Professional Documents
Culture Documents
33-54, 2002
Pergamon © 2002 Published by Elsevier Science Ltd
0364-5916/02/$ - see front matter
PII: S 0 3 6 4 - 5 9 1 6 ( 0 2 ) 0 0 0 2 3 - 8
Abstract. Cluster Variation Method (CVM) has been very successful in the computations of alloy phase
diagrams as well as in many problems of the materials science related to the phase transitions. Originally, CVM
was developed in the framework of the so-called rigid lattice approximation, but it has recently been extended
to include continuous atomic displacements due to thermal lattice vibration and local atomic distortion due to
size mismatch of the constituent atoms. In the present study, we focus our attention on the latter continuous
displacement treatment of CVM. The continuous displacement (CD) formulation of the CVM is applied to
study the phase stability of the binary alloys. The basic idea is to treat an atom which is displaced by r from its
reference lattice point as a species designated by r. The effects of continuous atomic displacement on the
thermodynamic quantities and phase transitions of binary alloys are investigated in detail. We also discuss the
extension of the CD treatment of CVM to the calculations of solid-liquid and gas liquid phases transitions.
© 2002 Published by Elsevier Science Ltd.
Keywords: Cluster variation method, continuous displacement treatment, pair approximation, grand potential,
phase separation, order-disorder transition
I. Introduction
The Cluster Variation Method (CVM) has been developed, improved and applied for more than half
century. The key concept of the entropy formula was introduced in 1949 by Kikuchi and published in 1951
[51Kik]. The entropy and the free energy of a binary alloy system were written in terms of long- and short-
range order parameters by Takagi [41Tak]. He derived the equilibrium state by minimizing the free energy with
respect to these variables, and showed that the equations were identical to those of Bethe [35Bet]. When the
CVM was introduced, the immediate purpose was to find a technique to go beyond the approximation of Bethe
and Takagi. The derivation of the entropy expression in the first report of the CVM [51Kik] made use of the
requirement of translational symmetry as the lattice was constructed point by point. It was later found that the
method was closely related to the superposition approximation ofKirkwood [35Kir].
The CVM provides us analytical expressions of the internal energy, configurational entropy and free
energy of the system in terms of the cluster probability variables. The basic cluster is usually taken in such a
way that it contains the maximum interaction range one wants to take into account. The equilibrium state is
determined by a variational principle, i.e., by a variation of these cluster probabilities until a maximum of the
thermodynamic potential is obtained. The simple computational technique, natural iteration method, is often
used to get the free energy minimum equilibrium states.
The CVM has been successfully applied, for example, to investigate temperature-composition phase
diagrams of alloys for complex Hamiltonians [94Fon], which include pair and many-body interactions. In these
CVM calculations, the entropy expression was formulated for permutations of atoms among rigid lattice points
[94Kik]. This type of approach will be referred to as the conventional CVM in the present paper. In real alloys,
however, there are problems which need more than permutations of atoms among rigid lattice points. These
33
34 R. KIKUCHI AND K. MASUDA-JINDO
problems include local lattice distortion caused by the atomic size differences [95Hor,88Moh,89Ter], thermal
lattice vibration and the effect of elastic interactions on phase diagrams [73Lar]. Responding to the need of
removing the restriction of the rigid lattice, the continuous displacement (CD) formulation has been introduced
recently [92Kik,97Kik,94Fin]. Since the CD formulation is an extension of the conventional CVM, we present
the basis of the CD by comparing it with the conventional CVM treatment.
transition involves more 'dynamic' phenomena originating from atomic diffusion and lattice dynamics. Recently,
the pre-transition phenomena (pseudo critical slowing down phenomena) of Ll0 ordered phase have been
studied by Mohri [00Moh] within the CVM and path probability method (PPM). The extension of conventional
CVM to the time dependent PPM has been developed by Kikuchi [66Kik,74Kik].
