Computational Materials Science 50 (2011) 3153–3161

Contents lists available at ScienceDirect

Computational Materials Science

journal homepage: www.elsevier.com/locate/commatsci

Tracking multi-phase boundaries using an integration-based approach

Q. Du a,⇑, M.A. Wells b
a
University of British Columbia, Department of Materials Engineering, Vancouver, BC, Canada
b
University of Waterloo, Department of Mechanical and Mechatronics Engineering, Waterloo, Ontario, Canada

a r t i c l e i n f o a b s t r a c t

Article history: An integration-based method to track multi-phase boundaries for multi-component alloys systems is pro-
Received 5 May 2011 posed. The foundation of the method is considering the phase boundary tracking as an initial value prob-
Received in revised form 25 May 2011 lem. It starts with the utilization of Gibbs–Duhem relation to find the slope of a phase boundary (the partial
Accepted 29 May 2011
derivatives of the solvus temperature with respect to the equilibrium phase composition) and the partial deriv-
atives of partition coefficients with respect to the equilibrium phase composition. Then, with these derivatives
and a pair of known initial points on a phase boundary, the rest of the phase boundary can be calculated
Keywords:
using a numerical integration technique such as the Runge–Kutta or Predictor–Corrector method in a step-
Gibbs–Duhem relation
Phase diagram calculation
wise manner. The integration method has been applied to track various types of phase boundaries in bin-
Integration technique ary, ternary and quaternary systems. The calculated ones are in good agreement with the ones obtained by
the commercial CALPHAD software Thermo-Calc. It provides a different computational method for the
tracking of multi-phase boundaries and might be useful in the calculation of phase diagram when com-
bined with the overlapping of Zero Phase Fraction (ZPF) lines technique and global minimization
technique.
Ó 2011 Elsevier B.V. All rights reserved.

1. Introduction terms which help introduce auxiliary (side) conditions to deter-

mine a specific phase equilibrium [6,7]. The necessity of introduc-
Phase diagrams, considered by the metallurgist as ‘‘road map’’ ing auxiliary (side) conditions is due to the fact that in
in the design of alloys and selection of processing parameters, multi-component systems at given temperature and pressure, the
could be computed by CALculation PHAse Diagram (CALPHAD) equilibrium state between two phases are not unambiguously
software such as FactSage, Pandat and Thermo-Calc. CALPHAD determined as indicated by Gibbs phase rule (mathematically
software consists of two essential components, which are thermo- there might be more unknowns than equations, i.e., the system
dynamic models and an Gibbs energy minimization method. In may be underdetermined). Hence other constraints need to be
general, thermodynamic models are constructed on the experi- imposed to reduce the degree of freedom to zero. One way of doing
mental measurements such as Differential Thermal Analysis this involves using the average compositions of the whole system
(DTA) and heat flux Differential Scanning Calorimetry (DSC). There and assuming one phase fraction is zero (Zero Phase Fraction con-
have been great efforts to develop open thermodynamic databases cept) for the system with two phases in equilibrium. The minimi-
such as SGTE database that could be used for phase diagram calcu- zation finally transforms into the solution of a set of non-linear
lation [1]. equations, and a common feature of CALPHAD software is that an
As discussed in [2–5], it was common during the minimization a iteration technique is employed to solve them numerically. Cur-
local minimization rather than global minimization was found. rently, computer codes that calculate phase diagrams (CALPHAD)
Therefore it is important to utilize a global optimization method. such as use numerical methods based on the two-step method of
However a global optimization method is CPU-expensive and in Hillert [6] or the one-step method of Lukas et al. [7] to minimize
commercial software packages an appropriate mixing of local and the Gibbs energy [8].
global minimization has to be maintained in order to find the cor- Alternatively, there is another phase diagram calculation strat-
rect equilibria within reasonable calculation time. This paper egy, namely the solubility product method, in which there is no
restricts itself to the category of the local minimization technique. need to introduce auxiliary (side) conditions. This implies the aux-
The minimization of the total Gibbs energy of the system is in iliary (side) conditions are not essential conditions in calculating a
general accompanied with some additional Lagrange multiplier phase diagram. The basic idea behind the solubility method is that
those mathematically underdetermined non-linear equations are
⇑ Corresponding author. Tel.: +1 604 822 7886; fax: +1 604 822 3619. considered as the implicit equations defining the phase diagram.
E-mail address: [email protected] (Q. Du).
Unfortunately the use of the solubility product based method is lim-

