EMU Notes in Mineralogy, Vol.

3 (2001), Chapter 3, 37–69

Thermodynamic modelling of solid solutions

JIBAMITRA GANGULY

Department of Geosciences, University of Arizona,

Tucson, AZ 85721, USA;
e-mail: ganguly@geo.arizona.edu

Introduction
One of the major goals of petrology is to retrieve values of the intensive parameters,
such as pressure (P), temperature (T) and fluid composition, under which the major
mineralogical properties of a rock were established, along with the time scales of
evolution of the mineralogical properties. The basic approach in the retrieval of the
intensive properties involves comparison of the mineralogical assemblages and the
mineral compositions of the rock with the phase equilibrium constraints. The latter are
calculated from the internally consistent thermochemical properties of the
stoichiometric end members or determined in the laboratory on relatively simple
systems, usually involving only the end-member phases, and then corrected for the
effects of the compositional departures as observed in a specific natural assemblage. In
addition, many mineral pairs (e.g. garnet and biotite) respond to changes of P–T
conditions, especially temperature, by continuous ion exchange reactions, and thus
register the P–T condition or P–T history of the rock in their compositions. The ion-
exchange reactions are also calibrated in the laboratory on relatively simple systems,
but require corrections for the effects of additional components that enter into solid
solution in the minerals in natural environments. The corrections for the compositional
effects rely critically on the thermodynamic mixing properties of the components in the
mineral solid solutions and fluid phase which are involved in a specific reaction (e.g.
Ganguly & Saxena, 1987). There has, thus, been a sustained effort over the last few
decades on the determination of thermodynamic mixing properties of phases in
geologically important systems.
In this review, I will summarise the general concepts of thermodynamic solution
theory and a number of macroscopic models that have been used in the treatment of
experimental data on mineral solid solutions. Some of the solution models were
originally developed for polymer and liquid solutions, but are also applicable to oxide
and solid solutions. For reasons of both time and space, I have not included order-
disorder theory specifically, although some aspects of the thermodynamics of minerals
showing such behaviour may also be treated within the framework of the models
discussed in this review, as will be pointed out in the appropriate sections. I have
assumed that the reader has a working knowledge of thermodynamics, including the
fundamentals of thermodynamic solution theory. Thus, only a brief discussion of the
38 J. Ganguly

solution theory has been included in this review in the spirit of recapitulation and re-
emphasis of some of the important points. The presentation of the topics follows a
systematic order in that the concepts presented in a particular section builds on what has
been presented in the preceding sections. The primary aim here has been to convey the
general physical ideas and thermodynamic concepts underlying various mixing models.
Consequently, in many cases the reader will have to consult the appropriate references
cited to find the equations that are needed for the use of a specific model, or derive these
equations following the methods that I have outlined.
To conform to the modern usage of terminologies (e.g. Hillert, 1998), and also for
the sake of brevity, Gibbs free energy will be referred to as Gibbs energy and a partial
molar quantity as simply a partial quantity. In the following list of symbols, the old
terminologies are shown within the parentheses.

Selected list of symbols and abbreviations:

ai and a i0 activity of the component i in a solution and in the pure state, respectively
D diffusion coefficient
G and Gm total and molar Gibbs (free) energy of a solution, respectively
Gi partial (molar) Gibbs (free) energy of a component i

G0(rec) the standard state Gibbs (free) energy change of a reciprocal reaction
L Avogadro’s number
ni and N number of moles of the component i and the total number of moles of
components in a solution, respectively
Q activation energy of diffusion
qi contact factor of the component i in the quasi-chemical formulation
Xi mole fraction of a macroscopic component i
Xij a binary mole fraction defined by the normal projection of a
multicomponent composition onto the binary join i–j
Xi a compositional function used to express the activity of a component i
X composition
j atomic fraction of j within a crystallographic site (usually specified by a
left superscript, e.g. Ij)
ji local mole fraction of a component j around a central component i

Ymxs an excess molar property of a solution

Ymmix a molar property of mixing of a solution
Y i* and Y 0i a property at the standard state and at the pure state, respectively, of the
component i
Z nearest neighbour coordination number of a central atom within a
sublattice
 i activity coefficient of a component i related to the compositional function
X i, i.e.  i = ai/ i
i and  (i) activity coefficient of the component i
µi and µ(i) chemical potential of the component i
 standard deviation.
 Thermodynamic modelling of solid solutions 39

i site fraction of the component i in a polymer solution

i local volume fraction of the component i

Ab albite (NaAlSi3O8)
An anorthite (CaAl2Si2O8)
CaTs calcium-Tschermak pyroxene (CaVIAlIVAlSiO6]
Di diopside (CaMgSi2O6)
En enstatite (MgSiO3)
Grs grossular (Ca3Al2Si3O12)
Kfs potassium feldspar (KAlSi3O8)
Ol olivine
Prp pyrope (Mg3Al2Si3O12)
Spl spinel

DQF Darken’s quadratic formulation

NRTL non random two liquid model
QC quasi-chemical model
UNIQUAC universal quasi-chemical model

Thermodynamics of solutions: a brief outline

In this section, I will present a brief summary of the thermodynamic formalisms relating
to the properties of a solution so that the subsequent sections can be followed in a self-
contained manner. For more detailed discussions, the reader is referred to Ganguly &
Saxena (1987) and Hillert (1998).

Gibbs–Duhem relation and partial quantities

At a constant P–T condition, the Gibbs energy (G) of a solution is given by
G = niµi, (1)
where ni and µi are, respectively, the number of moles and chemical potential of the
component i. Thus,
dG = nidµi + µidni. (2)
Now, since we also have
dG = – SdT + VdP + µidni, (3)
where S and V are the total entropy and total volume of the solution, respectively, we
obtain, on equating, at constant P–T condition, the right hand terms of the last two
equations
nidµi = 0 (4a)
or
Xidµi = 0, (4b)
40 J. Ganguly

where Xi is the mole fraction of the component i. This expression, which is known as
the Gibbs–Duhem relation, provides a relationship among the change of chemical
potentials of the components in a solution at constant P–T condition.
The partial property, Yi (e.g. partial volume Vi), of a component in a solution at a
fixed composition (X) is the rate of change of the corresponding total property of the
solution, Y (e.g. total volume, V), with respect to the change of the number of moles of
that component when P, T and the number of moles of all other components are held
constant. The chemical potential, µi, of a component at a fixed P–T condition is
equivalent to its partial Gibbs energy, Gi. However, using Equations 1 and 4b (i.e. the
Gibbs–Duhem relation), and some manipulations (e.g. Darken & Gurry, 1953), the
chemical potential of a component in a binary solution can be expressed in terms of the
molar Gibbs energy, Gm, as

. (5)

A relation analogous to Equation 5 also holds for a multicomponent solution if the

derivative of Gm with respect to the Xi is taken at constant P,T and constant relative
amounts of all components other than i, i.e. at constant Xj : Xk : Xl : … (Darken, 1950).
This multicomponent expression is often referred to as the Darken equation, and has
been applied by Sack & Loucks (1985) and Ghiorso (1990) to the problems of
multicomponent mineral solid solutions.
An alternative to the Darken equation for the derivation of partial properties in a
multicomponent solution is due to Hillert (e.g. Hillert 1998), which is as follows

G m G
 i  G m  (1  X i )  m (6a)
X i k
i X k

or, collecting the first derivative term within the summation,

G m G
i  Gm   m. (6b)
X i k X k

Here the derivative terms with respect to the atomic fraction of a component are taken
at either constant atomic fractions of all other components, or of all other (n – 2)
independent components in an n component system. The equivalence between the two
procedures follows from the fact that if one component, say the n-th component, is
chosen to be the dependent component, then

 Y   Y   Y 
     , (7)
 X i  i  n , j  X i  j
i  X n  j
n
where the subscripts outside the parentheses refer to the atomic fractions of the specified
components (i.e. i Xi), and j' refers to the atomic fractions of all other components
 Thermodynamic modelling of solid solutions 41

except those of i and n. Here Y is any property which is a function of the Xi-s. For the
derivation of partial properties in a multicomponent system, Hillert’s expression is
simpler to deal with than Darken’s (and is incorporated in the commercial Thermo-Calc
program for phase equilibrium calculations: http://www.thermocalc.se).
Similar expressions also hold for other partial quantities. Thus, in general, we can
replace µi and Gm by the generalised partial property, Yi, and the generalised molar
property, Ym, respectively.

Activity and standard state

The chemical potential of a component (i) at a given pressure, temperature and
composition of the solution is expressed as
µi(P,T,X) = µ*i (P*,T,X*) + RTlnai
(8)
= µ*i (P*,T,X*)+ RTln(X i i).
Here µ*i (P*,T,X*) is the chemical potential of the component i at the temperature of
interest and at some reference pressure (P*) and composition (X*) of the solution, ai
is the activity of the component i in solution at the specified P–T–X condition, X i is
a conveniently chosen compositional function, and  i is the corresponding activity
coefficient. X i  1, and is defined completely by the content of the component i in
the solution. Note that X i is not necessarily the atomic fraction of Xi, but is related
to it in such a way that X i = Xi at the terminal compositions, i.e. at Xi = 0 and 1.0.
For example, in a solid solution of the type (A,B,…)m(C,D,…)nP, an appropriate
expression of the activity of the component AmCnP is [()m(C)n] (AmCnP), as
discussed below, where  stands for the site fraction of the specified species. In this
case X i (where i AmCnP) stands for the entire compositional term within the square
bracket.
The state defined at (P*,T,X*) is usually called the standard state. According to
Equation 8, the explicit expression or magnitude of  i depends on the choice of standard
state of the component and the compositional function X i. This is because of the fact
that µi (P,T,X) is an absolute quantity so that only two of the three parameters on the
right hand side of Equation 8 may be chosen independently. It also follows from this
equation that the activity of a component in its standard state is unity.

Laws of dilute solutions

In the dilute range, the activity of a solute (i) that is actually present in a solution is
found to obey Henry’s law, which may be stated as follows.
LtXi  0, ai = KHXi, (9)
where KH is a constant at a fixed P–T condition. If i represents a strong electrolyte
(e.g. HCl, which dissociates almost completely to H+ and Cl– in aqueous solution),
then Xi in the above expression is replaced by (Xi)n, where n is the number of species
to which the solute dissociates (e.g. as XNaCl  0, aNaCl  (XNaCl)2) (see Ganguly &
Saxena, 1987, for the proof).
42 J. Ganguly

In the compositional range in which the activities of all solute components obey
Equation 9, the solvent (j) activity obeys Raoult’s law, which may be stated as
LtXj  1, aj = aj0(P,T)Xj, (10)
0
where a (P,T) is the activity of the pure component at the P–T condition of interest.
j
It is unity when the standard state of the component i is chosen to be the state of pure
component at the P–T condition of interest.
Consider now a two-site (I and II) solid solution of the type I(A,B,…)mII(C,D,…)nP,
e.g. garnet: (Fe,Mg,…)3(Al,Cr,…)2Si3O12. In this case, as Xj  1, the activity of the
molecular component AmCnP ( j) will approach [(1A)m(IIC)n], when pure AmCnP at
the P–T condition of interest is chosen to be the standard state of the molecular
component, i.e.
LtXj  1, aj = [(1A)m(IIC)n] Xj (11)
Although Henry’s law and Raoult’s law were proposed independently, one follows from
the other because of the Gibbs–Duhem relation (Eqn. 4b).