where
In[X] -= X lnX - X. (10)
In the above eq.(9), g2(rl, r2) denotes the pair probability function with atomic displacements ri and r2 at
atomic sites 1 and 2, respectively, and ~, the Lagrange multiplier for the normalization ofh(rt, r2, r3) probability
function. ~(ri, rj,) and tx(ri) are the Lagrange multipliers for the lattice symmetry constraints and R denotes the
rotation operator of angle ~/3. The trimer probability function h(rb r2, r3) is obtained by minimizing the free
energy • of the system with respect to h(rh r2, r3) as
[3~, 1
h(rl, r2,r 3) = exp[-~-] ex'p[-~ [c(rl, r2) + e(r2,r3) + c(r3,rl)]] exp(-(Tl(r1,r2) - vl(-r2,-rl))
-(q(r2~r3)-T~(-r3~-r2))-(~t(r3~rz)-~(-r1~-r3)))xexp(-(cx(r~)-~(Rr1))-(a(r2)-~(~r2)) (11)
(oft. r r ~V/2
- (ct(r3 ) - ~ [ R r "~))'~" i, 2, 3,J .
3~ (f(rl,r2,r3))l/6
Here, g(rb rz, r3) and f(rb r2, r3) functions are defined by
g(r~,h,r3) =_g2(rl,r2)g2(r2,r3)g2(r3,rl) , (12)
and
f(rl,r2,r3) =- f ( r ~ ) f ( h ) f ( r 3 ) . (13)
The similar formulations can also be done for other single-component materials with different crystal
structures, e.g., the pair approximation for 2D-square, fcc and bec crystals, and tetrahedron cluster
approximation for fcc crystal. The CD treatment of CVM for the three dimensional (3D) single component
systems can be formulated quite similarly to the case of 2D systems. For treating 3D systems, the atomic
displacements from the reference lattice sites are taken to finite numbers of the order of ~500 sites as shown in
Fig.2. Accordingly, the volume integrals are replaced by the discrete summation over the cubic meshes [34].
Crystal symmetries are taken into account in the constraint terms as in eq. (9). For instance, the rotational
symmetries, R2x, R2y and R2z are taken into account for fcc lattice as shown in Fig.3a. The CD-CVM
formulations for cubic alloys will be given in the subsequent subsections.
In Fig.4, we present the thermal lattice expansions a(T) both for 2D and 3D crystals, calculated by the
CD treatments of CVM. The thermal lattice expansions are given by normalized quantities a(T)/a(0), as divided
by those of the absolute zero temperature, and 12-6 type of Lennard-Jones potentials are used. The interatomic
interactions are taken into account for ZI nearest-neighbor atom pairs, Zl being 4, 6 and 12 for 2D-square,
2D-hexagonal and face centered cubic crystals, respectively. In general, the normalized thermal lattice
expansion decreases with increasing the number of nearest-neighbors Z1 using the same interatomic potentials.
It is also noticeable that the thermal lattice expansion becomes smaller when the better CVM approximation is
used [99Kik]. For instance, for fcc crystals, the quasi-chemical tetrahedron cluster approximation gives much
smaller thermal expansion than those calculated by the CVM pair approximation, as shown in Fig.4a.
The more realistic calculations of thermal expansions of Cu and Au metals are performed by using the
electronic many body potentials [93Cle]. The calculated thermal lattice expansions of Cu and Au metals are
compared in Fig.4b and 4c, respectively. The thermal expansions calculated by pair and quasi-chemical
tetrahedron cluster approximations are presented by symbols C) and I-l, respectively. In the figures 4b and 4c,
they are also compared with the corresponding experimental results (solid curves) [72Tou] as well as with the
molecular dynamics simulations (dashed lines) [90Hol]. The pair approximation of the CD treatment of CVM
tends to overestimate to s.ome extent the thermal lattice expansion of fcc metals as in the MD simulations, but
considerable improvements can be made by using the quasi-chemical tetrabedron cluster approximation.
For treating fec binary alloys, the point distribution and pair probability functions are denoted byflr) and
g(r~, r2), respectively. When it is needed to identify the two end points of a pair, we use 1 and r, and the point
functions for the two end points are given as
fil (r0 and fir (r2) with i, j=l, 2. (14)
On the other hand, the pair probability functions are written as
gij (rl, r2); rl is for 1, and rz for r.