0927-0256/$ - see front matter Ó 2011 Elsevier B.V. All rights reserved.
doi:10.1016/j.commatsci.2011.05.044
3154 Q. Du, M.A. Wells / Computational Materials Science 50 (2011) 3153–3161

ited to the systems where the thermodynamic functions of the

molar Gibbs energies are relatively simple such that the tempera-
ture–composition relationship (solubility line) and tie-line equa-
tions can be explicitly derived from the equilibrium conditions by
algebraic operations. For complex thermodynamic functions of a
multi-component alloy system, the algebraic operations obviously
cannot produce those explicit temperature–composition relations
and the solubility product method cannot be used. But the idea that
the phase diagram is defined implicitly by those underdetermined
non-linear equations motivates us to find a technique that is able
to obtain those explicit relations from implicit functions. The tech-
nique is numerical integration. Indeed, it had been used for binary
phase diagram calculations. It is well known that for binary alloy
systems the integration of Gibbs–Konovalov equation leads to the
phase boundary separating the two equilibrium phases, but this
integration method has never been applied to multi-component
alloys. It is the aim of this paper to exploit the integration technique
so that complex multi-component multi-phase boundary can be Fig. 1. Binary phase diagram for the illustration of the proposed integration
tracked. It is believed that the integration technique might be methodology.
computationally more efficient than the classical methods as no
iterations are required to obtain the phase diagram. The key point
to achieve this is to derive the integration equation for the partition calculated in a stepwise manner starting from a pair of initially
coefficients which govern the evolutions of tie-lines. known equilibrium points. However, due to our explicit identifica-
The phase boundary between two equilibrium phases (a and b) tion of the tracking as an initial value problem, this method differs
will be considered in this paper to illustrate the integration concept. from [9] in some important ways that lends itself to more practical
It starts with the deriving from the Gibbs–Duhem relationship of the use as to be demonstrated:
slope of the a phase boundary and the partial derivatives of the par-
tition coefficients with respect to the equilibrium a phase composi- 1. It is a general treatment of the two phase system that may con-
tion, X aj . Then the extension to multi-phase boundary is described. It tain a stoichiometric or partial stoichiometric phase.
is followed by the numerical integration methods for solving the 2. Mass balance relations (the auxiliary conditions) are not neces-
evolution equations governing the two important phase diagram sary to be parts of the equations, which reduces the number of
parameter, the solvus temperature and partition coefficient. Finally unknowns.
the proposed approach is applied to some binary and multi-compo- 3. The sophisticated integration technique such as adaptive
nent alloy systems to demonstrate its ability of tracking general Runge–Kutta integration technique or the Predictor–Corrector
multi-phase boundaries. All calculated phase boundaries are com- (PC) technique described in [10] could be employed rather than
pared to the ones obtained by Thermo-Calc in order to verify the a second-order Taylor series expansion used in [9].
methodology and to validate our implementation.
2.1. Mathematical calculation of the phase boundary slope and the
partial derivatives of partition coefficients
2. Development of an integration based numerical algorithm
Referring to Fig. 1, according to [11] the slope of the phase
The basic idea of the proposed approach is considering the boundary between two ab solution phases is given by the
boundary tracking as an initial value problem. The methodology Gibbs–Konovalov equation:
could be illustrated by taking a binary a–b phase diagram as an
2 a
example. The binary phase diagram can generally be described ðX aB  X bB ÞT @ Gm
dT @X Ba
2

using solvus temperature and partition coefficient: a ¼ ð2Þ

dX B DHb
T a ¼ T a ðX aB Þ
ð1Þ where DHb is the molar enthalpy (J) of first precipitation of an infin-
kB ¼ kB ðX aB Þ itesimal amount of the b phase from the a phase, Gam is the molar
Gibbs energy (J) of the a phase. It could be derived by examining
In which Ta is the solvus temperature for the a phase, kB is the
the total differential form of equal chemical potential equations as
Xa
partition coefficient defined as XBb . X aB and X bB are the atomic fractions demonstrated by Goodman et al. [11]. The same relation can also
B

of component B in phase a and b respectively as shown in Fig. 1. be derived based on the concept of implicit functions as pointed
Referring to Fig. 1, assume we initially know T a ðX aB Þ; a solvus out by Lukas et al. in their recent book [3]. Hillert [12] gave a more
temperature located on the solvus line of the phase diagram corre- general derivation based on the Gibbs–Duhem relation, which will
sponding to the equilibrium a phase composition X aB , and k B (hence be adopted here to derive both of the slope of the a phase boundary
X bB , the b phase composition in equilibrium with the a phase). Had and the partial derivatives of partition coefficients.
dT a dkB
;
dX aB dX aB
been derived, the shape of solubility limit line (solvus) and
the explicit relations between the phase compositions (tie-lines 2.1.1. Calculation of the phase boundary slope
equations) could be calculated using an analytical or numerical As discussed by Hillert [12], for an n-component two-phase
integration technique. dX dT a dkB
a; a could be derived from the thermody-
equilibrium system, the Gibbs–Duhem relation is applicable to
B dX B
namic models of this binary system using the Gibbs–Duhem rela- each phase, e.g., for the phases a and b by ignoring other effects
tions and the derivation is extendable to multi-component alloy such as magnetism one could write:
systems as described later in this section. X
n

The proposed technique is similar to the method reported by X ai dli þ Sam dT  V am dP ¼ 0 ð3aÞ
i¼1
Gaye and Lupis in [9] in the sense that the phase boundary is
 Q. Du, M.A. Wells / Computational Materials Science 50 (2011) 3153–3161 3155