Mixing and excess functions

Using Equations 1 and 8, one obtains the following expression for the molar Gibbs
energy, Gm, of a solution.
Gm =  Xi µi + RTXilnXi + RTXiln i
(12)
=  Xi µi +
Gmmix.
The first term on the right represents a mechanical mixing term of the standard state
chemical potentials, whereas
Gmmix represents the deviation of the molar Gibbs energy
of the solution from that of the mechanical mixture. Imposing the laws of dilute
solutions (Eqns. 9 and 10) on the above relation, it can be shown (e.g. Ganguly &
Saxena, 1987) that the G vs. X curve must be convex downwards near the terminal
regions (dG/dXi  –as Xi  0).
A solution is said to be thermodynamically ideal when  i = 1 for all components.
Thus, the Gibbs energy of an ideal solution is always less than that of the mechanical
mixture of the end-member components. The difference between the Gibbs energy of a
real solution and that of an hypothetical ideal solution is known as the excess Gibbs
energy of mixing of the solution,
Gmxs. Thus, we have

Gmxs = RT  Xiln i. (13)
The other properties of a solution follow from Equation 12 through appropriate
thermodynamic operations, viz., H = (G/T)/(1/T), S = –G/T and V = G/P (note
that since G = H – TS, it is not necessary to differentiate G to derive both H and S).
Thus, one obtains the following expression for the molar entropy of mixing in an
ideal solution

Smmix(ideal) = – RXilnXi. (14)
 Thermodynamic modelling of solid solutions 43

By definition, the quantity RTlni represents the excess chemical potential (or
excess partial Gibbs energy) of the component i. Consequently, from the definition of a
partial property
 
G xs 
RT ln  i   . (15)
 n i  P ,T , n j
n i

In terms of the molar property, we can obtain RTlni from Equation 5, or its
multicomponent extension due to Darken (1950), and from Equation 6 on simply replacing
Gm by
Gmxs. The latter operation is more convenient for a multicomponent solution.

Reciprocal solutions
Let us now consider a two-site (I and II) solid solution such as I(A,B)mII(C,D)nP in which
there is no stoichiometric relation between the substitutions in the two sites, that is the
ratio A/B is independent of the ratio C/D. This type of solutions are known as reciprocal
solutions. An example is the two-site garnet solid solution (Fe,Mg)3(Al,Cr)2Si3O12, where
the inert part Si3O12 corresponds to P. The molar Gibbs energy of such a two-site binary
reciprocal solution may be expressed using the reference surface illustrated in Figure 1
(Hillert & Staffansson, 1970; Wood & Nicholls, 1978; Hillert, 1998) according to
Gm = [ACG0(AmCnP) + BCG0(BmCnP) + ADG0(AmDnP)
(16)
+ BDG0(BmDnP)] + [mRT I(ilni) + nRT II(ilni)] +
Gmxs ,
where the summation is carried out separately for each site indicated by a left hand
superscript. The collection of terms within the second square brackets represents
–TSmix(ideal). This approach of representing Gm of a reciprocal solution with reference
to the Gibbs energies of the end-member compounds, whether these compounds are real
or hypothetical, has been called the compound energy model by Hillert and co-workers
(e.g. Andersson et al., 1986; Hillert, 1998).

0
G AmDnP
BmDnP

AmCnP BmCnP

Fig. 1. Illustration of the Gibbs free (G0) surface defined by the mechanical mixtures of end-member
components in a two-site binary reciprocal solid solution, (A,B)m(C,D)nP. The bounding binaries define the
G0 for the mechanical mixtures of end members in one-site solution. The ruled surface is non-planar.
44 J. Ganguly

Assuming that the interactions within each site are ideal, in which case
Gmxs in
Equation 16 is zero, the chemical potential of an end-member component in a binary
reciprocal solution is given by
µ(AmCnP) = µ0(AmCnP) +
G0(rec)(1 – A)(1 – C) + mRTln(A) + nRTln(C),
(17)
where µ0(AmCnP) is the Gibbs energy of the pure component AmCnP at the P–T condition
of interest and
G0(rec) is the Gibbs energy change of the homogeneous reciprocal
reaction
A m C n + B m D n = A m D n + B m C n. (18)
(It is obvious that in Equation 17, the standard state of a component is assumed to be
the state of the pure component at the P–T condition of interest.)
Equation 17 was first derived by Flood et al. (1954), and has subsequently been
derived by others (e.g. Blander, 1964; Hillert & Staffansson, 1970; Wood & Nicholls,
1978; Hillert, 1998). This equation, and its extension to multisite-multicomponent
solution, can be derived in a systematic way by using Equation 6, and carrying out the
differentiations for the appropriate components within each sublattice (Wood &
Nicholls, 1978; Sundman & Ågren, 1981; Hillert, 1998). Thus, for example, using
Equation 6b, we have for the component AmCnP in the two-site solution,

G m G m G
 A m Cn P  G m    m , (19)
X A X C k X k

where the last summation is carried over both sublattices (the corresponding expression
using Equation 6a should be obvious).
Equation 17 highlights an important property of a reciprocal solution, i.e. it
behaves non-ideally (in the sense that the chemical potential of a component cannot be
determined completely from a knowledge of the content of that component in the
solution), even when the interactions within the individual sites are ideal. Comparing
Equations 8 and 17, we then write the following expression of the activity coefficient of
the component AmCnP, when the intrasite interactions are ideal:
RTln (AmCnP)rec =
G0(rec)(1 – A)(1 – C), (20)
where the subscript rec is used to highlight that this is a component of the overall
activity coefficient that is simply due to the reciprocal nature of the solution. The above
equation may be written in a general form as
RTln (im jnP)rec = ± (1 – i)(1 – j)
G0(rec), (21)
where the positive sign holds when im jnP is a reactant component, and the negative
sign holds when it is a product component of the reciprocal reaction.
Multicomponent extension of the above expression can be found in Wood & Nicholls
(1978) and Hillert (1998), or derived from Equation 19 following the procedure
outlined above.
 Thermodynamic modelling of solid solutions 45

Comparing the statistical-mechanical and thermodynamic derivations of the activity

of a component in a reciprocal solution, Førland (1964) suggested that the entropy change
of a reciprocal reaction should be very small, which implies that the
G0 of a reciprocal
reaction should be quite insensitive to temperature change. This fact was utilised by
Liermann & Ganguly (1999) to model the (reciprocal) effect of the variation of Al/Cr ratio
in spinel on the (Fe,Mg) fractionation between orthopyroxene, (Fe,Mg)SiO3, and spinel
(Fe,Mg)(Al,Cr)2O4. Natural data on the Fe–Mg distribution coefficient, KD(Fe–Mg),
between olivine, (Fe,Mg)2SiO4, and spinel show a systematic variation with Cr/(Cr+Al)
content of spinel in an isothermal suite of metamorphic rocks (Evans & Frost, 1975). These
data show a linear relation between lnKD vs. Cr(spinel) that can be derived by expressing
the activity coefficients of FeAl2O4 and MgAl2O4 components in spinel according to
Equation 21, and assuming ideal mixing between Fe and Mg in both spinel and olivine
(Ganguly & Saxena, 1987), viz., RTlnKD = RTlnK +
G0(rec)XCr(Spl) where KD =
(Mg/Fe2+)Ol/(Mg/Fe2+)Spl, where K is the equilibrium constant of the exchange reaction
Fe2+–Ol + Mg–Spl Mg–Ol + Fe2+–Spl.
When the sites behave non-ideally, the overall activity coefficient of a component
has to be expressed by a combination of Equation 21 and additional terms reflecting the
non-ideal interactions within the individual sites including their mutual inter-
dependence. The reason for the cross interactions between the sites in a solid solution
may be appreciated by noting that the bond distance within site I may be affected by a
change of composition in site II, and vice versa.

Ionic solution model

The classical and statistical thermodynamic basis of the so-called ionic solution model,
which has been used extensively in the treatment of mineral solid solutions, was discussed
in detail by Ganguly & Saxena (1987). Thus, only a brief exposition of the model is given
in this section. It should be noted at the outset that all expressions for the activity of a
component are equivalent as long as these are based on the same standard state of the
component. The explicit expression of the activity coefficient of a component in a
solution, therefore, depends on the form of the activity expression. The ionic solution
model provides a rational approach towards the development of such expressions.

Single-site solutions
The ionic solution model for a solid solution involving substitution in a single type of
site may be illustrated by considering a solution such as olivine, (Fe,Mg)2SiO4.
According to this model, the activity of an end-member component, AmF (e.g. Mg2SiO4)
is given by
a(AmF) = (AA)m, (22)
where A is the atomic fraction of A within its specific site, and A is the activity
coefficient of the ion A reflecting non-ideal interactions with other ions within the same
site. A may be viewed as the activity coefficient of the normalised component AF1/m
(e.g. MgSi0.5O2). Note that A equals the mole fraction of the molecular component AmF
46 J. Ganguly

(e.g. Mg = X(Mg2SiO4) in olivine). A simple justification of this relation lies in the
expression for the configurational entropy of mixing, which can be derived from the
Boltzmann relation (e.g. Ganguly & Saxena, 1987), assuming the simplest possible model
for the distribution of atoms, namely that these are distributed randomly in the solid
solution within their structural sites. For a solution of the type (A,B,…)mP, this relation is

Sm(conf) = – mRilni
(23)
= – Riln(i)m.
Comparing the above equation with the simplest thermodynamic expression for the
entropy of mixing, namely that of an ideal solution as given by Equation 14, we obtain
Xi = (i)m. (24)
Thus, in general,
a(AmP) = (XAA) = (AA)m, (25)
which yields the ideal behaviour, a(AmP) = (A) , as X(AmP)  1.
m