The reduction relations of the pair probability functions can be written as
f, (r,)= J'dr_,E gi~(r, , r 2 ) i = l , 2, (15)
J
fj, (r2)= ~dr, Z g,j (r, , r 2 ) j = l , 2. (16)
i
We denote the point probability of site i on the left end by x, and that of site j on the right end by xjr. They are
written in terms of the point and pair probability functions as
x,,= J'dr,fia(r,)= Idr, j'dr2Zg0(r,,r2) . (17)
J
The normalization equations for the probability functions are given by
1= Z xij = Idr, E f,~,)= ~dr_.x f,,(r.,)= j'dr, Idr,_E g 0 (r,,rz) . (18)
i i j ij
In deriving the probability functions, one must take into account the crystal symmetries of the given lattice.
The fcc lattice has the crystal symmetries of 4-fold rotations, with respect to the [001 ] crystal axes as shown in
Fig.3a. Clearly, for the 4-fold rotation symmetries, i.e., R4x, R4y and R4~ on the x, y and z axis, we have the
following relations:
f,(r0=fil(R4xrl); fir(r2)=fi~(Raxr2)
fil(rl)=fil(R4yrl ); fir(r2)=fir(R4yr2)
fil(rl)=f,(R4~rl); fir(r2)=fi~(R4zr2). (19)
In terms of the components, we have the following transformations:
R4~(x,y,z)=(x,-z,y); R4xt(x,y,z)=(x,z,-y),
R4y(x,y,z)=(z,y,-x); R4yq(x,y,z)=(-z,y,x).
R4~(x,y,z)=(-y,x,z); R4,q(x,y,z)=(y,-x,z). (20)
Then we see that the relations hold for R4x, R4y and R4~ as
R4,R4, (x, Y,Z)= R4,R4y (x,Y,Z)= R4yR4~(x, Y,z)= (z, x.Y). (21)
Therefore, the three rotation operations R4~, R4y and R4, can be taken as independent, and one may use R4~ and
R4~ operations as independent.
We now introduce the constraint terms C,, and C,~, in terms of the Lagrange multipliers ct
C~il =-~ drlaxil (rl){fil (rl)- f il (R4xrl )}
= ~drl {Ctxil (rl) - atxil (R4x rl )}fi/(rl ) =- J drl Axil (rl)fi/(rl ), (22)
where A~, (r,)-- et ~,,( r l ) - c%, (R,~ r, ). (23)
The entire constraint terms are Cax + Caz for which
Coa" =- Y" Coxil + ~ Cajl = I drl I dr2 ~. {Ax//(rl )+ Axjr (r2 )~q (rl ,r2 ). (24)
i j (j
and
Ca: -~ I drl I dr2Z {A=il(rl )+ A:jr (r2 )}gij(rl,'2 ) . (25)
ij
Since there are no sublattices in the phase-separating system, symmetries among A's are the same and gi ven as
A,,(r)= A,i,(R4, r), A,i,(r)= A,i.(R4:r). (26)
Thus, when one calculates A,ij(r) and Azil(r), and then A's for r are derived from them.
For the energy expressions, we use the 12-6 type of Lennard-Jones potentials, otherwise stated, and we
choose
which makes the chemical potential IXincrease as B-atom concentration increases. Using eq.(31), we rewrite
(30) in terms of the pair probability functions g(rj,rD as
,
÷,LTA 1- fdrl Idr,~.s go(rl,r,)
L
] (35)
When we minimize the grand potential W with respect to gij(ri,r2) function, we obtain
z~s~itaxi
(r~,~2) =-6~0 (n,.2)- 71n{fil(rl)f J"(r2)}+61n{gij(rl'r2)}- l fl{ui + I'tJ} (36)
- +.,-,o;/+
Then. it is straightfo~'ard to derive the basic equations of the pair distribution functions as
CLUSTER VARIATION METHOD IN THE COMPUTATIONAL MATERIALS SCIENCE 39
gij(rl,r2)=exp[fl-~--~ -. 1,£7c
/jr(r ,/+ ~'"
fl2/-~lkii+lAj }l{fil(rl)fjr(r2)~l/12 (37)
x exp{Axl(rl)+ Axjr(r2)+ Azil(rl)+ Azjr(r2)}i,j=l.2.
When these equations are satisfied and the free energy is a minimum, the Lagrangean constant for normalization
in (35) is L=~/kT.
The extension of the CD treatment of CVM to the quasi-chemical tetrahedron cluster approximation
[49Li] is straightforward. For this treatment, we take into account the symmetry of fcc lattice as shown in fig.3b.
