terms of a random network
model of the glass structure. By adjusting the inter-
atomic distances and angles, they were able to devise
a model which fit the RDF very accurately to at least
3.5 8, and gave interpretations to the features of the
experimental RDF to 5.1 A. The structural par-
ameters derived from this analysis are listed in Table 3.
Their additional angular variables, /3 and y, describe
configurational differences caused by rotation of
tetrahedra about the Si-0 bonds but, unlike CL,they
found no preferred values for /I and y. KONNERT and
KARLE (1973) refined the RDF of silica glass using
a least squares technique to adjust the scale factor,
the background shape, and the first three interatomic
distances and peak widths. The structural parameters
derived from their analysis are listed in Table 3; the
angle a is inferred from their published results. In
addition, KONNERT et al. (1973) compared their silica
glass RDF with RDF’s derived from calculated and
experimentally measured powder patterns of various
polymorphs of SiO, and concluded that silica glass
has a tridymite-like bonding topology based on six-
membered rings of Si04 tetrahedra. Others (cf.
WRIGHT and LEADBETTER, 1976) have suggested an
a-cristobalite structure.
The distinction between tridymite and cristobalite
is well defined crystallographically, but in terms of
a tridymite-like or cristobalite-like structure for SiOZ
glass, the distinction is very small. In the idealized
structures, cristobalite consists of six-membered rings
where all four tetrahedra are in the staggered con-
figuration with respect to a central tetrahedron,
whereas in tridymite, three are staggered and one is
eclipsed. This leads to all “chair” conformation rings
in cristobalite and both “boat” and “chair” rings in
tridymite. The energy difference between the stag-
model gered and eclipsed configurations of SiZO, is quite
small (DE JONG and BROWN, 1978) and we expect both
configurations as well as intermediates between them
to be present in SiO, glass. However, the density of
albits the glass and the correlation of peak positions in the
RDF tend to favor the tridymite model.
If the idealized structures were the only ones
present in silica glass, we would expect Mozzi and
Warren’s #I and y parameters to take on discrete
values. However, more realistic structures for low tri-
dymite and low cristobalite (DOLLASE, 1965; D~LLASE
and BAUR, 1976; KONNERT and APPLEMAN, 1975;
KAYO and NUKIJI, 1976) show enough distortion from
pair. the idealized structures to suggest that p and y take
on a wide range of values in the crystalline poly-
morphs as well as in the glass. Thus the “tridymite-
like” or “cristobalite-like” models of silica glass would
lead to the same interpretation of the short range
RDF features as would Mozzi and Warren’s random
network view.
70 M. TAYLOR and G E. BROWN.JK

The observed radial distribution functions for Ab,
Or and An glass are shown in Fig. 7 along with the
RDF of vitreous silica calculated from the data of
KONNERT and KARLE (1973). For consistency, the
silica and feldspar RDF’s have been calculated using
a direct Fourier transform of the observed data with-
out least squares refinement of any parameters. The
similarity among the RDF’s of silica, Ab, and Or can
be seen imm~jately. The heights of the curves are
slightly different owing to the differences in average
electron concentration, but there are no major shifts
which would suggest different structures. Substitution
of NaAIO, and KAIO, for SiOt (Ab and Or vs silica)
causes a distinct shift in the PO1 maximum consis-
tent with the formation of longer AI-O bonds. This
shift is even more pronounced in An glass where half
the tetrahedral atoms are aluminum.
The An glass RDF is clearly different from the pre-
vious three with regard to both the peak positions
in the range from 3.5 to 5.08, and the lack of struc-
tural detail between 5.0 and 7.5 A. The difference in
peak positions suggests that An glass is based on a
different structure than silica_ Ab, and Or.
The similarity in the silica, Ab, and Or glass RDFs
suggests that a tridymite-like structure based on a
three-dimensional framework of six-membered rings
of tetrahedra might be a good model for Ab and Or
glasses. Aluminum would substitute for silicon in
some of the tetrahedral sites. If the rings were fairly
regular. such a structure would naturally have void
spaces or “interstices” between the rings (one for
every two tetrahedral atoms) to accommodate the
alkali cations in approximately 12-fold coordination.