Reciprocal and disordered solutions

Comparing Equation 17 with the expression of µ(AmCnP), written in terms of the
activity of the component AmCnP, i.e.
µ(AmCnP) =µ0(AmCnP) + RTlna(AmCnP), (26)
we obtain for a reciprocal solution with ideal intrasite interactions
a(AmCnP) = [(A)m(C)n](AmCnP), (27)
where  (AmCnP) is given by Equation 21. If, however, the intrasite interactions were
non-ideal, but the sites behave independently of one another, then we could write (Wood
& Nicholls, 1978)
 (AmCnP) = [(IA)m(IIC)n]exp[± (1 – A)(1 – C)(
G0(rec)/RT)], (28)
where the sign convention is the same as in Equation 21. The interdependence between
the mixing properties in the two sites would require additional terms, or may be
absorbed in some cases within the site activity coefficient terms.
As discussed by Ganguly (1982) and Ganguly & Saxena (1987), the above
equations are also applicable to solid solutions of the type I(A,B)mII(A,B)nP in which the
species A and B disorder or fractionate between the structural sites I and II. An example
of this type of solution is orthopyroxene, M1(Fe,Mg)M2(Fe,Mg)Si2O6, in which Fe and
Mg disorder between the two non-equivalent octahedral sites, M1 and M2 (Ghose,
1982). (The state of the Fe–Mg disordering in orthopyroxene is an important indicator
of the cooling rates of the host rocks, e.g. Ganguly et al., 1982.) In the absence of
adequate data on the thermodynamic mixing properties, it has been a common practice,
however, to express the activity of an end-member component in such a disordered two-
site solid solution in terms of what has been known as two-site ideal model, i.e.
a(AmAnP) = (IA)m] (IIA)n. (29)
 Thermodynamic modelling of solid solutions 47

By comparing with Equations 27 and 28, it should be noted that this expression not only
implies that I A = II B = 1, but also that
G0(rec)/RT = 0. In addition, because of
stoichiometric relation between the atomic fraction of A in the bulk crystal, XA, and
within the two sites, i.e. XA = p(IA) + q(IIA), where p = m/(m + n) and q = (1 – p), the
right hand term of Equation 29 is  XA, the equality holding only in the case of complete
disorder, i.e. when IA = IIA. Thus, an activity expression in terms of two-site ideal
model implies a negative deviation from ideality for the mixing of the macroscopic
components, i.e. a(AmAnP) < XA. For example, the often used expression a(MgSiO3) =
(M1Mg)1/2(M2Mg)1/2 for orthopyroxene solid solution, M1(Fe,Mg,…)0.5M2(Fe,Mg,…)0.5,
implies that a(MgSiO3) < XMg, which is not compatible with the available calorimetric
and phase equilibrium data (Stimpfl et al., 1999).

Coupled substitutions
There are many solid solutions that require coupled substitutions of ions in order that
the macroscopic electrical neutrality can be preserved. An example is plagioclase
feldspar, which has the end-member components NaAlSi3O8 and CaAl2Si2O8, involving
the coupled substitution (Na+Si4+) (Ca2+Al3+). When local electroneutrality is
maintained in the solution, a replacement of Na+ by Ca2+ is accompanied by a
replacement of the nearest neighbour Si4+ by Al3+ in the tetrahedral site. In this case, the
expression for the activity of an end-member component (e.g. NaAlSi3O8) in terms of
the ionic solution model should consider X in Equation 22 as the mole fraction of the
coupled species (e.g. XNaSi), and equate the exponent m to the number of moles such
species per mole of the solid solution. Thus, recasting the formula for plagioclase solid
solution as (NaSi,CaAl)(AlSi2O8),we have
a(Ab) = XNaSiNaSi XAbAb. (30)
The interested reader is referred to Ganguly & Saxena (1987) for further discussion on
this problem, especially when the substitution of Na for Ca2+ is partly or completely
decoupled from the substitution in the tetrahedral site at relatively high temperature.

One-site binary mixing models

The mixing models described in this section apply to solid solutions in which the
substitutions are restricted to one site or in which the substitutions in different sites
are coupled so that a site atomic fraction equals a molecular fraction, such as in the
case of the plagioclase solid solution. When the solid solution involves independent
multiple site substitutions, such as in garnet, the mixing properties within each site
may also be treated in terms of the models described in this section. The different
solution models deal with the different types of representation of the excess
thermodynamic quantities as functions of composition. The fundamental expression
is that of
Gxs m as a function of composition, from which all other excess
thermodynamic properties may be derived through standard thermodynamic
operations, as discussed above (e.g. Eqns. 14 and 15). I will first deal with binary
solutions and then ternary and higher order solutions.
48 J. Ganguly

Guggenheim or Redlich–Kister, Simple Mixture and Regular Solution models

Guggenheim (1937) suggested that the molar excess Gibbs energy of mixing of a binary
solution may be represented by the polynomial expression,

Gmxs = X1X2[A0 + A1(X1 – X2) + A2(X1 – X2)2 + ….], (31)
where the A’s are constants at a fixed P–T condition. This polynomial satisfies the
requirement that Gmxs must vanish at the terminal compositions (i.e. X1 = X2 = 0). Operating
on this relation according to Equation 5, and noting that RTlni =
Gixs, we have
RTln1 = X 22[A0 + A1(3X1 – X2) + A2(X1 – X2)(5X1 – X2) + …] (32a)
and
RTln2 = X 12[A0 – A1(3X2 – X1) + A2(X2 – X1)(5X2 – X1) + …]. (32b)
These expressions for the activity coefficients were first derived by Redlich & Kister
(1948) and are usually referred to as Redlich–Kister relations. Somehow, even the
Guggenheim polynomial is often referred to as the Redlich–Kister expression of excess
Gibbs energy, which does not seem justified (this is possibly due to the fact that these
authors recommended an extension of the Guggenheim polynomial to the ternary
system, as discussed below).
When the A constants with odd subscripts, A1, A3 etc. are zero, the
Gmxs becomes
symmetric with respect to composition, and are, thus, called symmetric solutions by
Guggenheim (1967). The simplest functional form of a nonideal solution is the one in
which all but the first constant in Equation 31 is zero. In this case,
Gmxs has a parabolic
symmetry with respect to composition. Guggenheim (1967) called this type of solution
a Simple Mixture, as it represents the simplest form of deviation from ideality.
Conventionally, A0 is replaced by the symbol W or WG when the solution behaves as a
Simple Mixture so that, according to the last two equations

Gmxs = WGX1X2, (33)
and
RTlni = WG(1 – Xi)2, (34)
where i is either component 1 or 2 (see Eqn. 51 for the microscopic interpretation of W)
The dependence of WG on P and T are given, respectively, by
 W G  1  
G xs 
V xs
   
 P X  P , (35)
 T 1X 2  T X 1 X 2

 W G  1  
G xs 
S xs
   
 T  T . (36)
 P X1X 2  P X1X 2

Hildebrand (1929) introduced the term Regular Solution for the type of solutions
which obey Equation 33, but in which the interaction parameter WG is independent of P
 Thermodynamic modelling of solid solutions 49

and T. Thus, a regular solution is a special class of a simple mixture with ideal volume
and entropy of mixing. However, this distinction is not strictly followed in modern
usage in that any solution which obeys the functional form of Equation 33 is often
referred to as regular solution. I will also use the term regular solution in the sense of
simple mixture. Following Thompson (1967), WG is commonly decomposed into
enthalpic (WH), entropic (WS) and volumetric (WV) terms. It can be shown that

bar bar . (37)

Assuming WV to be independent of P, the last term is often written as a PWV term in

geological literature, since, usually P >> 1 bar under geological conditions. The
temperature dependence of WH and WS is related to the excess heat capacity of mixing,
which is due to non-linear change of vibrational properties as a function of composition.
Due to an extreme paucity of heat capacity data for solid solutions, the WH and WS terms
are almost invariably assumed to be constants. However, as shown by Vinograd (2001)
from analysis of spectroscopic data in Prp–Grs and Di–CaTs solid solutions, there could
be significant temperature dependence of these parameters.
When the thermodynamic mixing properties of solid solutions appear symmetrical
with respect to composition, or show moderate deviation from symmetric behaviour, these
are usually fitted by regular or subregular (see below) solution models. However, the data
are rarely good and sufficient enough to permit discrimination if the symmetry is truly
parabolic in nature. Recently, Stimpfl et al. (1999) carried out a detailed study, by single
crystal X-ray diffraction, of the distribution of Fe2+ and Mg between the non-equivalent
octahedral sublattices, M1 and M2, in essentially binary orthopyroxene solid solution,
(Fe,Mg)SiO3, as a function of temperature. From these data, they calculated the
Smix of Fe
and Mg, assuming that the distribution is random within each sublattice. Their results show
that the
Smix is essentially symmetric with respect to composition, but the relation is not
parabolic. Instead, the best fit to the data requires two even parameters, A 0s and A 2s, in
Guggenheim’s polynomial expression, where the superscript s denotes terms related to the
expression of
Smxs, when expressed according to the form of Equation 31.

Subregular model
This is the simplest model for asymmetric solutions, and has been used most
extensively in the petrological and mineralogical literature. It represents a simple
extension of the regular solution model by making the parameter WG in Equation 33 a
simple function of composition as
WG(SR) = W 21G X1 + W 12GX2 (38)
so that

Gxs(SR) = (W 21GX1 + W 12GX2)X1X2, (39)
G
where SR implies subregular, and W is a function only of P and T. It is obvious that
ij
near the terminal regions X1 = 1 and X2 = 1, WG(SR) is approximated by W 21G and W 12G,
50 J. Ganguly

respectively. Thus, the subregular model is simply a weighted average of two regular
solution models fitted to the data near the two terminal segments of a binary solution.
Each subregular WGij may also be decomposed according to Equation 37.
The subregular formulation follows from Guggenheim’s polynomial expression
for
Gmxs, Equation 31, by truncating it after the second term and using the identity
A0 = A0(X1 + X2), which yields

Gmxs(SR) = [(A0 + A1)X1 + (A0 – A1)X2]X1X2 . (40)
G G
On substitution of W and W for the collection of constants within the first and second
12 21
set of parentheses, respectively, Equation 40 reduces to the standard subregular form.