We introduce the tetrahedron variables w in addition to the pair probability variables. Then, the energy of a
system with N lattice point is given by
When one minimizes the free energy ~ of the system with respect to the tetrahedron variable, one obtains
w0j25(ro; r~;r2; r5) = exp(flA - flSolzs(ro; r~;r2; rs ))(fo (ro)f (r~)fz (r2)f5 (r5)) 4
x exp(A~ (r0) + A~(fi) + A,(r2) + Ax(rs) ) (43)
x exp(A, (ro) + A, (rt) + A, (r2) + Az(r5)).
When the equilibrium state is solved, the Lagrangean constant for normalization 3. gives the free energy F per
single atom as in the pair approximation
X =F/N. (44)
Before going to the continuous displacement calculations, in order to examine the accuracy and to see
the effects of the new CD treatment, we briefly discuss the phase separation of binary fcc alloys calculated by
the conventional CVM. The three kinds of approximation schemes of CVM, i.e., pair, quasi-chemical
tetrahedron and CVM tetrahedron approximations are used. The entropy expressions for the three cases are
obtained by taking into account the characteristic cluster variables as [6]
S~i r/(ksN ) = 1 l~-~L(x,) - 6~]L(y 0) + 5, (45)
i /j
Sq-ch,r/(ksN) = 3 ~ L ( x , ) - ~ L ( W e , . ) + 2, (46)
i ijmn
-
, )
~Pl +gzp2 = -
f±s
I-tl drlf
\ i=l
rl +
j=l (,)t
I.tj dr'if 'j . (52)
In order to get general understanding of the continuous atomic displacements, we use the Lennard-Jones
potential for the interatomic potentials. Specifically, we choose the potential parameter values symmetrically as
CLUSTER VARIATION METHOD IN THE COMPUTATIONAL MATERIALS SCIENCE 41
general trends in the phase separating curves obtained by quasi-chemical cluster approximation remains
unchanged compared to those of the pair approximation.
C/2 1
- fdrlldr2•{itl+l.t2}gij(rl,r2). (66)
N 2 ij
Then the grand potential of the system is written as
where ~pis the damping factor. Using the H functions, one can solve
AAxiL (rl) = I ~HiL (R4xrl)+ 2HiL (R4x2 rl)+ HiL (R4x3 rl)} . (73)
The similar iterations can be done by changing x to z for AAz, and also for the R sublattice as those for L.
As one of example calculations, we take the following two sets of energy parameters:
second choice of the size parameters: ril0=l.0, r 120=1.05 and r 220=1.1. It can be seen in Fig. 7(b) that the
transition temperatures become asymmetric and show skewed structure. One can then come to the conclusion
that the critical temperature of order-disorder phase transformation of bcc alloys is reduced drastically by
amount o f - 40%, when the continuous atomic displacements arising from the thermal lattice vibration and the
size-misfit effects of the constituent atoms are taken into account.
In the light of the present theoretical findings of the order-disorder phase transition of the B2 ordered
alloys, we briefly comment on the previous theoretical calculations and comparison with the experimental
results. It has been pointed out that when the interaction potentials calculated by the first principles electronic
theory are used together with the conventional CVM in constructing the free energy, the order-disorder
transition temperature is estimated much higher than experiments and the area of single phase region is
narrower than experiments [95Hor,88Moh,89Ter]. It is interpreted as caused by neglecting the local distortion
of the lattice due to the size difference of the constituent atoms in the alloys. So far, the first principles phase
diagram calculation has been extensively performed for noble metal based alloys. For instance, the
characteristic topological feature of Cu-Au system has been reproduced well [95Hor,88Moh,89Ter]. However,
the order-disorder transition temperature is greatly overestimated. This tendency results from the
over-stabilization of the ordered phases due to the neglect of lattice distortion and thermal vibration effects
[92Kik,97Kik,94Fin]. The present calculations using the CD scheme of CVM, although for bcc alloys, clearly
shows that the order-disorder transition temperature are reduced significantly by taking into account the
continuous displacement of atoms around the reference lattice sites. In addition to the above mentioned noble
metal based alloys, it has also recently been pointed out that large discrepancy arises in the congruent
temperature of L10 FePd alloy, when neglecting the effects of the continuous atomic displacements [39].