This stuffed-tridymite model for the glass is analogous
to the crystalline derivatives concept developed by
BUERGER (1954).
Several factors make this an attractive model. First.
there are several stable crystalline analogs such as
nepheline, NaAISiO.+, and kalsilite, KAISiOd, which
give the model a crystal chemical precedent. Nephe-
line and kalsilite have tetrahedral frameworks similar
to that of tridymite, with the alkali ions occupying
the spaces between the rings. These structures are
similar to those envisioned for the Ab and Or glass,
though they have all the void spaces filled with alkali
ions whereas Ab and Or glass would have only half
the spaces filled. We find further evidence for this
type of coupled substitution at the silica end of the
join. Examination of chemical analyses of several
SiO, polymorphs (DEER et al., 1963) shows that the
tridymite structure is the most capable of accepting
alkali and aluminum ions. There is evidence that tri-
dymite is not a stable polymorph of SiO, but requires
small amounts of impurities to nucleate (HOLMQUIST,
1961).
Second. the densities calculated from a stuffed tri-
dymite model are in reasonably good agreement with
the observed densities of the alkali feldspar glasses.
We have estimated a density for our model by assum-
ing that the tetrahedral framework of crystalline
nepheline and kalsilite is approximately the same as
that of Ab and Or glass. respectively. If we write the
unit cell formula of nepheline (eight “molecules” of
NaAlSiO,) as NasAlsSisOJz and substitute four Si02
for four NaAlO,, we have Na~Al~Si,~O~~ or four for-

60000

Fig. 7. Comparison of Ab. Or. An and silica RDF’s. The short vertical lines show positions of maxima
in D(r). the form of the RDF which represents the difference between G(r) (solid curve) and the radial
distribution of a structureless medium with the same average electron density (dashed curve}. Silica
curve has been resealed to an eight oxygen formula unit for comparison with the feldspar%
 Structure of mineral glasses-l
mula units of albite. Using the cell volume of nephe-
line we obtain the density of the model:
(formula units per unit cell)
x (formula weight)
density (g/cm3) =
(cell volume) x 0.6022
The density of Ab glass calculculated in this way is
2.40 compared to the observed density of 2.38. Per-
forming the same calculation for the kalsilite struc-
ture, we estimate a density of Or glass of 2.31 com-
pared to the observed density of 2.37. The densities
estimated from these structural models are in much
better agreement with the observed glass densities
than densities from the high albite (2.62) and sanidine
(2.57) structures.
The third attractive feature of the stuffed tridymite
model is its consistency with a continuous network
view of glass structure. The observed densities and
radial distribution functions can be rationalized with-
out postulating ordered and disordered regions, large
void spaces, or a high concentration of structural irre-
gularities (such as six coordinated aluminum or non-
bridging oxygens). We do not wish to imply that the
glass structures are perfectly ordered or that irregu-
larities do not occur: the shapes of the peaks in the
RDF suggests that even within the rings there is a
certain amount of conformational randomness.
The last attractive feature of the stuffed tridymite
model is that it gives a reasonable explanation of the
features of the Ab and Or glass radial distribution
functions. Figure 8 shows the observed Ab glass RDF
and the RDF calculated for the stuffed tridymite
model. This model has been constructed from the
nepheline structure reported by SIMMONSand PEACOR
(1972) by weighting the contribution of a given pair
of atoms in proportion to the number of pairs of
that type which would be found in the feldspar cell.
For example, there are twice as many sodium atoms
60000
I 2
Fig. 8. Stul%d tridymite model for Ab glass based on the nepheline structure. Ab glass RDF (solid
line); stuffed tridymite model (dashed line); T-On, O-On, and T-Tn contributions to the stuffed tri-
dymite model (dot-dash, dotted, and small dashed lines, respectively).
71
in a cell with the nepheline formula as one with the
albite formula. The contribution of each Na-0 dis-
tance calculated in the nepheline cell is multiplied by
l/2 to give its contribution to the feldspar model. In
crystallographic terms, this corresponds to a half-
occupied sodium site. Unfortunately, as of this writing
there are no published structural data on a pure
sodium nepheline: the crystal used by Simmons and
Peacor had l/4 of the sodium atoms replaced by
potassium. The stuffed tridymite model for Or glass
has been calculated in a similar manner using the
structure of kalsilite reported by F'ERROTTA and SMITH
(1965) and is shown with the Or glass RDF in Fig. 9.