X
n
From the definition of partition coefficient, one has:
X bi dli þ Sbm dT  V bm dP ¼ 0 ð3bÞ
i¼1
X ai
where li is the chemical potential of component i (J/mol). P is the X bi ¼ ð13Þ
pressure (Pa). Smm and V mm are the molar entropy (J) and molar vol- ki
ume (m3) of the m (m = a or b) phase respectively and they can be Therefore,
linked to molar Gibbs energy using the following equations:
m
@X bi 1 Xn X ai @ki
m
Sm ¼  @G m a ¼  j¼2 2 @X a
ð14Þ
@T @X j ki ki
m @Gmm
m ¼ a or b ð4Þ j
V m ¼ @P
Please note Eq. (14) also holds for non-stoichiometric compo-
Hereafter it is assumed that the two phases have identical mo- nents. As to be demonstrated later, Eq. (14) is useful to illustrate
lar volume. Subtracting Eq. (3a) from Eq. (3b) yields: the thermodynamic origin of partition coefficient.
Bringing Eq. (12) into Eq. (14), the evolution equation of parti-
X
n
ðX ai  X bi Þdli þ ðSam  Sbm ÞdT ¼ 0 ð5Þ tion coefficient of a stoichiometric component i is:
i¼1

Rearranging indexes and introducing the constrains of ki Xn @ki

 ¼0 ð15Þ
X
n X
n X ai j¼2 @X a
j
a
X i ¼ 1; X bi ¼1 ð6Þ
i¼1 i¼1
For any non-stoichiometric alloying components (the amounts
of these alloying components are allowed to vary) of b phase, those
It yields: evolution equations can be deduced knowing that dli must have
X
n the same value for the two phases staying in equilibrium with each
ðX ai  X bi Þdðli  l1 Þ þ ðSam  Sbm ÞdT ¼ 0 ð7Þ other according to Gibbs–Duhem relation [12]. Therefore the total
i¼2 differential of the diffusion potential is the same, i.e.,
As we have    
d lbi  lb1 ¼ d lai  la1 ðfor non-stoichiometric componentÞ
a
@Gm ð16Þ
¼ li  l1 ð8Þ
@X ai
The equal diffusion potential condition guarantees that there is
The deferential form of Eq. (8) can be written as: no exchange of component i and component 1 (the base compo-
 a P
n nent) between the two equilibrium phases. It could be reformu-
@ 2 Gam a
@ Gm 2 a
d @G m
@X a
¼ dðli  l1 Þ ¼ @X ai @X aj
dX j þ @X a @T dT ð9Þ lated into the following form:
i i
j¼2

Therefore bringing Eq. (9) into Eq. (7), one could have: !  a
@Gbm @Gm
Pn Pn 2 a a d ¼d ðfor non-stoichiometric componentsÞ ð17Þ
i¼2
b
j¼2 ðX j  a
X j Þ @X@ aG@Xm a dX i @X bi @X ai
j i
dT ¼ a Pn 2 a
ð10Þ
@Gbm
@T
 @Gm
@T
 b
j¼2 ðX j  X aj Þ @X
@ Gm
a @T Expanding the total differential in Eq. (17), one obtains:
j

@T
This equation gives the phase boundary slope @X a (the partial
i !
derivatives of solvus temperature with respect to X ai ), and are the X
n
@ 2 Gbm @ 2 Gam @ 2 Gbm Xn
@ 2 Gam
b a
generalized version of the Gibbs–Konovalov equation for dX j ¼ a  dT þ dX j ð18Þ
j¼2 @X bi @X j b b
@X i @T @X i @T j¼2
@X ai @X aj
multi-component systems. It defines the evolution of the solvus
temperature upon the equilibrium a phase composition. It forms Eqs. (11) and (18) give n1 linear equations relating the infini-
the first set of partial differential equations used in the proposed tesimal variations of the b phase compositions to the infinitesimal
integration-based technique. The authors would like to acknowl- variations of the a phase compositions and temperature (Please
edge one of the reviewers who had pointed out this concise deriva- a b
note dT has been already related to dX i by Eq. (10)). If dX i is con-
tion of Eq. (10). a
sidered as the unknown (the dependent variables) and dX i known
2.1.2. Calculation of the partial derivatives of partition coefficients (the independent variables), solving this set of linear equations
Below the relations between the infinitesimal variations of the gives the relations of each infinitesimal variation of the b phase
b
two equilibrium phase compositions are derived to obtain the par- composition, dX i , to the infinitesimal variations of the a phase
a
tial derivatives of partition coefficients with respect to the equilib- compositions, dX i . Therefore all partial derivatives of X bi with re-
rium a phase composition. These partial derivatives have received @X bi
spect to X aj , @X a , can be found. Recalling Eq. (14), the equations gov-
relatively little consideration in the literature to our knowledge. i

The stoichiometric component (if any) in the b phase is exam- erning the evolution of partition coefficients could be obtained. For
ined first. For any stoichiometric component in the b phase, its clarifications, the governing equation is given only for the binary
composition is constant. One can easily write: case below. However, it is just a matter of linear algebra operations
to derive those equations for multi-component systems. In such a
b
dX k ¼ 0 ðfor stoichiometric component kÞ ð11Þ dX bi
binary system, dX ai
could be written readily from Eq. (18) as
Therefore all of the partial derivatives of the composition of a  
stoichiometric component with respect to the equilibrium a phase ðX b2  X a2 Þ @
2 a
Gm @ 2 Gam @ 2 Gb
 @X b @T
m
@X a2 @T @ 2 Gam
b
composition are zero. @ 2 Gbm dX i @X 2a
2
2
2 a ¼ a 2 a
þ 2
ð19Þ
@Gbm
@X bk @X b2 dX 2
@T
 @G
@T
m
 ðX b2  X a2 Þ @X
@ Gm
a @T @X a2
2
¼0 ð12Þ
@X aj Bringing Eq. (19) into Eq. (14), one obtains
3156 Q. Du, M.A. Wells / Computational Materials Science 50 (2011) 3153–3161