Darken’s quadratic formulation

Darken (1967) pointed out that when the activity coefficient of a solvent component
(say component 1) obeys the regular solution relation, as given by Equation 34, then the
Gibbs–Duhem relation (Eqn. 4b) only requires that the activity coefficient of the solute
component must obey the relation
RT ln2 = WG(1 – X2)2 + I, (41)
where I is an integration constant. In order that the component 2 conforms to Raoultian
behaviour (Eqn. 10) as X2  1, the integration constant must be zero when 1 obeys the
regular behaviour over the entire range of composition. If 1 conforms to the regular
solution property over a restricted compositional range only near the terminal region 1,
then 2 will conform to Equation 41 over the same compositional range with I
0. Thus,
according to Equation 13,
Gmxs near the terminal region 1 is given by,

G mxs = WGX1X2 + IX2. (42)
This is known as Darken’s Quadratic Formulation (DQF).
Using Equation 12 and rearranging terms, the molar Gibbs energy of a solution
obeying DQF in the terminal region 1, where the solvent component 1 obeys regular
solution behaviour, is given by

Gm = X1G01 + X2(G20 + I) +
Gmmix

(43)
= X1G01 + X2G2' +
Gmmix ,
where
Gmmix has the same expression as a regular solution, viz.,

Gmmix =
Gmmix(ideal) + WGX1X2 (44)
Thus, as noted by Powell (1987), a solution obeying DQF in the terminal region 1 may
be viewed as a regular solution between the real end member 1 and a hypothetical end
member whose molar Gibbs energy, G2', is given by that of the end member 2 plus the
value of the integration constant, I.
By analysing the experimentally determined activity coefficient data on a number
of liquid binary alloys, especially those in which Fe was the solvent, Darken (1967)
showed that while the solvent 1 followed regular solution behaviour up to a certain level
of addition of the solute component 2, the latter followed the relation described by
 Thermodynamic modelling of solid solutions 51

Equation 41, with I

0, over the same range of concentration. From these observations,
Darken (1967) suggested that it may be possible to treat many solutions in terms of the
Quadratic Formulation in the two terminal regions, each with characteristic values of
W and I. The behaviour of the intermediate compositional region would be more
complex since it has to make the transition from the quadratic properties of one terminal
region to those of the other.
For those solutions which conform to DQF in the two terminal regions, the
intermediate region could follow a relation that is a weighted average of those of the
terminal regions in much the same way as the expression of
Gmxs of a subregular
solution represents a weighted average of the regular solution expressions in the
terminal regions (Eqn. 39). In that case,
Gmxs of the intermediate segment of the solution
conforming to DQF in the terminal regions (Eqn. 42) is given by

Gmxs(1–2) = X1[W 21G X1X2 + I21X2] + X2[W 12G X1X2 + I12X1]

(45)
= X1X2[W 21G X1 + W 12GX2 + I21 + I12],
where the subscript ij represents the property of the terminal region j. Note that when
Xi  1, the
G xs
m is given by only the Xi[…] term after the first equality in the above
expression.
Powell (1987) analysed the available molar volume data of several binary
mineral solid solutions, and showed that the data in the two terminal segments are
better described by DQF than by regular solution model. His analysis is presented
below. Using Equation 45 and the identity I = G2' – G20 (Eqn. 43), we obtain the
following expression for
Vmix for the terminal region 1, if the solution obeys DQF
in that region.

(46)
0

or

. (47)

Similarly for the terminal region 2, we have

0 . (48)

On the other hand, if a solution obeys the subregular behaviour over the entire
compositional range, then using Equation 39, and following the same procedure, we have

V mixm
 W21V X 1  W12V X 2  W12V  X 1 (W21V  W12V ) . (49)
X 1 (1  X 1 )
 52 J. Ganguly

Thus, if a solution obeys the subregular behaviour, then the function on the left-hand
side of the above expression should change linearly with composition. Figure 2a shows
the data for the microcline (KAlSi3O8) and low-albite (NaAlSi3O8) solid solution (Kroll
et al., 1986) plotted in the manner suggested by Equation 49. It is evident that the data
do not describe the linear relation expected from the subregular model. Figures 2b and
2c show the same volumetric data plotted according to the forms of DQF in the terminal
regions 2 (Eqn. 48) and 1 (Eqn. 47), respectively. The expected linear relations (solid
lines) are followed in the terminal regions. [However, Powell (1987) seems to have
overextended the linear relation in the terminal region 1 (Fig. 1c). A more appropriate
linear fit to the data in this region should have been one with a smaller slope and better
satisfying the data near X = 1].
20
30
15
∆V m /XK(I – XK)

∆V m /XK

20 10
mix

10 5
mix

0
0

0 0. 2 0. 4 0. 6 0. 8 1. 0 0 0. 2 0. 4 0. 6 0. 8 1. 0
a b

30 Fig. 2. Volume–composition relation of micro-

cline (KAlSi3O8)–low albite (NaAlSi3O8) solid
∆V m /(I – XK)

20 solution displayed according to (a) subregular

model (Eqn. 49) and (b–c) Darken’s quadratic
formulation (DQF). V represents unit cell volume
mix

10 in Å3. In Figs. (b–c), the data are displayed

according to the DQF equations for the terminal
regions 2 and 1, respectively (see text). The
0 figure is reproduced from Powell (1987).
0 0. 2 0. 4 0. 6 0. 8 1. 0
c
In applying DQF to treat the mixing property data, one needs to be careful
about the quality of the data. Since the data are divided into three segments, two
terminal regions and one central region, there is greater flexibility in fitting the data,
which permits better conformity of relatively poor quality data to DQF than to the
subregular model. This point can be illustrated by considering the volumetric data
along the pyrope–grossular join. Powell (1987) showed that the available data are
much better described by DQF than by the subregular model. Since then, precise
volumetric data were obtained on this join by Ganguly et al. (1993) and Bosenick &
Geiger (1997). As illustrated in Figure 3, neglecting two anomalous measurements,
these data conform well to the subregular model.
 Thermodynamic modelling of solid solutions 53

4.0

3.0
V/X(1 – X)

2.0

1.0

0.0
0 0.2 0.4 0.6 0.8 1.0
Prp X(Ca) Grs

Fig. 3. Volume–composition relation of the pyrope (Mg3Al2Si3O12)–grossular (Ca3Al2Si3O12) solid solution

displayed according to the linear form suggested by subregular behaviour. V is the molar volume in cm3. The
data are from Ganguly et al. (1993; circles) and Bosenick & Geiger (1997; filled squares). The vertical bars
represent ± 1 centred on the symbols. The fitted line neglects the anomalous data marked by asterisks.

Quasi-chemical and related models

In the simple mixture model, the distribution of species in a solution has been
considered to be random, even though the pair-potential energies are different, thus
leading to nonzero enthalpy of mixing. However, this cannot be strictly correct since a
species would tend to be preferentially surrounded by the ones with which it has a
relatively stronger potential energy of interaction. The atomic distribution would be
effectively random at high temperature when the thermal energy per mole, RT, is
sufficiently high to prevent clustering of such species. Guggenheim (1952) sought to
remedy this logical problem with the simple mixture model by considering that in a
binary solution, the distribution of the 1–1, 2–2 and 1–2 pairs is related to the energy
change of the homogeneous chemical reaction
1–1 + 2–2 = 2 (1–2). (50)
The resultant thermodynamic mixing model is known as the quasi-chemical (QC)
model because of its appeal to a chemical reaction among the different pairs in the
solution, and representation of the equilibrium concentration of these pairs in much the
same way as the equilibrium concentration of components is expressed by an
equilibrium constant of a chemical reaction.
54 J. Ganguly

From consideration of the total potential energy of a lattice consisting of 1–1, 2–2
and 1–2 pairs, and neglecting the effect of long range forces, Guggenheim (1952)
introduced an interchange energy, W, according to
W = LZ[!12 – ½(!11 + !22)], (51)
where L is Avogadro’s number, Z is the nearest neighbour coordination number of the
atom 1 or 2, and !ij is the potential energy of interaction between the species i and j.
Here the term Z represents the coordination number of an atom within its specific
sublattice instead of the usual polyhedral coordination number around a central atom.
For example, in the solid solution between NaCl and KCl, each alkali atom has six
nearest neighbour Cl atoms (and vice versa). However, in Equation 51, Z must be
taken as 12, which represents the number of nearest neighbour alkali atoms
surrounding a central alkali atom in the crystal structure. WH in the simple mixture
model is exactly the same as the interchange energy defined above. It should be noted
that although Guggenheim neglected the effects of long range forces in the derivation
Equation 51, inclusion of these forces leads to a similar expression (see Vinograd,
2001), except that the Z[..] term is replaced by Z(k)[!12(k) – ½(!11(k) + !22(k))], where the
summation is carried out over k-nearest pairs (i.e. 1st nearest, 2nd nearest, 3rd nearest,
and so on). The contribution of the distant pairs may be important even though the
energy of interaction may decrease rapidly with distance since Z(k) could increase
rapidly with distance (Vinograd, pers. comm.).
In order to account for the mixing of molecules or atoms of different sizes,
Guggenheim (1952) also introduced parameters known as contact factors, q, which
represent the geometrical relation of an atom to another atom of different type in a nearest
neighbour site. The contact factors have the property that q1/q2  1 as either contact factor
tends to unity. (One can think of a number of relations between q1 and q2 that would satisfy
this limiting property. For example, Green (1970a) assumed q1/q2 = 1, whereas Fei et al.
(1986) assumed q1 + q2 = 1. Both relations satisfy the required limiting behaviour of the
ratio q1/q2.) Guggenheim (1952) showed that the deviation from a random distribution of
the species is given by a parameter ", which is defined as
1
)  2W & 2
"  1  4# 1# 2 '1  exp $ , (52)
(  ZRT %

where # is related to the contact factors according to

X 1 q1
#1  1  # 2  . (53)
X 1 q1  X 2 q 2

As the solution approaches a random distribution, i.e. W/RT  0, "  1, whereas for
positive (W > 0) and negative (W < 0) deviations from ideality, " > 1 and < 1,
respectively.
Within the above framework, Guggenheim (1952) derived the following QC
expression for the molar excess Gibbs energy of mixing in a binary solution.
 Thermodynamic modelling of solid solutions 55

G mxs Z ,2) X 1 q1 ln *"  # 1  # 2 +& ) X 2 q 2 ln *"  # 2  # 1 +& /, .

 1' $' $. (54)
RT 2 ,0( # 1 ( "  1) % ( # 2 ( "  1) % ,-

Using H = (G/T)/(1/T), one then has

)4X X W &  2W  ) X 1 q1# 2 X 2 q 2# 1 & .

H mxs  ' 1 2 $ exp .'  $ (55)
( " ( "  1) %  ZRT ( "  # 1  # 2 "  # 2  #1 %

The
Smxs can be derived from the relation G = H – TS.

The unknown parameters, q and W, for a binary system can be retrieved from
phase equilibrium or enthalpy of mixing data, if the quasi-chemical model provides an
adequate analytical representation of the data. From an analysis of NaCl–KCl solvus
data, Green (1970a) showed that qNa+/qK+ nearly equals the ratio of the cationic radii or
molar volumes of the two end members. However, Fei et al. (1986) failed to find any
such relation between the ratio of the retrieved contact factors and molar volumes of the
end members in the pyrope–grossular, diopside–Ca-Tschermak and diopside–enstatite
solid solutions.
Modifications and extensions of the quasi-chemical theory have been suggested by
Abrams & Prausnitz (1975), Green (1970a), Powell (1983) and Pelton & Blander
(1986). The Abrams–Prausnitz modification, which is known as the universal quasi-
chemical theory or UNIQUAC, involves only two adjustable parameters for a binary
solution and seems to have a wide range of applicability for liquid solutions. Its success
for solid solutions has not yet been tested. In the classic QC theory, the extrema in both
enthalpy and entropy of mixing appearing for the negative deviation from ideality
(which favours the formation of 1–2 pairs) are at X1 = X2 = 0.5, whereas such extrema
often occur at other compositions in real systems. (In a binary system,
Hmix exhibits
negative deviation from ideality with a “V” shaped form, whereas
Smix shows an
inverted “W” form in which the sagging of the central portion depends on the degree of
ordering; see Vinograd, 2001, Fig. 11.) This problem with the location of extrema was
remedied using a semi-empirical approach in the modification proposed by Pelton &
Blander (1986). Powell (1983) also discussed the incorporation of asymmetry in the QC
formulation. Green (1970a), who was the first to apply QC model to mineralogical
systems, developed an expression of
Gxs/RT for the QC model as a power series of
W/RT and of a function of the contact factors. The simple mixture or the regular solution
model follows in a straightforward way as a special case of Green’s equation. It shows,
as discussed by Ganguly & Saxena (1987), that in order for the QC model to reduce to
the simple mixture model, not only should W/RT be small, but also the species 1 and 2
should be sufficiently alike in shape and size so that their contact factors, q1 and q2, are
also similar. However, the magnitude of W/RT and the dissimilarity of the contact
factors must be intrinsically related in that W/RT cannot be a small quantity unless the
mixing units are sufficiently alike.
The simple mixture (or regular solution) and QC models follow as zeroth and first
approximations, respectively, of a powerful approach developed by Kikuchi (1951),
56 J. Ganguly

which is known as the cluster variation method. This method was applied by Burton &
Kikuchi (1984a, 1984b) to treat order-disorder in CaCO3–MgCO3 and Fe2O3–FeTiO3
solid solutions. An extended quasi-chemical model was developed by Truskinovskiy et
al. (1987) to successfully treat order-disorder relations in a variety of mineral solid
solutions. Vinograd (2001) discussed in detail the cluster variation method and applied
it to garnet and pyroxene solid solutions.