No attempt has been made in either model to account
for the effect of substituting SiO, for NaA102 on the
average T-01 bond length: the first peak in the model
RDF’s are at slightly higher values of r than would
be expected for the Ab and Or compositions. The
nepheline and kalsilite O(1) oxygen has been moved
to the on-axis position to remove the effects of oxygen
disorder on T-O interactions.
One problem with these model structures is that
the hexagonal crystallographic symmetry forces a
very narrow distribution of certain atomic distances
giving rise to very sharp maxima in the quasi-crystal-
line model. The T-T2 peak at 5.1 A in the Or stuffed
tridymite model (Fig. 9) is a good example of this
problem. We have not as yet devised a means of
relaxing this symmetry, though the length dependence
of the disorder parameter is an attempt.
The features of the model RDF’s are much sharper
than those of the glasses, but the positions with re-
spect to r are in fairly good agreement. The maxima
in the glass RDF’s between 3.0 and 5.5 A are shifted
to somewhat lower values of r, suggesting that the
models would be in better agreement if they were
collapsed a bit. For example, if the T-O-T angle in
nepheline were slightly smaller, the T-T1 and T-T2
distances would be shorter. Such a collapse might be
3 4 5 6 7
72 M. TAYLORand G. E. BROWN. JH

60000 - STUFFED YRI~YMITE

?z
*

Fig. 9. Stuffed tridymite model for Or glass based on the kasiiite structure. Or glass RDF (solid
line); stuffed tridymite model (dashed line); T-On, O-On. and T-Tn contributions to the stuffed tri-
dymite model (dot-dash. dotted. and small dashed lines, respectively).

expected since the models are derived from fully through adjacent tetrahedra, a third four-membered
“stuffed” structures, whereas the Ab glass should have ring of similar conformation is produced between
only half the voids filled. them. This ring will contribute to the T-T2 maximum
The main points of the comparison between at 4.6 A just as the original two rings. A smaller pro-
observed glass RDF’s and the calculated stuffed tridy- portion of this type of adjacent tetrahedra or “‘crank-
mite RDF’s for both Or and Ab glasses are (1) the shaft” linkage in the glass as in the anorthite structure
maxima at 4.2 A are due mainly to T-02 vectors: would explain the reduction in the area under the
(2) relatively few vectors occur between 4.4 and 4.7 A, glass RDF peak at 4.2 A relative to that calculated
causing a distinct minimum in the RDF; (3) the max- for the anorthite structure. It would also be in keep-
ima at 5.1 A are due to the distribution of T-T2 vec- ing with the similar but slightly lower density of An
tors. These results are in sharp contrast to the high glass (2.69 g/cm3) relative to that of the anorthite
albite and sanidine quasi-~ystalline models where the structure (2.76 g/cm31.
T-T2 vectors are at about 4.6 8, due to the presence
Implications for melt phenomena from glass structures
of four-membered rings. The difference between the
feldspar and stuffed tridymite bonding topologies is Making the assumption that these materials retain
represented schematically in Fig. 10, with the tetra- the same atomic arrangements in passing from the
hedra representing SiO, or AlO groups and the large glassy to the molten state, we find that the postulated
circles representing alkali positions. Although six-
membered (and larger) rings are present in the feld- ,A’ A. B
spar structure. four-membered rings are predominant.

Structurul models for An glusses

As pointed out earlier, the anorthite structure

accounts for most of the features of the glass RDF
to approximately 5.OA. Beyond this point there is
little indication of regular interatomic distances. We
would therefore envision a glass structure based on
feldspar-like four-membered rings of Si04 and AlOr
tetrahedra (Fig. 10) which are connected to other
four-membered rings in a more or less random fash-
ion. For example, some rings may be connected
through only one T-O-T bridge whereas others may Fig. 10. (a) Schematic representation of reIdspar four-mem-
be joined through adjacent tetrahedra, pr~ucing the bered rings of tetrahedra. Front and back rings are joined
so-called “crankshaft” arrangement characteristic of through opposite tetrahedra whereas top and bottom rings
are joined through single tetrahedra. The circles represent
feldspar crystal structures. Still others may be con-
alkali ion positions. (b) Representation of stuffed tridymite
nected through opposite tetrahedra to form the struc- six-member rings. The vertical ring on the left side of the
ture shown in the top portion of the molecule in drawing has the “boat” conformation and the horizontal
Fig. 10a. When two four-membered rings are joined top and bottom rings have the “chair” conformation.