 
2 a
! @ 2 Gam @ 2 Gb
ðX b2  X a2 Þ @X
@ Gm
 @Xb @T
m
1 X a2 dk2 @ 2 Gam
a @T @X a2 @T
@ 2 Gbm 2 2
 ¼ þ ð20Þ
@X b2
2
k2 k22 dX a2 @Gbm
 @G m
a
 ðX b2  X a2 Þ @X
@ Gm
a @T
2 a
@X a2
2

@T @T 2

Eq. (20) reveals the thermodynamic nature of the partition coef-

ficients. It forms the second set of partial differential equations
used in the proposed integration-based approach.

2.2. Extension to multi-phase boundary

Consider the equilibrium of a solid solution phase with other

p1 phases (b, c, d, e, etc.) where p (larger or equal to 2) is the total
number of phases in equilibrium. Eq. (10) could be written in the
following form for the equilibrium between a and m phase (m is
one of the other p1 phases (b, c, d, e, etc.)):
Fig. 2. Illustration of the numerical integration technique.
Pn Pn m 2 a a
i¼2 j¼2 ðX j  X aj Þ @X@ aG@Xm a dX i
j i
dT ¼ Pn ð21Þ
@Gmm a
m a @ Gm 2 a

@T
 @G
@T
m
 j¼2 ðX j  X j Þ @X a @T lution equations. For the simplest case, the procedure is that at a
j
known starting point, denoted by A0 in Fig. 2, the slope to the curve
Eq. (18) could also be extended to component i of each pair of at A0 can be computed using evolution equations. Take a small step
equilibrium phases (between a and m phase): along that tangent line up to a point A1. If it is assumed that A1 is
! still on the curve, the same reasoning as for the point A0 above
Xn
@ 2 Gmm m @ 2 Gam @ 2 Gmm Xn
@ 2 Gam a
dX j ¼  m dT þ dX j ð22Þ can be used. After several steps, a polygonal curve A0A1A2A3. . . is
j¼2
@X mi @X mj a
@X i @T @X i @T j¼2
@X ai @X aj computed.
Mathematically, the tracking problem has the following generic
Together with some appropriate constrains, Eqs. (21) and (22)
form:
allow to track multi-phase boundaries. The tracking procedure will
be illustrated in Sections 3.3 and 3.4 respectively by taking two dy
typical phase boundaries as examples. One of the examples is the ¼ f ðx; yÞ; yðx0 Þ ¼ y0 ð24aÞ
dx
mono-variant line in a ternary system along which three phases
are in equilibrium. The second is the a + b and a + b + c phase Then the following formula can be used to calculate the phase
boundary in a quaternary system. These two examples represent diagram:
typical scenarios one would encounter in the calculation of a Zero Z xn
Fraction Line (ZPF) in CALPHAD.
yn ¼ y0 þ f ðx; yðxÞÞdx ð24bÞ
x0
2.3. Verification of the mathematical representation
To deal with multi-dimensional numerical integration, one has to
The correctness of these derivatives and the integration meth- phrase the multiple integral as repeated one-dimensional integrals.
odology can be tested in the first place by applying them to an ideal The following equations can be used:
dilute binary A–B system, in which the molar Gibbs energy func- R xn1 ;x02 ;...;x0m
tions of the two equilibrium phases can be described using simple ynx1 ¼ y0 þ x01 ;x02 ;...;x0m
f ðxi ; yðxi ÞÞdx1
solution model. It is well known that the solubility line and parti- R xn1 ;xn2 ;...;x0m
tion coefficient for this system can be calculated using a solubility ynx1 ;x2 ¼ ynx1 þ xn1 ;x02 ;...;x0m
f ðxi ; yðxi ÞÞdx2 ð25Þ
product method as given below: ...
! R xn1 ;xn2 ;...;xnm1 ;xnm
GbB  GaB ynx1 ;x2 ;...;xm ¼ ynx1 ;x2 ;...;xm1 þ xn1 ;xn2 ;...;xnm1 ;x0m
f ðxi ; yðxi ÞÞdxm
a
X B ¼ exp ð23aÞ
RT
There are many numerical integration techniques one could
  choose from. Three integration techniques, namely Euler, Runge–
GbB GaB Kutta and Predictor–Corrector (PC) methods respectively, have
exp RT
  been implemented. The simplest one is the first order Euler meth-
kB ¼ ð23bÞ
GaA GbA od which uses the following equation to evaluate the one step
1  exp RT integration.