Athermal, Flory–Huggins, Wilson and NRTL (non-random two liquid) models

The statistical thermodynamic study of Fowler & Rushbrooke (1937) and calorimetric
measurements of Meyer and co-workers (e.g. Meyer & van der Wyk, 1944) showed that
molecules of different size and shape mix with a significant non-random distribution or
non-ideal entropy effect even when
Hmix = 0. This type of solution is known as an
athermal solution. Athermal behaviour is closely approximated, but not followed
exactly, by several polymer solutions in which the components differ in size but have
very similar energetic properties. However, as emphasised by Ganguly & Saxena
(1987), mineral solid solutions are not expected to show this type of behaviour as
substitutions of atoms of different size invariably lead to nonideal enthalpic effects
owing to the distortion of the lattice and nonlinear change in the bonding energies.
However, athermal solutions offer a starting point for the development of models which
have been successfully used to treat mineral solid solutions.
It was shown independently by Flory (1941, 1944) and Huggins (1941) that the
entropy of mixing resulting from the non-energetic solution of a polymer component (2)
in a monomer solvent (1) is given by

Smmix = – R(X1ln31 + X2ln32), (56)
where 31 and 32 are the fraction of sites occupied by the solvent and the polymer,
respectively. If there are N1 molecules of the solvent and N2 molecules of the polymer,
and there are p segments in a polymer molecule, then

N1 pN 2
31  , 32  , (57)
N1  pN 2 N1  pN 2

where N1 + pN2 are the total number of sites in the solution. It is assumed that each lattice
(or quasi-lattice) site is occupied by either a solvent molecule or a polymer segment.
When a lattice site is occupied by a polymer segment, the adjacent sites are occupied by
the rest of the segments so that each polymer molecule occupies p lattice sites.
Wilson (1964) extended the Flory–Huggins formulation to include the mixing
of molecules which differ not only in size but also in their energetic properties.
This extension involved calculation of the relative probabilities of finding
molecules of the two components around a central molecule or atom, say of the
type i, taking into account the energies of interaction of the i–j and i–i pairs, and
from that deriving expressions for the local volume fractions of the components
around the central component. Wilson assumed that the ratio of the “local mole
fractions” of the components i and j around a central component i is given by
 Thermodynamic modelling of solid solutions 57

 ji X j exp(  E ji / RT )
 , (58)
 ii X i exp(  Eii / RT )

where Eij is the molar interaction energy between i and j. The local volume fraction
of a component around a central component of the same type is then given by

Vi  ii
i  , (59)
Vi  ii  V j  ji

where Vi and Vj are the molar volumes of the components i and j, respectively.
Wilson used these local volume fractions in place of the overall site fractions in the
Flory–Huggins expression (Eqn. 56). This procedure leads to
(60)
with
V2 ) E  E11 &
4 12  exp '  12
RT $%
(61a)
V1 (

and
V1 ) E  E 22 &
4 21  exp '  12 $ . (61b)
V2 ( RT %

It should be noted that the local volume fractions in Wilson’s formulation do not
always add up to unity (Prausnitz et al., 1986). Also the Wilson expression has no rigorous
theoretical justification, but is rather an intuitive extension of the Flory–Huggins
formulation to account for the energetic effects on mixing. However, it has been
successfully applied to many binary systems (Orye & Prausnitz, 1965), and seems to have
some appeal in the treatment of multicomponent solutions as discussed later. On the other
hand, there are two important formal limitations of the Wilson expression (Wilson, 1964;
Prausnitz et al., 1986). First, it cannot produce a maximum in the ln vs. X relation. Second,
no values for the parameters 412 and 421 can be found that produce phase separation or
unmixing, that is produce a ‘hump’ (or a convex upwards segment) in the Gm vs. X curve
(which is always convex downwards near the terminal regions). In other words, there are
no values of these parameters which satisfy the condition 2Gm/X i2 < 0. Thus, if the
solution has a miscibility gap, the application of the Wilson equation must be restricted to
the P–T–X domain where the solution is continuous.
Renon & Prausnitz (1968) modified Wilson’s formulation so that it can produce
phase separation by introducing a correction factor, 512, as a multiplier of the energy
terms in Equation 58. This model, which is known as the Non Random Two Liquid
Model (NRTL), leads to the following expression for
Gmxs
) ( E  E22 )G21 ( E21  E11 )G12 &

Gmxs  X 1 X 2 ' 12  $ , (62)
( X 1  X 2G21 X 2  X 1G12 %
58 J. Ganguly

where
. (63)

Expressions for the other thermodynamic excess functions can be derived from
Equation 62, and are given in Prausnitz et al. (1986).
When 512 = 0, Gij = 1, and also the ratio of the local mole fraction reduces to that
of the bulk mole fractions (Eqn. 58). Under this condition, the
Gmxs in the NRTL model
reduces to

Gmxs =
E(X1X2), (64)
where, using E12 = E21,

E = 2E12 – (E11 + E22) (65)
Equation 64 is formally similar to the regular solution expression, Equation 33.
However,
E reduces to the expression for W in terms of pair potential energies, as
given by Equation 51, only if Eij = Z/2(L!ij), where Z is the number of nearest
neighbours of the components i and j around a central component of i or j. Thus, Eij
should be treated as a quantity proportional to the pair potential energy between i and j.
By comparing the NRTL and QC models, Renon & Prausnitz (1968) suggested that 512
should be similar to 1/Z. Consequently, 512 should be < 1. However, as discussed later,
the value of 5 retrieved from experimental mixing property data on mineral solid
solutions sometimes depart very significantly from the expected value of 1/Z.

Comparison of one-site binary mixing models

Fei et al. (1986) applied the Guggenheim, Subregular, QC, Wilson and NRTL models
to several binary mineral solid solutions for which calorimetric data for
Hmix were
available. These are diopside–Ca-Tschermak (Di–CaTs), diopside–enstatite (Di–En),
pyrope–grossular (Prp–Grs) and anorthite–albite (An–Ab) solid solutions. The Wilson
equation was found to give a much worse fit to the data than the other models, whereas
the Guggenheim polynomial using three constants, A0, A1 and A2, had the best overall
fit. The other three models also produced quite good fits to the data with the NRTL
model giving a slightly better fit. The 5 value for the NRTL model varied between 0.05
and 0.30, but did not show any correlation with the number of nearest neighbour cations
around a central cation in a solid solution series. For example, in both Di–En and
Prp–Grs solid solutions, each divalent cation has two nearest neighbour divalent
cations. From the analysis of Renon & Prausnitz (1968), as discussed above, the
expected value of 5 is 0.50, which is similar to the fitted value of 0.30 for the Prp–Grs
solid solution, but much different from that for the Di–En solid solution, which is 0.05.
Green (1970b) showed that while the phase diagram of the NaCl–KCl solid
solution may be fitted well by both subregular and quasichemical models, the retrieved
thermodynamic properties from the two different models were quite different. The
Hxs
and
Sxs values recovered from the fitted subregular parameters were much larger than
the measured values, while those recovered from the fitted QC parameters were much
 Thermodynamic modelling of solid solutions 59

closer to the latter. Thus, the success of fitting a limited amount phase equilibrium data by
a given model does not guarantee that the extracted thermodynamic mixing properties are
correct. One must simultaneously treat as large a data set as available, and also evaluate if
the theoretical basis of the model is appropriate for the specific solid solution being
modelled. Thus, for example, although the NRTL model provides a good fit to the
Hxs
data for the Di–En solid solution, the fact that the retrieved value is grossly different from
the value expected from theory may be a warning about the inadequacy of the NRTL model
for this solid solution.

Multicomponent solutions
An important practical problem in solution thermodynamics lies in the formulation of
the method of prediction of the properties of multicomponent solutions from those of
the bounding binaries. The problem has been discussed by a number of workers earlier
(e.g. Jiran & Jacob, 1983; Hillert, 1980, 1998; Cheng & Ganguly, 1994) to which the
interested reader should refer for more extensive discussions. There have been two
common approaches in the development of multicomponent excess Gibbs energy
models. One is to begin with an appropriate polynomial function to represent
Gmxs of
the multicomponent solution, and then truncate it after a certain number of terms, which
leads to special forms for
Gmxs for the bounding binaries. The other approach is to
combine the binary excess free energies according to certain empirical schemes. These
approaches, which have been called respectively the “power series multicomponent
models” and “projected multicomponent models” by Cheng & Ganguly (1994), are
discussed below.
I will refer to a multicomponent solution by the nature of its most asymmetric
binary. For example, a subregular multicomponent solution is the one for which the
most asymmetric binary has subregular behaviour. I will first discuss solution models
which deal with mixing within a single structural site, and then a method of combination
of mixing within the individual sites to express the multisite mixing properties.