 Structure of mineral glasses-l
glass structures provide some qualitative interpre-
tations of phenomena associated with Ab, Or and An
melts. For example, FL~CID and KNAPP (1968) have
calculated liquidus curves for the SiO,-NaAlSi,Os,
SiO*-KAISiJOs, and Si02-CaAl&Os joins, assum-
ing various structural models for the substitution
alumina into silica liquid. They found that random
substitution of Al for Si was the best model for the
alkali feldspar melts, though some evidence for minor
amounts of a “polymeric” species (presumably units
with a structure different from that of the SiOs melt)
was observed on the albitic side of the SiO,-NaAl-
Si30s join. Random substitution of Al for Si implies
that the basic structure of the liquid is not changed.
This implication is consistent with the view that silica,
Ab, and Or melts all have tridymite-like tetrahedral
arrangements. On the other hand, Flood and Knapp
found it necessary to postulate aluminosilicate
anionic “clusters” with a structure different from that
of the SiOZ melt to explain the silica-rich side of the
Si02-CaA12Si,06 liquidus. The best formula for
these clusters was found to be 4A10,.3SiOi-.
Although this is not exactly the same formula as we
would suggest for the four-membered rings in An melt
(Al&O~- or, for two rings, AL&O:;), Flood and
Knapp’s results are consistent with our idea that An
melts are based on a different polymerization scheme
than silica, Ab, or Or composition melts. They con-
cluded that Al and Si tetrahedra show “an increasing
deviation from random distribution” in the sequence
K < Na <<Ca
which are interpreted to mean that Al tetrhedra in
CaAl,O,-containing melts are present as pairs
bonded to Ca in order to satisfy local charge require-
ments. This pairing is not required in KAIO; and
NaAIO; containing melts. The most energetically
favorable arrangement of Al and Si tetrahedra in the
Ca2Si3A14014 clusters proposed by Flood and Knapp
is thought to be an alternating one like that in crystal-
line anorthite. This suggestion follows from the alu-
minum avoidance principle (LOWENSTEIN, 1954) and
the observation that anorthite retains perfect ordering
of Al and Si tetrahedra up to the melting temperature
(LAVES et al., 1970) whereas the crystalline forms of
NaAlSi,Os and KAISi,Os do not. Based on our RDF
results for An glass, we also suggest that the molecu-
lar clusters present in An-composition melts consist
of four-membered rings of tetrahedra rather than
Flood and Knapp’s clusters which show three-mem-
bered rings of A12Si09 composition.
Nucleation and growth of alkali feldspar crystals
from their anhydrous melts near the liquidus is notor-
iously difficult (SCHAIRER and BOWEN, 1956), while
crystallization of anorthite proceeds relatively rapidly
(DAY and ALLEN, 1905). We can rationalize this differ-
ence in the case of crystallization in terms of the melt
structures by noting that at least two T-0 bonds
must be broken and reformed to form a four-mem-
bered ring from a six-membered ring such as in the
73
alkali feldspar liquid --) crystal transition. This would
have to happen many times within a limited volume
of alkali feldspar melt to form a stable nucleus, and
the melt network around the nucleus would have to
readjust to accommodate the resulting reduction in
of volume. If the melt has a feldspar-like bonding topo-
logy, nucleation could proceed by simple conforma-
tional adjustments where a relatively small number
of T-0 bridges would have to be made or broken.
We would therefore expect a rather large barrier to
nucleation in Ab and Or composition melts relative
to An composition melts.