j zn~x . ,[
CLUSTER VARIATION METHOD IN THE COMPUTATIONAL MATERIALS SCIENCE 45
When we minimize [3t~{XAB(j2j1)} with respect to the point probability variables f0z,j2,r), the equilibrium state
with the minimum value of the grand potential ff~ can be found as
flrgt({XXB(J2, Jz )}) ___1 fly, E 6(J2, k2,lr - R(m~)XAB(J2,m)+ In f(Jz, j2,r) - fl2(Jz)= 0. (77)
.c91e(J'z,j2,r) z m j2
This leads to the point probability function in the following form:
~H [ N!{Pairjak}N{Pairkbm}N{Pairmej}N ],
(87)
G3-cell [ {Cell J}N {Cell k}N {Cell m}N {3 - pts j akb me}N
f {~m}{n}{Akmn}{Ajkm}{Ajmn}{Ajnk }
G4-ce. -- n l (88)
LN, {km}{kn}{kj'}{mn}{nj}{jm}{Ouartetjkmn} J "
This expression is written for a small but finite size lattice constant, and the space is divided into N cells. Each
cell is either occupied by an atom or vacancy. The first factor is for permutation of occupied and vacant cells
among N cells. The product factor is the Correlation Correction Factor for the pairs, and goes over all j-k cell
pairs; a is the vector connecting two cells j and k. It turns out that the pair approximation of
[exp(S/kB)=WccuG2.ceu] does not lead to any phase changes. Also, the triplet approximation does not lead to a
stable liquid phase.
Then, the quartet approximation using all shapes and all orientations of 4-ceil clusters is applied. When
the grand potential is minimized, the 4-cell probability is derived as
f g~(a,b + C)
pl/n (a,b) ---/9 J dc[l - g(c) - g(b + c) - g(a + b + c) ÷
g(a) (91)
+ ga(a,b + c) + ga(b,c) _ ga(a,b,c) 1 .
g(a+b) g ( a + b ) g3(a,b) "
These are a part of the set of equations to be solved for the pair distribution function g(rl2). This set needs two
tiers of interactive procedures and forms the bottleneck in the numerical computation and has not been
successfully solved. For obtaining an approximate solution, we introduce an additional simplification on
equations, and obtained a pressure versus density isotherm of the van der Waals type.
For the numerical calculation of the pressure P and chemical potential ~t of the system, we use the
relations
t i p = l n u + I n , o - / 9 J'da[g(ayl)]+ l o 2 j'da j'db[3- g(a) + g(b)g(c)
-(g(a)g(b)+ (b)g(c) + g(c)g(a~i+ 393 (a, b )], (92)
and
~P=p-P~ Sda[g(a)-l]+-~v.3 j'da ~db[2-(g(a)+g(b)+g(e))+g3(a,b)], (93)
for the triplet case without the 3-body interaction.
The pressure P versus density (r03) values calculated by quartet treatment of CD-CVM formalism are
shown in Table 5. The critical temperature in the quartet approximation is found to be about kT/e=l.7. This
value can be favorably compared with the value of 1.3 by Hansen and McDonald [76Han] calculated for Ar.
The liquid phase is also confirmed by the radial distribution function g(r). Thus, it may be concluded that the
quartet formulation leads to the stable liquid phase.
4. Conclusions
The continuous displacement treatments of the CVM have been applied to the calculations of phase
diagrams of binary alloy systems. We have worked out with the binary alloy systems with phase separating
imcractions in the fcc and bcc structures. The phase-separating diagrams of a number of model systems are
calculated using both pair and quasi-chemical tetrahedron cluster approximations. The CD treatments of the
CVM have also been applied to investigate the order-disorder phase transition of bcc alloys using the pair
approximation. In general, it has been shown that the phase-transition temperatures are reduced drastically when
the continuous atomic displacements are taken into account. Therefore, the large discrepancy between the
experimental and theoretical phase diagrams [95Hor,88Moh,89Ter] calculated so far by the conventional CVM
and ab initio electronic theory can be removed by applying the CD scheme of the CVM.
In addition, the CD treatment of CVM has also been applied to study the surface segregation phenomena
of fcc binary alloys as well as the gas-liquid phase transition of a mono-atomic molecular fluid. In both
calculations, the CD treatment of CVM is successful and provides us the simple physical insights into the
problems.