Power series multicomponent models

One of the earliest and most successful multicomponent models is that due to Wohl
(1946, 1953), who introduced a power series expression for
Gmxs of a ternary solution.
Wohl’s ternary expression is as follows
)1 &

Gmxs   X i X j (WijG X j  W jiG X i )  X i X j X k '  (Wij  W ji )  Cijk $ . (66)
i
j '( 2 i
j $%

Subsequently, power series ternary and quaternary expressions were developed by

several workers in the geochemical literature, viz. Anderson & Lindsley (1981), Berman
& Brown (1984), Helffrich & Wood (1989) and Mukhopadhyay et al. (1993). However,
Cheng & Ganguly (1994) showed that all these expressions are either equivalent to
Wohl’s ternary expression or represent its extension to quaternary solution. They also
developed a power series quaternary expression following Wohl’s ternary formulation,
60 J. Ganguly

and showed that upon truncating it after the third degree terms (i.e. those that contain
three-suffix constants reflecting three-component interactions), the excess Gibbs energy
of mixing can be expressed as

Gmxs   X i X j (WijG X ji  W jiG X ij )   X i X j X k Cijk , (67)

i
j i
j,
k

where Cijk is a ternary interaction term, as defined below, the WG-s are the binary
interaction parameters, and Xji and Xij are the projected mole fractions of the
components j and i, respectively, in the binary join i–j. These binary mole fractions are
obtained by the normal projection of the multicomponent composition onto that join.
Analytically, Xij is given by ½(1 + Xi – Xj). It is interesting to note that the first right
hand term in the above expression represents a summation of
Gmxs of the subregular
bounding binaries at compositions that are at the shortest distance from the
multicomponent composition. Furthermore, a quaternary or higher order solution does
not involve a quaternary or higher order term when the binaries have sub-regular
behaviour (this conclusion was also independently reached by Jordan et al., 1950;
Helffrich & Wood, 1989; and Mukhopadhyay et al., 1993). Using Equations 67 and 15,
one can obtain the expression for the activity coefficient of a component in a
multicomponent subregular solution (see Cheng & Ganguly, 1994).
In terms of the coefficients of the power series expression, the binary subregular
parameters are given by (Wohl, 1946; Cheng & Ganguly, 1994)
Wij = aij + aiij and Wji = aij + aijj, (68)
whereas the ternary term represents
Cijk = ½(2aijk – aiij – aijj – aiik – aikk – ajjk – ajkk). (69)
Here the ‘a’ terms are related to the interactions of the subscripted species. Now, if
aiij = aijj, then Wij = Wji, and consequently the first right hand term of Equation 67
reduces to a combination of binary excess free energies with regular solution behaviour.
However, under this condition, that is the equivalence of three body interactions
between two species, Cijk = [aijk – (aiij + aiik + ajjk)], which is not necessarily equal to zero.
Thus, in general,
Gmxs of a ternary or higher order regular solution may not equal that
obtained by the summation of the subsidiary binaries. However, it is likely that when
the binaries conform to a regular solution, the Cijk terms are small. This is an intuitive
suggestion in the sense that a ternary regular solution should involve a smaller number
of higher order terms than a ternary subregular solution.
[It should be noted that the so-called ternary term of Berman & Brown (1984) is
not the ternary interaction term Cijk defined above, which we will call Cijk(Wohl), but
represents the first term within the square bracket of Equation 66, that is one-half of the
sum of the binary interaction terms. The Cijk term of Mukhopadhyay et al. (1993) is also
not the Cijk(Wohl), but stands for the collection of terms within the square brackets of
Equation 66. Also note that their Wijk = – Cijk(Wohl).]
 Thermodynamic modelling of solid solutions 61

Redlich & Kister (1948) utilised Guggenheim’s polynomial for binary solution
(Eqn. 31) to express
Gmxs of multicomponent solution. Their expression, which is
commonly referred to as the Redlich–Kister model, involves summation of the binary

Gmxs plus multicomponent correction terms, i.e.

Gmxs = XiXj(
Gmxs)ij + multicomponent correction terms, (70)
xs
where (
G ) is calculated at Xi and Xj from Equation 40 (note that these are total
m ij
atomic or mole fractions of the components in the multicomponent system). The
Redlich–Kister model has enjoyed popularity in the metallurgical literature (e.g. Hillert,
1998). However, it should be noted that it is equivalent to the Wohl model as long as the
bounding binaries follow regular or subregular behaviour (Cheng & Ganguly, 1994), i.e.
when the constant terms higher than A1 are zero in the polynomial expression of the
binary excess free energies (Eqn. 31).

Projected multicomponent models

Several schemes have been proposed for combining the binaries to predict the
multicomponent behaviour with or without involving multicomponent interaction
terms. These methods, which are due to Kohler (1960), Colinet (1967), Muggianu et al.
(1975) and Bonnier & Caboz (1965, quoted in Hillert, 1998), are illustrated in Figure 4.
Their expressions for the
Gmxs in a ternary solution have been summarised by Hillert
(1998), and are, thus, not repeated here. The method of shortest distance suggested by
Muggianu et al. (1975), which is usually referred to as the Muggianu method, was also
suggested independently by Jacob & Fitzner (1977). We will, thus, refer to this model
as the Muggianu–Jacob model. The expression for
Gmxs given by Bonnier & Caboz

1 1

3 2 3 2
(a) Kohler (b) Muggianu–Jacob
1 1

3 2 3 2
(c) Colinet (d) Toop
xs
Fig. 4. Schematic illustration of the “Projected Multicomponent Models” for a ternary system. The
Gm of a
xs
ternary solution is calculated by combining the Gm of the terminal binaries at the projected compositions.
62 J. Ganguly

(1965) was modified later by Toop (1965) in order that the ternary
Gmxs appears as a
summation of the
Gmxs in the binaries when the latter have regular solution behaviour,
i.e.
Gmxs(ternary) = XijXji
(Gmxs)ij where Xij and Xji are the shortest distance binary
compositions from the ternary compositional point. The modified Bonnier–Caboz
formulation is often referred to as Toop’s method in the literature. The primary
motivations behind the different projected multicomponent formulations is the
prediction of the multicomponent behaviour from only the binary properties, that is to
somehow ‘absorb’ the effects of multicomponent interactions within the scheme of
combination of the binaries.
Toop’s method is an asymmetric formulation in that it treats one component
(component 1 in Fig. 4d) differently from the other two. Thus, this method ought to be
applied only to ternary systems where one component has a distinctly different property
from the other two. For example, Pelton & Blander (1986) used a modified QC
formulation for the silicate slag system SiO2–CaO–FeO in which the method of
combination of binaries is analogous to that of Toop’s method. They chose SiO2 as the
special component 1, since it is an acidic component while the other two are basic
components. The predicted ternary properties from combination of the binary data were
found to be in good agreement with the experimental data. For solid solutions, one may
also be able to identify a component which behaves quite differently from the others.
For example, in aluminosilicate garnet, (Fe,Mg,Ca)3Al2Si3O12, Ca is the most nonideally
mixing component, while Fe and Mg mix nearly ideally (e.g. Ganguly et al., 1996).
Thus, Ca may be treated as the unique component in the asymmetric formulation.
The expression of
Gmxs obtained for a multicomponent subregular solution
(Eqn. 67) using the power series approach of Wohl (1946) involves a combination of
binary excess Gibbs energies at compositions that are at the shortest distance from the
multicomponent composition. This is exactly the method of combination of the binaries
suggested in the Muggianu–Jacob projected multicomponent model (Fig. 4b). Thus,
there seems to be an independent theoretical justification in the scheme of the
combination of binaries in this model. It was found (Jacob & Fitzner, 1977; Jacob, pers.
comm.) that for metallic systems the shortest distance method predicts the ternary
Gmxs
somewhat better than the other methods. However, the quality of agreement between the
predicted and measured ternary values becomes worse with increasing non-ideality of the
binaries, especially when a binary
Gmxs exceeds 15 kJ/mole, implying increasing
importance of the higher order terms.

Estimation of higher order interaction terms

In principle, it is impossible to determine the multicomponent interaction terms from
only the binary data. The higher order interactions specific to a given model can only
be determined from comparison of the predicted multicomponent behaviour from the
binary data with the multicomponent properties determined experimentally. It is,
however, doubtful if experimental data are going to be sufficient for such purpose in the
foreseeable future, at least for systems of geological interests. A viable alternative
would be to determine the binary and multicomponent enthalpic properties from the
 Thermodynamic modelling of solid solutions 63

crystallographic data and pair potential energies of the ions participating in the solid
solutions (e.g. Ottonello, 1992). Since there are a large body of crystallographic data for
a variety of rock-forming mineral solid solutions in both binary and multicomponent
systems, it may be possible to use this approach to at least approximately evaluate the
relative magnitudes of the higher order terms in the different multicomponent
formulations. One may, thus, determine the effectiveness of the different approaches in
predicting the multicomponent behaviour from only binary data.
From theoretical analysis, Cheng & Ganguly (1994) developed a method of
approximation of the Cijk term (Eqn. 67) in the special case that one of the binaries, j–k,
behaves nearly ideally. Ilmenite series, (Mg,Fe,Mn)TiO3, and garnet series,
(Ca,Fe,Mg)3Al2Si3O12, are examples of this type of solid solution. In both cases, the last
two components mix nearly ideally (Shibue, 1999; Ganguly et al., 1996). Indeed, Fe2+
and Mg mix with small deviation from ideality in all ferromagnesian silicates for which
the thermodynamic mixing properties are known (Ganguly & Saxena, 1987). Thus, this
approximation scheme should be applicable to ternary joins of rock-forming minerals
involving Fe2+–Mg as one of the subsidiary binaries. The method is as follows.
) Xj Xk &
Cijk 6   ij ik '(Wij  W ji )  (Wik  Wki ) $ , (71)
(' X j  Xk X j  X k %$

where ij = 0 when i  j, and ij = 1 when i

j. For garnet, this method predicts CCaMgFe
to be of the order of a kJ per cation-mole of the ternary garnet solid solution. This is
consistent with the conclusion of Berman & Koziol (1991), based on the analysis of
natural and experimental phase equilibrium data involving garnet solid solutions.

Multicomponent property without multicomponent terms

The Wilson, NRTL, QC and UNIQUAC formulations have been extended to
multicomponent solutions without requiring any multicomponent term. These extensions,
which can be found in Prausnitz et al. (1986) and Pelton & Blander (1986), have special
appeal in that the calculation of the multicomponent properties requires knowledge of only
the binary properties. However, it remains to be tested how well these formulations predict
the behaviour of multicomponent silicate and oxide solid solutions.
Using only the binary mixing parameters, Shibue (1999) applied both the ternary
Wilson and Margules (which is the same as Wohl’s) formulations to the (Fe,Mn,Mg)TiO3
solid solution to evaluate which one predicts the behaviour of the ternary solution better.
He utilised the data on cation exchange between aqueous chloride solution and the three
binary solid solutions in this system to derive binary cation mixing properties in the solid
solution, assuming the chloride solution to be ideal. Of the three binaries, the Fe2+–Mn2+
join was found to be nearly ideal. From the retrieved binary mixing data, Shibue (1999)
calculated the ternary mixing properties, and from them the compositions of aqueous
chloride solutions in equilibrium with ternary solid solutions of various compositions.
The predicted compositions of the chloride solution were then compared with the
experimental data of Kubo et al. (1992) on the equilibrium compositions of coexisting
solid and chloride solution. Shibue (1999) found that the sum of the distances (di)
64 J. Ganguly

between the predicted and observed ternary chloride solution using Wilson’s
formulation for the solid solution is 0.0094, compared to that of 0.0127 using Wohl’s
formulation with only the binary terms. He, thus, concluded that the prediction of
ternary behaviour from the Wilson formulation is better than that from Wohl’s
formulation. However, incorporation of the ternary parameter in Wohl’s formulation
according to Equation 71, which can be calculated from the binary data, leads to very
similar quality of prediction of the ternary property as that obtained from the Wilson
formulation (Shibue, pers. comm.).
One problem with the above analysis of the relative merits of the two models is
that the values of
G0 of the binary solid–fluid exchange reactions that are retrieved
from the experimental data depend on the adopted binary mixing models for the solids.
However, in principle,
G0 is a unique quantity. In addition, the retrieved values of
G0
depend on whether the sum of the values for the three binaries have been constrained to
be zero, as there are only two independent binaries, or whether these are derived solely
from the best fit to the binary data. The resultant variation of
G0 leads to changes of
the correlated binary parameters and, hence, in the quality of prediction of ternary
properties. For example, the subregular parameters retrieved from the constraint that

G0(binaries) = 0 are somewhat different from those that are derived simply from best
fits to the binary data, and used in Shibue (1999). The first set leads to a poorer
prediction of the ternary properties (di = 0.0154) compared to those from the second
set (di = 0.0099) (Shibue, pers. comm.).
Barron (1976) compared the predictive success of the ternary properties in the
system Ab–Kfs–An at 900 °C according to Wohl’s and Kohler’s model using only the
binary terms. In this ternary system, there is a large solvus in the An–Kfs binary, and
continuous solid solution in the two other joins at 900 °C. Both models predict the
experimentally determined ternary solvus at 900 °C quite closely. Ottonello (1992) also
found that Wohl’s and Kohler’s models have comparable predictive abilities of ternary
properties from the binary data.