Please refer to Appendix A for the arrival of Eq. (23) by both the Z xnþ1
solubility method and the proposed integration approach. In this f ðx; yðxÞÞdx ¼ f ðxn ; yn ÞDx ð26Þ
xn
special case, instead of numerical integration, analytical integra-
tion was applicable due to simplicity of the involved solid solution The widely used Runge–Kutta integration method has the
model and the ideal dilute approximation. advantage of error monitor and adaptive step control. In other
words, it is able to achieve the pre-specified level of accuracy by
2.4. Numerical integration technique automatically adjusting the integration step size. The practical
method could be the Predictor–Corrector (PC) algorithm proposed
In the previous sections by deriving the evolution equations of in [10]. As its name suggests, this algorithm consists of two parts:
two parameters that control the shape of the phase boundaries, predictor and corrector. The predictor performs a numerical inte-
i.e., the solvus temperature and partition coefficient, the phase gration, usually very coarse, to give a prediction of the yn+1. The
boundary tracking has been formulated as an initial value problem. corrector is a procedure where the prediction is ‘‘polished’’ to the
Numerical integration techniques could be used to solve these evo- required level of accuracy.
 Q. Du, M.A. Wells / Computational Materials Science 50 (2011) 3153–3161 3157

3. Validation phase compositions. This feature has been well produced with the
integration approach. The good agreement as observed from Figs.
3.1. Tracking of two-phase boundary in binary systems 3 and 4 verifies the applicability of the proposed method to real
systems.
The integration-based approach was used to track the phase
boundaries of two binary systems, an Al–Sc system with a FCC
phase and Al3Sc phase in equilibrium and an Al–Mg system with 3.2. Tracking of two-phase boundary in ternary system and a
liquid and A12_AlMg in equilibrium respectively. Then the predic- quaternary system
tions were compared to the ones by Thermo-Calc. The thermody-
namic functions of molar Gibbs energy functions for these two In this section the proposed approach is used to calculate phase
systems were taken from the Thermo-Calc TTAL6 and TCBIN data- boundary of an Al–Mn–Si ternary system consisting of an Al–Mn–Si
base respectively. FCC phase in equilibrium with ALPHA phase and an Al–Fe–Mn–Si
The calculated solubility limit line of the Al–Sc is shown in Fig. 3 quaternary system consisting of an Al–Fe–Mn–Si FCC phase in equi-
with the markers. The solid line is the results obtained by Thermo- librium with LIQUID phase. The thermodynamic model of each
Calc. The solubility line of the Al–Mg system calculated by the inte- phase is taken directly from TTAL6 database and the programming
gration technique is shown in Fig. 4 with markers along with the interface of Thermo-Calc, TQ, is called to compute the partial deriv-
ones calculated using Thermo-Calc (solid lines). A feature of the later atives of molar Gibbs energy of each phase.
diagram is the existence of a turning point along the temperature It is useful to illustrate the multi-dimensional integration proce-
axis, i.e., for a given temperature, there are two pairs of equilibrium dure by taking the Al–Mn–Si system as an example. The numerical
integration is performed in a square in FCC phase compositional
space with the Mn and Si composition as its two axes. It starts at
the bottom-left corner. The values of solvus temperature and the
equilibrium ALPHA phase compositions at this corner are the initial
values for the numerical integration and have been computed by
Thermo-Calc prior to the integration. Then the integration goes step
by step along the Mn compositional axis until the right boundary of
this square is reached. Thereafter the integration goes one step up
along the Si compositional axis. At this new Si composition value,
the integration continues along the Mn compositional axis heading
to left boundary. Upon reaching the left boundary, it goes one step
up again along the Si compositional axis and then heading to the
right boundary. This process is repeated until the whole square
has been covered. By this ‘‘sweeping’’ process the multi-dimen-
sional integration process has been transformed into a series of
one-dimensional integration.
Fig. 5 shows the solvus calculated by the integration method
(the color map) and Thermo-Calc (the wire frame) in the square
compositional space. The X axis is Mn atomic fraction and Y axis
is Si atomic fraction with (0.0036, 0.0014) as the bottom-left corner
and (0.0072, 0.0028) as top-right corner. They are in very good
Fig. 3. Comparison of the solubility limit lines for a binary system with Al–Sc solid agreement.
solution phase in equilibrium with Al3Sc phase calculated by Thermo-Calc (the solid
line) and the integration method (the markers).

Fig. 5. The solvus surface of FCC in equilibrium with ALPHA phase in an Al–Mn–Si
Fig. 4. Comparison of the numerical integration technique with Thermo-Calc on the system calculated by the integration method (the color map) and Thermo-Calc (the
prediction of the two-phase boundary of an Al–Mg system. wire frame).
3158 Q. Du, M.A. Wells / Computational Materials Science 50 (2011) 3153–3161

According to Eq. (21), the following two equations arrive:

P3 P3 2 L L
i¼2
FCC
j¼2 ðX j  X Lj Þ @X@ LG@Xm L dX i
j i
dT ¼ P3 ð27aÞ
@GFCC L
FCC @ Gm 2 L
m
@T
 @Gm
@T
 j¼2 ðX j  X Lj Þ @X L
@T j

P3 P3 2 L L
i¼2
Mg2Si
j¼2 ðX j  X Lj Þ @X@ LG@Xm L dX i
j i
dT ¼ Mg2Si L P3 2 L
ð27bÞ
Mg2Si @ Gm
 X Lj Þ @X
@Gm
@T
 @Gm
@T
 j¼2 ðX j L
@T j