References
Table 1. Number of small mesh points inside an nb sphere and the mesh size b/a when the sphere redius is 0.4a
n 2 3 4 5
Number o f points 33 123 257 51
b/a=O.4/n 0.2 0.133 0.1 0.08
Table 3. Parameter values for the many body potentials, Cu-Cu row is for the alternative choice.
Cu-Au values are for the L12 structure.
Table 4. Interatomic distance r0(13) in the bulk CuAu alloy with Xcu=0.5
Table 5. The pressure p as a function of density (r0"3) calculated by quartet treatment of CD-CVM
using the Lermard-Jones potential (kT/c=l.5)
r0"3 0.05 0.1 0.15 0.25 0.3 0.35 0.4 0.45 0.48 0.5 0.56 0.58 0.6
Pr03/e 0.07 0.115 0.14 0.17 0.16 0.14 0.13 0.121 0.12 0.121 0.135 0.155 0.165
50 R. KIKUCHI AND K. MASUDA-JINDO
a)
t • • ,
• .
• ! • • •
• • • •
• . ~:...
• • • • * ° • * * • •
• • •
• t •
• • • • • • • •
o o • ,t , • * • . •
• ...... . ..'..'..::j.~.
• • • • t • • • j • •
. . . . . e . -,.. • •
a ~-
b)
00000000000
00000000000
00000000000
00000000000
ooooo ooooo
000000-00000
00000000000
00000000000
00000000000
00000000000
00000000000
Fig. 1 The concentric circles of radius nb, b being an appropriate length: 8n points on the n-th circle.
CLUSTER VARIATION METHOD IN THE COMPUTATIONAL MATERIALS SCIENCE 51
Z
a)
b)
Fig. 2 Sketch of the cubic mesh points within the sphere used for 3D discrete CVM
(a). Relative atomic displacements in the CD-CVM scheme in bcc lattice (b).
52 R. KIKUCHI A N D K. M A S U D A - J I N D O
R2z
a)
1.025 1 1.025
t a) b) o pair /
D qch tetrahedronI c) o/I
~
g 1.06 1.020
12-6 L.-J - - - MD O,~ 1.020
exp "" /
1.05 /B ]
1.015 0." 1.015
0 1.04 / zl~'l
",~ 1.010 1.010-
¢~ 1.03 i ~ s J ~
X 1.005 1.005 -
1.02
Cu Au
[-" 1.0 0.995 0.995 I I I I
0.05 0.1 0.15 I I I I
200 400 600 800 200 400 600 800
kT/~ --* Temperature(K) Temperature(K)
Fig. 4 Thermal lattice expansions calculated by CD scheme of CVM for the crystals with
12-6 type o f Lennard-Jones potential (a), Cu (b) and Au (c). For Cu (b) and Au (c) metals,
the electronic many body potentials are used.
CLUSTER VARIATION METHOD IN THE COMPUTATIONAL MATERIALS SCIENCE 53
8-4
.
O ~
0 0.2 0.4 0.6 0.8 1.0
X8
5.5 5.5
a)
5.4 5.4
t 5.3
~o
5.2 5.2
5.1 5.1
5 5
0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9
XB---~ X B --~
Fig. 6 Phase diagrams of size-different binary parameters presented in Table 2. The curves in
6(a) from bottom to top are (a), (b) and (c) in Table 2, while the curves in 6(b) from bottom to
top are (e), (d), (a) and (f) in Table 2, respectively.
54 R. KIKUCHI AND K. MASUDA-JINDO
7 7.
a) b)
6 6-
5 5
--~4 ?-.4
3 3
2 2
1 I
0 0
0,0 0'.2 ' 0'.4 ' 0'.6 ' 0'.8 ' 1.0 0.0 012 014 016 018 1.0
XB_..~ XB---~
Fig. 7 The order-disorder phase diagrams of B2 alloys. The upper and lower curves are calculated by
conventional CVM and CD scheme of the CVM, respectively.
O 600K
0.2
--fi3-- 1000K
, - - × - - 1400K
0.15
0.1
d:Z
0.05
<
-0.05
I
-0,1 . . . . I . . . . . . . . I . . . . I . . . . I . . . . I . . . .
0 1 2 3 4 5
Surface Layer