Solid solutions with multi-site mixing

Hillert (1998) suggested the following expression to represent the
Gxs m of a two-site
binary reciprocal solution, I(A,B)bII(C,D)c, for which
Gm is given by Equation 16.

Gmxs = ABCIAB:C + ABDIAB:D + CDAICD:A + CDBICD:B, (72)
where A is the atomic fraction of the species A in the site I, IAB:C is the interaction
parameter between A and B in the site I when site II is completely filled by C, ICD:A is the
interaction parameter between C and D in the site II when site I is completely is filled by
A, and so on. The above expression allows different behaviour of
Gmxs within a specific
sublattice depending on the nature of the species occupying the other sublattice, or in other
words, different
Gmxs on opposite sides of the compositional square illustrated in Figure 1.
For example, when C = 1,
Gxs = ABIAB:C, but when D = 1,
Gxs = ABIAB:D. Each site
parameter, IAB:C and so on, may be expressed according to the Guggenheim or the so-called
Redlich–Kister form, that is, by the expression within the square bracket in Equation 31,
truncating it after the appropriate number of terms, as demanded by the data. Extensions of
 Thermodynamic modelling of solid solutions 65

this approach to multiple sublattices and multiple components have been discussed by
Hillert (1998) (and is incorporated in the commercial computer program Thermo-Calc:
http://www/thermocalc.se, which has been used extensively for the calculation of phase
diagrams).

Concluding remarks
I have summarised above the physical ideas and the basic theoretical structure for a
variety of solution models that have been used to treat the thermodynamic properties of
mineral solid solutions. These models are also applicable to melts. It may be possible to
fit a limited body of data by more than one-solution models equally well, but
Hmix and

Smix predicted by the different models could differ quite significantly. It is, therefore,
important to examine the theoretical basis of the solution models in such cases and to
ensure, as much as possible, compatibility of the adopted solution model with the known
microscopic properties of the solid solution (e.g. short range ordering; compatibility of
the retrieved fitting parameters with their expected values from theory such as discussed
in the section on “comparison of one-site binary mixing models”).
From the results of the comparative studies, as discussed above, Guggenheim’s
polynomial (Eqn. 31) or the so-called Redlich–Kister formulation seems to offer a simple
and flexible model for binary solid solutions, although in some specific cases another
model, especially QC model when there is short range order, may work better. Use of the
Guggenheim polynomial for the binaries affords an additional advantage in the treatment
of reciprocal solid solutions in terms of the form suggested by Hillert (1998), because the

Gxsm in this expression reduces to the form ij[..] for the terminal binaries. Binary
solutions obeying DQF can also be incorporated in this scheme by defining a solution
between a real component and a hypothetical component (Eqn. 43). Finally, if one is to use
only the binary terms to predict the multicomponent properties, then the “shortest distance
method” of combining the binaries (Fig. 4c) is probably the overall best method because of
the theoretical justification, as discussed above. The asymmetric Toop method could be
advantageous where there is a component with a distinctly different property.
There is currently a great paucity of experimental data on the multicomponent
mixing properties of rock-forming minerals which have been routinely used for thermo-
barometric and other phase equilibrium calculations. These types of data are needed to
understand the nature of the higher order terms and the method of combination of
binaries to predict the multicomponent properties from the binary data. In many
important systems, adequate binary data are also lacking. The lack of sufficient well-
constrained experimental data has led to a proliferation of empirically adjusted
formulations of a given geothermometer or a geobarometer. An example is the
garnet–biotite Fe2+–Mg exchange thermometer, for which there are approximately 20
different empirical or semi-formulations to take into account the nonideal mixing
effects of additional components, especially in biotite.
With the advancement of diffusion kinetic modelling of compositional zoning of
rock-forming minerals, such as garnet and pyroxene, to retrieve the time scales of
metamorphic processes (e.g. Spear, 1995; Ganguly et al., 2000) there is now an
66 J. Ganguly

increased need for accurate thermometric formulations. This is because of the expo-
nential dependence of the diffusion process on temperature (D  exp(–Q/RT)), so that
a small error in temperature estimate translates into a large error in the retrieved time
scales. The problem is especially acute in minerals such as garnet and pyroxene, which
have a relatively large activation energy, Q, of diffusion (Ganguly & Tazzoli, 1994;
Ganguly et al., 1998). In addition, resolution of many large-scale tectono-metamorphic
problems relies on the precision of thermo-barometers from which a spatial distribution
map of P–T conditions of metamorphism is produced (e.g. Neogi et al., 1998). The
extraction of thermodynamic properties of silicate melts, which have wide ranging
applications, from solid-melt phase equilibrium data requires reliable data on the
thermodynamic properties of the solid solutions. Mutual compatibility of the thermo-
dynamic properties of the melt and solid phases is necessary, but not sufficient, for the
successful prediction of melt properties beyond the range of experimental phase
equilibrium data from which these were extracted.
In view of the wide ranging applications of thermodynamic properties of solid
solutions to geological problems, it is hoped that there will be a major effort to provide
experimental data that would tightly constrain the properties of multicomponent rock-
forming minerals and lead to improvements in the theoretical formulations relating the
binary and multicomponent properties. However, because of their compositional
complexity, it may be a long time before we have adequate experimental data to
sufficiently constrain the thermodynamic properties of some important mineral solid
solutions. It is, therefore, necessary to improve our understanding of the microscopic
basis of thermodynamic nonideality of solid solutions so we can make appropriate
approximations where adequate experimental data are not available. A microscopic
understanding is also important to evaluate the quality of, and discriminate among,
conflicting experimental results.

Acknowledgements
I am indebted to Prof. Mats Hillert for a thorough and constructive review of an earlier
version of the manuscript in a very short notice. Thanks are due to Prof. Charles Geiger
and Prof. Sumit Chakraborty for their suggestions to improve the clarity of presentation
and to Dr. Victor Vinograd and Prof. Yasuhiro Shibue for helpful discussions. Prof.
Geiger’s careful editorial work is also gratefully appreciated. I, of course, bear the sole
responsibility of any error or lack of clarity that still exists in this paper. This work was
partly supported by a NASA grant No. NAG 5-7364.

References
Abrams, D.S. & Prausnitz, J.M. (1975): Statistical thermodynamics of liquid mixtures: A new expression for the
excess Gibbs energy of partly or completely miscible systems. Am. Inst. Chem. Eng. J., 21:116–128.
Anderson, D.J. & Lindsley, D.H. (1981): A valid Margules formulation for an asymmetric ternary solution:
Revision of olivine-ilmenite thermometer, with applications. Geochim. Cosmochim. Acta, 45:847–853.
Andersson, J.-O., Fernández Gillermet, A., Hillert, M., Jansson, B. & Sundman, B. (1986): A compound energy
model of ordering in a phase with sites of different coordination numbers. Acta Metall., 34:437–445.
Barron, L.M. (1976): A comparison of two models of ternary excess free energy. Contrib. Mineral. Petrol.,
57:71–81.
 Thermodynamic modelling of solid solutions 67

Berman, R.G. & Brown, T.H. (1984): A thermodynamic model for multicomponent melts, with applications to
the system CaO-Al2O3-SiO2. Geochim. Cosmochim. Acta, 48:661–678.
Berman, R.G. & Koziol, A.M. (1991): Ternary excess properties of grossular-pyrope-almandine garnet and their
influence in geothermometry. Am. Mineral., 76:1223–1231.
Blander, M. (1964): Thermodynamic properties of molten salt solutions. In Blander, M. (ed.): Molten salt
chemistry. New York (N.Y.): Wiley, 127–237.
Bosenick, A. & Geiger, C.A. (1997): Powder X-ray diffraction study of synthetic pyrope-grossular garnets
between 20 and 295 K: a comparison of thermal expansion and heat capacity and volumes of mixing. J.
Geophys. Res., 102:22,649–22,657.
Burton, B. & Kikuchi, R. (1984a): Thermodynamic analysis of the system CaCO3-MgCO3 in the tetrahedron
approximation of the cluster variation method. Am. Mineral., 69:165–175.
Burton, B. & Kikuchi, R. (1984b): The antiferromagnetic-paramagnetic transition in 5-Fe2O3 in the single prism
approximation of the cluster variation method. Phys. Chem. Miner., 11:125–131.
Cheng, W. & Ganguly, J. (1994): Some aspects of multicomponent excess free energy models with subregular
binaries. Geochim. Cosmochim. Acta, 58:3763–3767.
Colinet, C. (1967): Relation between the geometry of ternary phase diagrams and the thermodynamic properties
of liquid solutions. Diplôme d’études supérieures, Univ. Grenoble, France. (PhD Thesis, in French)
Darken, L.S. (1950): Application of the Gibbs-Duhem equation to ternary and multicomponent systems. J. Am.
Chem. Soc., 72:2909–2914.
Darken, L.S. (1967): Thermodynamics of binary metallic solutions. Met. Soc. AIME Trans., 239:80–89.
Darken, L.S. & Gurry, R.W. (1953): Physical chemistry of metals. New York (N.Y.): McGraw-Hill.
Evans, B.W. & Frost, B.R. (1975): Chrome-spinel in progressive metamorphism – a preliminary analysis.
Geochim. Cosmochim. Acta, 39:959–972.
Fei, Y., Saxena, S.K. & Eriksson, G. (1986): Some binary and ternary silicate solid solution models. Contrib.
Mineral. Petrol., 94:221–229.
Flood, H., Førland, T. & Grjotheim, K. (1954): Über den Zusammenhang zwischen Konzentration und Aktivitäten
in geschmolzenen Salzmischungen. Z. Anorg. Allg. Chem., 276:290–315.
Flory, P.J. (1941): Thermodynamics of high polymer solutions. J. Chem. Phys., 9:660–661.
Flory, P.J. (1944): Thermodynamics of heterogeneous polymers and their solutions. J. Chem. Phys., 12:425–438.
Førland, T. (1964): Thermodynamic properties of fused salt systems. In Sundheim, B.R. (ed.): Fused salts. New
York (N.Y.): McGraw-Hill, 63–164.
Fowler, R.H. & Rushbrooke, G.S. (1937): An attempt to extend the statistical theory of perfect solutions. Trans.
Faraday Soc., 33:1272–1294.
Ganguly, J. (1982): Mg-Fe order-disorder in ferromagnesian silicates. II. Thermodynamics, kinetics, and
geological applications. In Saxena, S.K. (ed.): Advances in physical geochemistry, Vol. 2. Berlin-Heidelberg-
New York (N.Y.): Springer-Verlag, 58–99.
Ganguly, J. & Saxena, S.K. (1987): Mixtures and mineral reactions. Berlin-Heidelberg-New York (N.Y.):
Springer-Verlag.
2+
Ganguly, J. & Tazzoli, V. (1994): Fe -Mg interdiffusion in orthopyroxene: Retrieval from the data on
intracrystalline exchange reaction. Am. Mineral., 79:930–937.
Ganguly, J., Cheng, W. & O’Neill, H.St.C. (1993): Syntheses, volume, and structural changes of garnets in the
pyrope-grossular join: Implications for stability and mixing properties. Am. Mineral., 78:583–593.
Ganguly, J., Cheng, W. & Tirone, M. (1996): Thermodynamics of aluminosilicate garnet solid solution: new
experimental data, an optimized model, and thermometric applications. Contrib. Mineral. Petrol.,
126:137–151.
Ganguly, J., Cheng, W. & Chakraborty, S. (1998): Cation diffusion in aluminosilicate garnets: experimental
determination in pyrope-almandine diffusion couples. Contrib. Mineral. Petrol., 131:171–180.
Ganguly, J., Dasgupta, S., Cheng, W. & Neogi, S. (2000): Exhumation history of a section of the Sikkim
Himalayas, India: records in the metamorphic mineral equilibria and compositional zoning of garnet. Earth
Planet. Sci. Lett., 183:471–486.
Ghiorso, M. (1990): Application of the Darken equation to mineral solid solutions with variable degrees of order-
disorder. Am. Mineral., 75:539–543.
68 J. Ganguly