Starting from a known initial three-phase equilibrium, the two

equations allow us to calculate the increments of Liquid phase
composition for a given temperature increment. Therefore it allows
us to track the mono-variant line which lies in both of the L/L + FCC
phase boundary and L/L + Mg2Si phase boundary. Please note here
it is also possible to choose the increments of any involved vari-
ables other than temperature to track the mono-variant line.
To track tie-lines, Eq. (22) has to be invoked. One could write
Fig. 6. The composition–temperature isopleth of an Al–Fe–Mn–Si system calcu- down the following equations for FCC phase:
lated by the integration method (the markers) and Thermo-Calc (the solid lines).
!
X
3
@ 2 GFCC FCC @ 2 GLm @ 2 GFCC
m m
dX j ¼  dT
FCC FCC
j¼2 @X 2 @X j @X L2 @T @X FCC
2 @T
For the Al–Fe–Mn–Si system, a temperature–composition iso-
pleth was calculated with the Mn and Si content in liquid phase X
3
@ 2 GLm L
þ dX j ð28aÞ
fixed at 1 at% and the Fe content varying from 0.1 at% to 1 at%. j¼2 @X L2 @X Lj
The solidus and liquidus obtained by the integration method is
shown with the markers in Fig. 6 while the ones by Thermo-Calc !
X
3
@ 2 GFCC FCC @ 2 GLm @ 2 GFCC
with the solid lines. They are in a very good agreement. These m m
dX j ¼  dT
two case studies confirm the applicability of the proposed ap-
FCC FCC
j¼2 @X 3 @X j @X L3 @T @X FCC
3 @T
proach to the boundary tracking of multi-component alloys. X
3
@ 2 GLm L
þ dX j ð28bÞ
3.3. The tracking of mono-variant lines in a ternary system j¼2 @X L3 @X Lj

In which index 2 refers to Mg component and 3 to Si compo-

The ternary system taken as an example is Al–Mg–Si system. In
nent. This equation allows us to calculate increments of the FCC
this system along a mono-variant line Liquid phase co-exists with
phase composition for the given temperature increment.
any two of FCC, Mg2Si and Silicon phase. The three-phase equilib-
Because Mg2Si is a stoichiometric phase, Eq. (11) instead of
rium of Liquid, FCC and Mg2Si is chosen below to illustrate the
Eq. (22) applies here:
tracking procedure. The other three-phase equilibriums,
Liquid + FCC + Si and Liquid + Mg2Si + Si, were also computed. The Mg2Si
dX 2 ¼0
calculation results are compared with the ones obtained by Ther- Mg2Si
ð29Þ
dX 3 ¼0
mo-Calc in Fig. 7.

0.20

0.16 557.89 ºC
568.04 ºC
Mg Atomic Fraction

576.85 ºC

0.12 575.57 ºC

586.85 ºC

0.08

0.04

0.00
0.00 0.02 0.04 0.06 0.08 0.10
Si Atomic Fraction

Fig. 7. The calculated mono-variant lines and tie lines by the integration method (markers) and the Thermo-Calc (solid lines) for a Al–Mg–Si ternary system.
 Q. Du, M.A. Wells / Computational Materials Science 50 (2011) 3153–3161 3159

0.09

Mg atomic fraction
0.06

0.03

0.00
0.000 0.002 0.004 0.006 0.008
Fe atomic fraction

Fig. 8. An isothermal section (820 K) of an Al–Fe–Mn–Si phase diagram with the constraint of fixed Si content (0.96 at%). The solid lines were calculated by Thermo-Calc and
the markers by the proposed integration method.