If melting of crystalline albite and sanidine has to
proceed by a similar mechanism, i.e. the breaking of
four-membered rings to form six-membered rings, we
might expect a similar kinetic barrier to the formation
of the melt (AINSLIE et al., 1961). DAY and ALLEN
(1905) noted the persistence of albite crystals at tem-
peratures well above the melting point and, in similar
experiments, we have noted the persistence of crystals
even after 48 hr at temperatures 60” above the melting
point. This superheating phenomenon was studied in
detail for albite by DIETZ et al. (1970).
The viscosities of the feldspar melts may also be
correlated with the proposed structures. Or and Ab
melts have viscosities on the order of lo6 and lo4
poise, respectively, at 1550°C (cf. SHAW, 1972, Fig. 9)
whereas anorthite has a viscosity of only lo’-10’
poise at the same temperature (CUKIERMANand UHL-
MANN,1973). UBBELOHDE(1965) has stressed the im-
portance of “blocked volumes” (volume fractions of
the melt which must move as a unit) in understanding
viscosity anomalies in organic liquids. He has sug-
gested that the sharp rise in the viscosity of albite
near its freezing point may be due to a fraction of
the volume being blocked to flow by cluster forma-
tion. The stable network of six-membered rings in-
ferred to be present in alkali feldspar melts would,
in this context, represent relatively large volumes of
melt which are constrained to flow as units, whereas
single four-membered rings may be the flow units in
An melts. Structural interpretations of the viscosity
measurements tend to support this view. CUKIERMAN
and UHLMANN (1973) suggested a flow unit in An-
composition melts of about one formula unit; this
would correspond to one four membered ring in our
model. DIETZ et al. (1970) assumed a four-NaAlSi,Os
flow unit for Ab mefts based on the unpublished data
on Baak and Hornyak for the sodium disilicate-alu-
mina system; this would correspond to 2-3 six-mem-
bered rings or a cluster slightly larger than that
shown in Fig. lob. The smaller flow units in An com-
position melts could be due to the 1 : 1 Al :Si ratio
leading to a higher probability of AI-O-AI bridges
which are less stable than the AI-O-S1 and Si-0-Si
bridges most probably present in Ab and Or compo-
sition melts. Our structural models for Ab and Or
composition melts offer no explanation of the differ-
ences in viscosity of these melts because both are
based on similar tetrahedral frameworks. However,
74 M. TAYLOR and G. E. BROWNJR

the lower viscosity of Ab relative to Or composition
melts may be rationalized by noting that Na-0 bonds
are shorter than K-O bonds in the alkali feldspar
structures and that this difference presumably persists
in the molten state. The closer approach of Na to
the bridging oxygen leads to a greater inductive effect
of Na relative to K resulting in a greater weakening
of PO-Tlinkages in Ab relative to Or composition
melts.
The similarity of radial distribution functions of
glasses in the system Si02-NaAlSiaOs-KAlSi,Os
suggests that the stuffed tridymite structure might be
a good starting model for anhydrous granitic melts.
To test this, we have prepared a glass with the same
composition as Nockolds’ average hornblende biotite
granite (NOCKOLDS, 1954). The results will be reported
in detail with some other synthetic rock glasses cur-
rently under investigation, but it appears that addi-
tion of 5-6 wt“,, mafic oxides (Fe,O,i-CaO+MgO)
and an alkali/aluminum ratio of less than one does
not significantly alter the six-membered ring melt
structure.
Finally, the subject of immiscible natural magmas
has received increasing attention in the petrologic
literature (E. ROEDDER, 1977 Presidential Address,
Geochemical Society: PHILPOTTS, 1976). The separ-
ated pairs typically can be divided into an alkali alu-
mino-silicate member and an alkaline earth, transi-
tion metal silicate member. For example, PHILPOTTS
(1971, 1976) found droplets (ocelli) of alkali feldspar
and nepheiine or quartz which apparently separated
from the parent basic magma as a less dense phase.
Their chemical compositions show a strong enrich-
ment in the network forming oxides, SiO,, Al,O, and
alkali relative to the parent magmas. Their densities
have been inferred to be between 2.1 and 2.3 g/cm3.