Ghose, S. (1982): Mg-Fe order-disorder in ferromagnesian silicates. I. Crystal Chemistry. In Saxena, S.K. (ed.):
Advances in physical geochemistry, Vol. 2. Berlin-Heidelberg-New York (N.Y.): Springer-Verlag, 4–57.
Green, E.J. (1970a): Predictive thermodynamic models for mineral systems. I. Quasi-chemical analysis of the
halite-sylvite subsolidus. Am. Mineral., 55:1692–1713.
Green, E.J. (1970b): Perils of thermodynamic modelling. Geochim. Cosmochim. Acta, 34:1029–1033.
Guggenheim, E.A. (1937): Theoretical basis of Raoult’s law. Trans. Faraday Soc., 33:151–159.
Guggenheim, E.A. (1952): Mixtures. Oxford: Clarendon Press.
Guggenheim, E.A. (1967): Thermodynamics. Amsterdam: North Holland Publ. Co.
Helffrich, G. & Wood, B.J. (1989): Subregular models for multicomponent solutions. Am. Mineral.,
74:1016–1022.
Hildebrand, J.H. (1929): Solubility, XII, Regular solutions. J. Am. Chem. Soc., 51:66–80
Hillert, M. (1980): Empirical methods of predicting and representing thermodynamic properties of ternary
solution phases. Calphad, 4:1–12.
Hillert, M. (1998): Phase equilibria, phase diagrams and phase transformations: Their thermodynamic basis.
Cambridge (U.K.): Cambridge Univ. Press.
Hillert, M. & Staffansson, L.-I. (1970): Regular solution model for stoichiometric phases and ionic melts. Acta
Chem. Scand., 24:3618–3626.
Huggins, M.L. (1941): Solutions of long chain compounds. J. Phys. Chem., 9:440.
Jacob, K.T. & Fitzner, K. (1977): The estimation of the thermodynamic properties of ternary alloys from binary
data using the shortest distance composition path. Thermochim. Acta, 18:197–206.
Jiran, E. & Jacob, K.T. (1983): Computation of the thermodynamic properties of quaternary and higher order
systems from binary data using shortest distance composition paths. Calphad, 7:41–50.
Jordan, D., Gerster, J.A., Colburn, A.P., & Wohl, K. (1950): Vapor-liquid equilibrium of C4 hydrocarbon-furfural-
water mixtures. Chem. Eng. Prog., 46:601–613.
Kikuchi, R. (1951): A theory of cooperative phenomena. Phys. Rev., 81:988–1003.
Kohler, F. (1960): Zur Berechnung der thermodynamischen Daten eines ternären Systems aus dem zugehöringen
binären System. Monatsh. Chem., 91:738–740.
Kroll, H., Schiemann, I. & von Cölln, G. (1986): Feldspar solid solutions. Am. Mineral., 71:1–16.
Kubo, T., Uchida, E., Furukawa, Y. & Imai, N. (1992): Experimental study on ion exchange equilibria between
solid solution (Fe2+,Mn2+,Mg2+)TiO3 and aqueous (Fe2+,Mn2+,Mg2+)Cl2 solution. Kobutsugaku Zasshi (J.
Mineral. Soc. Jpn.), 21:59–67 (in Japanese with English abstract).
Liermann, H-P. & Ganguly, J. (1999): Thermodynamics and kinetics of Fe2+-Mg exchange between spinel and
orthopyroxene: Experimental determination an application to cooling rates (abstract). Meteorit. Planet. Sci.,
34 (suppl.): A75.
Meyer, K.H. & van der Wyk, A. (1944): Properties of polymer solutions. XVIII. Thermodynamic analysis of
binary systems with a chain-shaped component. Helv. Chim. Acta, 27:845–858 (in French).
Muggianu, Y.M., Gambino, M. & Bros, J.P. (1975): Enthalpies of formation of liquid alloy bismuth-gallium-tin
at 723 K. Choice of an analytical representation of integral and partial excess functions of mixing. J. Chim.
Phys., 72:83–88.
Mukhopadhyay, B., Basu, S. & Holdaway, M. (1993): A discussion of Margules-type formulations for
multicomponent solutions with a generalised approach. Geochim. Cosmochim. Acta, 57:277–283.
Neogi, S., Dasgupta, S. & Fukuoka, M. (1998): High P-T metamorphism, dehydration melting, and generation of
migmatites and granites in the higher Himalayan crystalline complex, Sikkim, India. J. Petrol., 39:61–99.
Orye, R.V. & Prausnitz, J.M. (1965): Thermodynamic properties of binary solutions containing hydrocarbons and
polar organic solvents. Trans. Faraday Soc., 61:1338–1346.
Ottonello, G. (1992): Interactions and mixing properties in the (C2/c) clinopyroxene quadrilateral. Contrib.
Mineral. Petrol., 111:53–60.
Pelton, A.D. & Blander, M. (1986): Thermodynamic analysis of ordered liquid solutions by a modified
quasichemical approach – Applications to silicate slags. Metall. Trans., B, Process Metall., 17:805–815.
Powell, R. (1983): Thermodynamics of complex phases. In Saxena, S.K. (ed.): Advances in physical geochem-
istry, Vol. 2. Berlin-Heidelberg-New York (N.Y.): Springer-Verlag, 241–266.
Powell, R. (1987): Darken’s quadratic formulation and the thermodynamics of minerals. Am. Mineral., 72:1–11.
 Thermodynamic modelling of solid solutions 69

Prausnitz, J.M., Lichtenthaler, R.N. & de Azevedo, E.G. (1986): Molecular thermodynamics of fluid-phase
equilibria. 2nd ed. Englewood Cliffs (N.J.): Prentice-Hall.
Redlich, O. & Kister, A.T. (1948): Thermodynamics of non-electrolyte solutions, x-y-t relations in a binary
system. Ind. Eng. Chem., 40:341–345.
Renon, H. & Prausnitz, J.M. (1968): Local compositions in thermodynamic excess functions for liquid mixtures.
J. Am. Inst. Chem. Eng., 14:135–144.
Sack, R.A. & Loucks, R.R. (1985): Thermodynamic properties of tetrahedrite-tennantites: Constraints on the
independence of the Ag Ca, Fe Zn, Cu Fe, Ag Sb exchange reactions. Am. Mineral., 71:257–269.
Shibue, Y. (1999): Calculations of fluid-ternary solid solutions equilibria: An application of the Wilson equation
to fluid-(Fe,Mn,Mg)TiO3 equilibria at 600 °C and 1 kbar. Am. Mineral., 84:1375–1384.
Spear, F.S. (1995): Metamorphic phase equilibria and pressure-temperature-time paths. Washington, D.C.:
Mineral. Soc. Am.
Stimpfl, M., Ganguly, J. & Molin, G. (1999): Fe2+-Mg order-disorder in orthopyroxene: equilibrium fractionation
between the octahedral sites and thermodynamic analysis. Contrib. Mineral. Petrol., 136:297–309.
Sundman, B. & Ågren, J. (1981): A regular solution model for phases with several components and sublattices,
suitable for computer applications. J. Phys. Chem. Solids, 42:297–301.
Thompson, J.B., Jr. (1967): Thermodynamic properties of simple solutions. In Abelson, P.H. (ed.): Researches in
geochemistry, Vol. 2. New York (N.Y.): Wiley, 340–361.
Toop, G.W. (1965): Predicting ternary activities using binary data. Trans. Am. Inst. Min. Eng., 233: 850–855.
Truskinovskiy, L.M., Sigalovskaya, Yu.I., Senderov, E.E. & Urusov, V.S. (1987): A general quasi-chemical model
of mineral disordering. Geokhimiya, 11:1511–1526.
Vinograd, V.L. (2001): Configurational entropy of binary silicate solid solutions. In Geiger, C.A. (ed.): Solid
solutions in silicate and oxide systems /EMU Notes Mineral., 3/. Budapest: Eötvös Univ. Press, 303–346.
Wilson, G.M. (1964): Vapor-liquid equilibrium. XI. A new expression for the excess free energy of mixing. J. Am.
Chem. Soc., 86:127–130.
Wohl, K. (1946): Thermodynamic evaluation of binary and ternary liquid systems. Trans. Am. Inst. Chem. Eng.,
42:215–249.
Wohl, K. (1953): Thermodynamic evaluation of binary and ternary liquid systems. Chem. Eng. Prog., 49:218–221.
Wood, B.J. & Nicholls, J. (1978): The thermodynamic properties of reciprocal solid solutions. Contrib. Mineral
Petrol., 66:389–400.