The initial value for the calculation is the invariant point (the which are applicable to the tie-line triangles. Eqs. (10) and (18)
co-existence of the four phases). Then the mono-variant line (the gave eight differential equations for the three equilibrium phases
three-phase equilibrium line) is tracked by repeatedly calculating considered here. The second set equation is from the fact that this
the increment of phase compositions induced by a small tempera- boundary is zero fraction line for LIQUID phase. Therefore, the
ture increment via Euler integration. summation of phase fraction of Mg2Si and FCC is equal to 1. It could
Fig. 7 shows the mono-variant lines together with the 5 tie lines be written in a differential form:
(dotted lines) linking the equilibrium liquid and FCC phase compo- fcc Mg2Si
sition. The equilibrium temperatures for these tie lines were df þ df ¼0 ð30Þ
shown within their nearby text boxes. The mono-variant lines Meanwhile, there are three mass balance equations for point P:
obtained by the integration are plotted with markers and the ones
by Thermo-Calc with solid lines. They are in good agreements X PFe ¼ f fcc X fcc
Fe þ f
Al6Mn Al6Mn
X Fe
validating the applicability of the proposed methodology to
X PMn ¼ f fcc X fcc
Mn þ f
Al6Mn Al6Mn
X Mn ð31Þ
multi-phase boundary tracking.
X PSi ¼f fcc
X fcc
Si þf Al6Mn
X Al6Mn
Si ¼ 0:0096
Please note in the last equation above the imposed constraint of
3.4. The tracking of three-phase boundary in a quaternary system
(the fixed Si content) is applied. Certainly the mass balance equa-
tions could be written in differential forms:
The three-phase boundary considered in this section is the one
between a two-phase zone and a three-phase zone. In contrast to P
dX Fe ¼ X fcc
fcc fcc
þ f fcc dX Fe þ X Al6Mn
Al6Mn Al6Mn
þ f Al6Mn dX Fe
Fe df Fe df
the mono-variant lines tracked in the previous sections, it is not part P fcc fcc Al6Mn Al6Mn
of any single-phase region implying none of the three phases has a dX Mn ¼ X fcc
Mn df þ f fcc dX Mn þ X Al6Mn
Mn df þ f Al6Mn dX Mn ð32Þ
fcc fcc Al6Mn Al6Mn
100% presence along the boundary. However, there is one phase 0¼ X fcc
Si df þf fcc
dX Si þ X Al6Mn
Si df þf Al6Mn
dX Si
whose fraction is equal to zero. Given the importance of these Zero
Here all the 12 differential equations have been found to use the
Fraction Lines (ZPF) in phase diagram calculation, it is necessary to
integration method.
demonstrate how to track them basing on Eqs. (10) and (18) together
For convenience, the tracking procedure started from the point
with some appropriate additional constrains. Fig. 8 shows a 2D iso-
where the four phases are in equilibrium. The integration axis was
thermal section (temperature 820 K) of an Al–Fe–Mg–Si phase
taken along X Liquid . All of the incremental parts of the other 12
diagram, in which Liquid, FCC and Mg2Si phase are in equilibrium Mg
unknowns were worked out by the 12 differential equations. The
with the constraint of the fixed average Si content (0.96 at%). The
calculated boundaries by the proposed method (the markers) are
two lines, which lie between FCC + Mg2Si zone, FCC + LIQUID + Mg2-
in good agreement with the one obtained by Thermo-Calc (the
Si zone and FCC + LIQUID zone, have been calculated by Thermo-Calc
solid lines) as shown in Fig. 8. It is a strong evidence for the ability
(solid lines) and the proposed integration method (markers). The
of the integration method to track the general multi-phase bound-
tracking procedure is illustrated below by taking the boundary
aries encountered in multi-component phase diagrams.
between FCC + Mg2Si and FCC + LIQUID + Mg2Si as an example.
Let’s denote a point located in this boundary, P, as (X PFe ; X PMg ).
These are two main unknowns. Meanwhile it is also mandatory 4. Conclusions
to track its associated  tie-line triangles,  which are nine  unknown

represented by X fcc fcc fcc
Fe ; X Mg ; X Si , X liquid
Fe ; X liquid liquid
Mg ; X Si , X Mg2Si
Fe ; For the tracking of multi-phase boundaries, integration equa-
X Mg2Si
Mg ; X Mg2Si
Si Þ. Please note the tie-line triangles are not located in tions have been derived for the two important phase diagram
this 2D section. Together with the two non-zero phase fractions parameters, solvus temperature and partition coefficients, based
for FCC and Mg2Si, there are 13 unknowns in total. The first set on the Gibbs–Duhem relation. Then the tracking of phase boundary
of equations to solve out the 13 unknowns are Eqs. (10) and (18), is transformed into an initial value problem, which could be solved
3160 Q. Du, M.A. Wells / Computational Materials Science 50 (2011) 3153–3161

by any numerical integration method. The methodology has been Below it is going to shown the same result could be derived
validated initially by the agreement with the solubility product from the integration method. It begins with the first and second
method for the system with ideal solid solution models. Its appli- partial derivatives of the molar Gibbs energies for the two phases:
cability to real binary and multi-component systems with com- @Gam @GaB @GaA
mercial thermodynamic databases has been verified by its good @T
¼ X aB @T
þ RX aB ln X aB þ ð1  X aB Þ @T
þ ð1  X aB ÞR lnð1  X aB Þ
agreement with the CALPHAD software Thermo-Calc calculation @Gbm @GbB @GbA
@T
¼ X bB @T
þ RX bB ln X bB þ ð1  X bB Þ @T
þ ð1  X bB ÞR lnð1  X bB Þ
results. This method is alternative technique for the tracking of
multi-phase boundary. It might be also useful for the calculation ðA5aÞ
of real phase diagram when combined with the ZPF overlapping
@Gam
technique discussed in [13] and a global minimization technique. @X aB
¼ GaB þ RT ln X aB  RT lnð1  X aB Þ  GaA
It may also be an alternative to the iterative methods that have @Gbm
ðA5bÞ
been widely used in precipitation kinetic modeling to solve the lo- @X bB
¼ GbB þ RT ln X bB  RT lnð1  X bB Þ  GbA
cal equilibrium condition equations.
@ 2 Gam @GaB @GaA
@X aB @T
¼ @T
þ R ln X aB  R lnð1  X aB Þ  @T
Acknowledgements ðA5cÞ
@ 2 Gbm @GbB @GbA
@X bB @T
¼ @T
þ R ln X bB  R lnð1  X bB Þ  @T
The authors would like to thank Prof. Warren J. Poole from the
University of British Columbia (Vancouver, Canada) for useful dis-  
@ 2 Gam
cussions on this paper and his encouragement. Prof. A. Pelton and 2 ¼ RT X1a þ 1X
1
a
@X aB B B
Dr. G. Eriksson of Ecole Polytechnique (Montreal, Canada) are   ðA5dÞ
@ 2 Gbm
acknowledged for their reading of the first version of the manu- 2 ¼ RT X1b þ 1X
1
b
@X bB B B
script and constructive criticisms.
Replacing the partial derivative in Eq. (10) with Eq. (A5), we
Appendix A. Solubility product method and the integration have
technique
dT ðX bB  X aB ÞRTðX1a þ 1X
1
aÞ
a ¼
B B
0 b
1
In this appendix, the comparison of the solubility product meth- dX B X
B
b
ð1X Þ
@Gb @GaB @Gb @Ga X bB ð1X bB Þ B
od with the integration method is conducted by taking two ideal X bB ð @TB  @T
Þ þ ð1  X bB Þ @TA  ð1  X bB Þ @TA þ R ln @ X
b
b
A
B 1X
dilute solid solution phases as an examples. The chemical potential X Ba ð1X aB Þ B