The ocelii approach the stoichiometry necessary for
a stuffed tridymite melt structure and have a density
similar to that which would be predicted for such
a melt. Because of their high alkaline earth and transi-
tion metal contents, the coexisting basalt fractions
presumably represent melts based on short chain
silica polymers and edge sharing metal oxide octa-
hedra. These structures would suggest melts of higher
density without the large void spaces afforded by the
fully polymerized six-membered ring structures and
would tend to exclude large ions such as Na and
melt structures. There is evidence (see Part II of this
series) that these structures are stable as far as the
nepheline and, presumably, kaisiiite compositions.
The An glass RDF can be explained by a four-mem-
bered ring feldspar-like structure. though the RDF
shows no regular structure larger than this. These
structures are consistent with the melt structure
models proposed by FKIOD and KNAPP (1968) for the
silica-rich side of their respective binaries with SiOr.
These results also have bearing on the differences in
heat capacities between crystalline feldspars and their
glasses. Furthermore, they can be used to rationalize
the nucleation and viscous flow behavior of feldspar
melts and may give some insight into the behavior
of more complex sialic magmas.
The models presented here are not necessarily
unique, and others may be devised which improve
the fit to the observed radial distribution functions,
Furthermore, the use of high intensity X-ray sources,
electron and neutron diffraction, and improvements
in data acquisition and reduction techniques should
provide more reliable radial distribution functions in
the near future. A point we wish to emphasize is that
the radial distribution function is a physical observ-
able of the glass or melt for which proposed atomistic
models must account,
Ackno~I~ngemenrs--We wish to acknowledge and thank
J. H. RoNNERT of the U.S. Naval Research Laboratory
for discussions and a copy of his radial distribution refine-
ment program: A. I. BIENENSTO~Kof the Stanford Mater-
ials Science Department and P. M. FENN of the Stanford
Geology Department provided critical readings of the
manuscript. The alkali feldspar glass samples were supplied
by D. B. STEWART of the U.S. Geological Survey and the
calcic feldspar glass was prepared by B.H.W.S. DE JONC
of Stanford. We also thank Standard Oil of California for
computer time on their IBM 3701168. Support was pro-
vided by NSF Grants GA 41731 and EAR-74-03056-OAl.
REFERENCES
AINSLIE N. G.. MACKENZIEJ. D. and TURNBULL D. (1961)
Melting kinetics of quartz and cristobalite. J. P~u,
Chem. 65, 1718--1724.
BALYUZI H. H. M. (1975) Analytic approximations to inco-
herently scattered X-ray intensities. Actu Crpstallogr.
A31, 600-602.
BOCKRIS J. O’M. and LOWE D. C. (1954) Viscosity and
the structure of molten silicates. Proc. Roy. Sot. A226,

K. The stuffed tridymite structure may thus represent 423-435.

Btzowrj G. E., HAMILTONW. C.. PREWIT?‘ C. T. and SUENO
a “preferable” structure for certain ions S. (1974) Neutron diffraction study of AljSi ordering in
sanidine: a comparison with X-ray data. NATO A&.
Stud. Imt. on Feldsgars. Symp. Vol.. Manchester. Eng-
CONCLUSIONS land, pp. 6880.
BUERCER M. J. (1954) The stuffed derivatives of the silica
By comparison of the observed radial distribution
structures. Am. ,~i~era~og~~t 39, 6&J-614.
functions to those calculated for model structures, we BURNHAM C. W. (197% Water and magmas: a mixing
have shown that Ab and Or glass are not consistent model. Georttim. Cnsmochim. Acta 39, 1077-1084.
with a feldspar-like four-membered ring structure. CAWTHORNR. G. and MCCARTHY T. S. (1977) Partitioning
The observed RDF’s are better explained by a six- of nickel between immiscible picritic liquids. Earth
Planet. Sci. Lett. 31, 339-346.
membered ring, stuffed tridymite structure leading to CROMER D. T. and MANN J. B. (1968) X-ray scattering
the view that all compositions within the system factors computed from numerical Hartree-Fock wave
SiO~-NaAlSi~O~-KAiSi~O* will have similar overall functions. Actn Cr~.stul~~gr. A24, 321.--324
 Structure of mineral glasses-l
CUKIERMANM. and UHLMANND. R. (1973) Viscosity of
liquid anorthite. J. Geophys. Res. 78. 49204923.