formulations for the two ideal solid solution phases are: ðA6Þ
a a
laB ¼ GB þ RT ln X B Employing the dilute solution approximation, Eq. (A6) could be
b
l ¼ þ RT GbB ln X bB simplified to
B
ðA1Þ
laA ¼ RT lnð1  X aB Þ þ GaA dT
RT
X aB
a ¼   ðA7Þ
lbA ¼ RT lnð1  X bB Þ þ GbA dX B @GaB @GbB
 @T
þ R ln X aB  @T
The equal chemical potential conditions yield:
a The above equation could be reformulated into
GB þ RT ln X aB ¼ GbB þ RT ln X bB
ðA2Þ
RT lnð1  X aB Þ þ GaA ¼ RT lnð1  X bB Þ þ GbA dðGbB  GaB Þ ¼ dðRT ln X aB Þ ðA8Þ

Therefore the following equations could be derived directly Therefore integrating Eq. (A8) gives
from Eq. (A2): !
GbB  GaB
Xa GbB GaB
X aB ¼ C 1 exp ðA9Þ
ln XBb ¼ RT
RT
B
ðA3Þ
1X aB GbA GaA This equation has the same formulation as Eq. (A4a). An appro-
ln 1X b ¼ RT
B priate initial condition is required to determine the constant C1 in
Eq. (A9).
Employing the dilute solution approximation X bB ! 1; X aB ! 0;
To derive the partition coefficient equation, Bringing Eq. (A6)
the two solubility limit lines of this binary system for a given tem-
into Eq. (20) and applying the dilute solution approximation, one
perature are:
! obtain:
a GbB  GaB Xa !
X B ¼ exp ðA4aÞ RT d kBB dT dðGaB  GaA  GbB þ GbA Þ
RT a ¼ a þ R ln X aB þ R lnð1  X bB Þ
1  X bB dX B dX B dT
! RT
GaA  GbA þ
X aB
ðA10Þ
1  X bB ¼ exp ðA4bÞ
RT
Eq. (A10) could be rewritten as
Eq. (A4a) and Eq. (A4b) lead to the partition coefficient   
Xa
equation: d RT ln 1  B ¼ dðGaB  GaA  GbB þ GbA Þ þ dðRT ln X aB Þ ðA11Þ
  kB
GbB GaB
exp RT Therefore integrating Eq. (A11) gives
kB ¼   ðA4cÞ !
GaA GA
b
Xa GaA þ GbB  GaB  GbA
1  exp RT 1  B ¼ C 2 X aB exp ðA12Þ
kB RT
 Q. Du, M.A. Wells / Computational Materials Science 50 (2011) 3153–3161 3161

The same partition coefficient equation as Eq. (A4c) could be de- [4] Y. Austin Chang, W. Alan Oates, Materials Thermodynamics, John Wiley & Sons,
2010.
rived from (A12) supposing an appropriate initial condition has
[5] W. Cao et al., CALPHAD 33 (2009) 328–342.
been provided to determine the constant C2. [6] M. Hillert, Bulletin of Alloy Phase Diagrams 2 (3) (1981) 265–268.
[7] H.L. Lukas, J. Weiss, E.T. Henig, CALPHAD 6 (3) (1982) 229–251.
[8] Ursula R. Kattner, JOM 49 (1997) 14–19.
References [9] H. Gaye, C.H.P. Lupis, Metallurgical Transactions A (6) (1975) 1057–1064.
[10] E.L. Allgower, K. Georg, Introduction to Numerical Continuation Methods,
SIAM, 2003.
[1] P.J. Spencer, Computer Coupling of Phase Diagrams and Thermochemistry 32 [11] D.A. Goodman, J.W. Cahn, L.H. Bennett, Bulletin of Alloy Phase Diagrams 2
(2008) 1–8. (1981) 29–34.
[2] N. Saunders, A.P. Miodownik, Pergamon (1998) 283. [12] M. Hillert, Bulletin of Alloy Phase Diagrams 3 (1982) 4–8.
[3] Hans Leo Lukas, Suzana G. Fries, Bo Sundman, Computational Thermodynamics [13] H. Gupta, J.E. Morral, H. Nowotny, Scripta Metallurgical 20 (1986) 889–894.
the Calphad Method, Cambridge University Press, 2007.