DAY A. L. and ALLENE. T. (1905) The isomorphism and
thermal properties of the feldspars. Publication No. 31.
Carnegie Institute of Washington, Washington, D.C.
DEER W. A., HOWIE R. A. and ZUSSMANJ. (1963) Rock
Forming Minerals Vol. 4, Framework Silicates. John
Wiley, New York.
DE JONG B. H. W. S. and BROWNG. E. (in preparation)
Molecular orbital studies of silicate molecules II. The
S&0$-, H&O:, H,SiAIO:- and H,Al,O:- mol-
ecules.
DIETZ E. D., BAAKT. and BLAU H. H. (1970) The super-
heating of an albite feldspar. Z. Krist. 132, 340-360.
D~LLASEW. A. (1965) Reinvestigation of the structure of
low cristobalite. Z. Krist. 121, 369-377.
D~LLASEW. A. and BAURW. H. (1976) The superstructure
of meteoritic low tridymite solved by computer simula-
tion. Am. Mineralogist 61, 971-978.
EVSTROPYEV K. S. and PORAI-KOSHITS E. A. (1972) Discus-
sion on the crystallite hypothesis of glass structure. J.
Non-crvsr. Solids 11, 17&172.
FLOOD H. and KNAPP W. J. (1968) Structural character-
istics of liquid mixtures of feldspar and silica. 1. Am.
Ceram. Sot. 51, 259-263.
HES P. C. (1971) Polymer model of silicate melts. Geochim.
Cosmochim. Acra 35, 289-306.
HOLMQUIST S. B. (1961) Conversion of quartz to tridymite.
i. Am. Ceram. Sot. 44, 82-86.
KARLE J. and KONNER~J. H. (1974) Analysis of diffraction
from amorphous materials and applications. Proc. Am.
Crystallog. Assoc. 10, 29-43.
KATO K. and NUKUI A. (1976) Die Kristallstructur des
monoklinen Tief-Tridymits. Acra Cr.vsrallogr. 32,
24862491.
KONNERTJ. H. and APPLEMAND. G. (1975) Terrestrial
low tridymite: refinement of the three dimensional struc-
ture. Geol. Sot. Am. Ann. Mrg. Absrr. Prog. 7, 1151.
KONNERTJ. H. and KARLEJ. H. (1973) The computation
of radial distribution functions for glassy materials. Acra.
Crysrallogr. A29, 702-710.
KONNERTJ. H., KARLEJ. H. and FERGUSONG. A. (1973)
Crystalline ordering in silica and germania glasses.
Science 179, 177-178.
KUSHIROI. (1975) On the nature of silicate melt and its
significance in magma genesis: regularities in the shift
of the liquidus boundaries involving olivine, pyroxene,
and silica minerals. Am. J. Sci. 275, 411-431.
LAVESF., CZANK M. and SCHULZH. (1970) The tempera-
ture dependence of the reflection intensities of anorthite
(CaAl,Si,Os) and the corresponding formation of
domains. Schweiz. Mineral. Perrog. Mitt. 50, 514-525.
LEADBETTER A. J. and WRIGHT A. C. (1972) Diffraction
studies of glass structure I. Theory and quasi-crystalline
model. J. Non-crysr. Solids 7, 2336.
LOWENSTEIN W. (1954) The distribution of aluminum in
the tetrahedra of silicates and aluminates. Am. Mineralo-
gist. 39, 92-96.
MASSONC. R. (1968) Ionic equilibria in liquid silicates.
J. Am. Ceram. Sot. 51, 13+143.
Mozzr R. L. and WARRENB. E. (1969) The structure of
vitreous silica. J. Appl. Crysrallogr. 2, 164-172.
NOCKOLDSS. R. (1954) Average chemical compositions of
some igneous rocks. Geol. Sot. Am. Bull. 65, 1007-1032.
NUKUI A., MORIKAWAH., IWEI S. and TAGAI H. (1972)
Study of molten silicas by X-ray diffraction (abstr.). Acra
Crysr. AZ8, S-128.
PEACORD. R. (1968) A high temperature single crystal dif-
fractometer study of leucite (K,Na)AISi,O,. Z. Krisr.
127, 213224.
7s
PERROTTA A. J. and SMITHJ. V. (1965) The crystal structure
of kalsilite, KAISiO.,. Mineral May. 35, 588-595.
PHILPOTT~A. R. (1971) Density, surface tension and visco-
sity of the immiscible phase in a basic alkaline magma.
Lirhos 5, 1-18.
PHILPOTTSA. R. (1976) Silicate liquid immiscibility: its
probable extent and petrogenic significance. Am. j. Sci.
276, 1147-l 177.
PRETNAR V. G. (1968) Beitrag zur ionen Theory der
silikat Schmeltzen. Ber. Bunsenyes. Phys. Chem. 72,
733-778.
PREWITTC. T., SUENOS. and PAPIKEJ. J. (1976) The crystal
structures of high albite and monalbite at high tempera-
tures. Am. Mineralgist. 61, 12131255.
RANDALLJ. T., R~OKSLIYH. P. and CARPER B. S. (1930)
X-ray diffraction and the structure of vitreous solids I.
2. Krisr. 75, 196-216.
RIEBLINGE. F. (1966) Structure of sodium aluminosilicate
melts containing at least 50 mole y, of SiOl at 1500°C.
J. Chem. Phys. 44, 2857-2865.
REIBLINGE. F. (1968) Structural similarities between a
glass and its melt. J. Am. Ceram. Sot. 51, 143-149.
ROBIER. A., HEMINGWAY B. S. and WII.SONW. H. (1978)
Low temperature heat capacities and entropies of
feldspar glasses and anorthite. Am. Mineralogist 63,
109-123.
SCHAIRERJ. F. hnd BROWNN. L. (1956) The system Na,O-
A1,03-Si02. Am. J. Sci. 254, 12%195.
SHAW H. R. (1972) Viscosities of magmatic silicate liquids:
An empirical method of prediction. Am. J. Sci. 272,
87&892.
SIMMONS W. B. and PEACORD. R. (1972) Refinement of
the crystal structure of a volcanic nepheline. Am. Miner-
alogist 57, 1711-1719.
TAYLORM. (submitted) A quasi-crystalline model for inter-
oretation of radial distribution functions: theorv and ex-
perimental confirmation. MS submitted to J. Appl. Crys-
tallogr.
TAYLORM. and BROWNG. E. (1976) A radial distribution
study of amorphous NaAlSi,O,, KAISi,O, and CaAI,
Si20s. Geol. Sot. Amer. 1976 Ann. Mrg., Absrr. Prog.
8, 1133.
TAYLORM. and BROWNG. E. (submitted) Structure of sili-
cate mineral glasses: III. NaAiSi,O, supercooled liquid
at 805°C and the effect of thermal history. MS submitted
to Geochim. Cosmochim. Acra.
Toop G. W. and SAMISC. S. (1962) Activities of ions in
silicate melts. Trans. AIME 224, 878-887.
UBBELOHDEA. R. (1965) Melting and Crystal Srrucrure.
Clarendon Press.
URNES S. (1969) Studies of the sodium distribution in
sodium silicate glasses by the chemical difference
method. Phys. Chem. Glasses 10, 69-71.
URNES S. (1972) X-ray diffraction studies in the system
Na*O-Al,O,-SiO*. Phys. Chem. Glasses 13, 77-78.
WAINWRIGH~J. E. and STARKEYJ. (1971) A refinement
of the structure of anorthite. Z. Krisr. 133, 77-84.
WARRENB. E. (1969) X-my Diflracrion, Addison-Wesley.
WINCHELLA. N. and WINCHELLH. (1964) The Microscopic
Characteristics of Artificial Inorganic Solid Substances:
Oprical Properties ofArrijcia/ Minerals.
Academic Press.
WRIGHTA. C. (1974) The structure of amorphous solids
by X-ray and neutron diffraction. Adc. Strut. Res. b)
DiJf Merh. 5, l-84.
WRIGKT A. C. and LEADBETTER A. J. (1976) Diffraction
studies of glass structure. Phys. Chem. G/asses 17,
122-145.
WYART J. (1940) gtude cristallographique d’une leucite
artificielle. Structure atomique et symttrie du minbral.
Bull. Sot. Franq. Min. 63,. 5-17.