Gale and Davis: The Computer Calculation of Phase Boundaries from Thermochemical Data 25

Acknowledgement 11. J. W. Christian, Ref. 7, p. 721.

This work was supported by the Advanced Research 12. H. D. Mendelberg, M. Meixner, and K. Lucke, Acta Met.,
1965, 13, 835.
Projects Agency, through the Center for Materials Researc h 13. M. Cook and T. Ll. Richards, J. Inst. Metals, 1947, 73, 1.
at Stanford University, and by the Department of Navy, 14. H. G. Van Bueren, "Imperfections in Crystals ",2nd edn.,
through the U.S. Naval Ordnance Test Station, China Lake, p. 491. 1961: Amsterdam (N. Holland).
C r~ .
a 1 ornIa,
USA
.. .
15. R. D. McCammon and H. M. Rosenberg, Proc. Roy. Soc.,
, 1957, [A], 242, 203.
16. A. J. Kennedy, "International Conference on Fatigue of
Metals", p. 401. 1956: London (Inst. Mech. Eng.).
References 17. T. Broom andR. K. Ham, Proc. Roy. Soc., 1957, [A], 242,166.
1. C. Zener, "Elasticity and Anelasticity of Metals ", p. 61. 18. K. Lucke and K. Detert, Acta Met., 1957,5,628.
1948: Chicago (Univ. Press). 19. R. A. Vandermeer and P. Gordon, Trans. Met. Soc. A.I.M.E.,
2. F. Lokshin and O. Andreeva, "Metallovedenie i Term. 1959, 215, 577.
Obrabot. Mat. Konf." (Odessa, 1960), p. 233. 20. P. Gordon and R. A. Vandermeer, ibid., 1963, 224,917.
3. R. I. Garber, L. M. Polyakov, and G. N. Malik, Phys. Metals 21. G. F. Bolling and W. C. Winegard, Acta Met., 1958,6,283.
Metallography, 1960, 10, (4), 97. 22. G. F. Bolling and W. C. Winegard, ibid., p. 288.
4. B. Weiss, D. OelschHigel, H. Kuzmany, and H. Pinger, Z. 23. K. T. Aust and J. W. Rutter, "Recovery and Recrystallization
Metallkunde, 1965, 56, 312. of Metals ", p. 131. 1963: New York and London (Inter-
5. G. A. Hayes and J. C. Shyne, Trans. Inst. EEE, Sonics and science Publishers).
Ultrasonics, 1969, SU-16, 68. 24. K. T. Aust and J. W. Rutter, Trans. Met. Soc. A.I.M.E.,
6. M. Cook and T. Ll. Richards, J. Inst. Metals, 1940, 66, 1. 1959, 215, 820.
7. J. W. Christian, "The Theory of Transformations in Metals 25. J. W. Rutter and K. T. Aust, ibid., 1960, 218, 682.
and Alloys ", p. 21. 1965: Oxford, &c. (Pergamon Press). 26. K. T. Aust and J. W. Rutter, "Ultra-High-Purity Metals ",
8. J. T. Michalak and W. R. Hibbard, Jr., Trans. Met. Soc. p. 115. 1962: Cleveland, Ohio (Amer. Soc. Metals).
A.I.M.E., 1957, 209, 101. 27. K. Lucke and H. P. Stuwe, Ref. 23, p. 27.
9. P. Gordon, ibid., 1963, 227, 699. 28. E. S. Machlin, Trans. Met. Soc. A.I.M.E., 1962,224, 1153.
10. B. F. Decker and D. Harker, ibid., 1950, 188,887. 29. J. C. M. Li, Ref. 23, p. 160.
© The Metals and Metallurgy Trust. 1971.

The Computer Calculation of Phase

Boundaries from Thermochemical Data
B. Gale and J. M. Davis

Inaccuracies in determinations of miscibility gaps and solu-

bilities in the solid state, as well as solidus curves, are typical
Some numerical problems associated with the calcula- examples of the shortcomings of conventional methods for the
tion of phase diagrams from thermochemical data are
quantitative assessment of equilibrium diagrams.
discussed. Two simple cases are considered. The mis-
cibility gap in the Cr-Mo system is calculated using the A number of recent investigations by Kubaschewski and
free energies derived from calorimetric and vapour- his co-workersl-4 have demonstrated this point. The method
pressure data. The liquidus and solidus curves for the employed has been to calculate Gibbs energies of the various
Mg-Cd system were calculated in a similar way. phases constituting the system from assessed heats and
These did not agree very well with the directly deter- entropies of formation. These are then plotted on a Gibbs
mined curves and a method is described for refining energy-concentration diagram and lines are drawn that touch
the estimated solidus curve using the measured liquidus the curves at not less than two distinct points, provided they
curve. For both systems it is shown that unless care is do not intersect them anywhere else. These points are the
taken in representing the thermochemical data spurious
phase boundaries.
features appear in the phase diagram.
The graphical method of equilibrium-diagram evaluation
is convenient and simple for binary systems that are not too
complicated. However, it appears that the evaluation of
phase boundaries from thermodynamic data is so superior to
The value of thermochemical data for the evaluation of
the conventional methods that it should usefully be extended
metallurgical phase diagrams is becoming increasingly recog-
to more complicated binary, ternary, and multicomponent
nized. Phase boundaries are, of course, part of the more
systems. For these, the graphical method is quite inadequate
general thermodynamic properties of a system.
and computer methods must be used.
The advantage of the thermodynamic approach is that
A fundamental difficulty is that the bond structure of
enthalpies and free energies of phase formation can be deter-
metallic phases is so complex and so little understood that no
mined under conditions where equilibrium is easily achieved.
well-founded theoretical form for the free-energy function is
The results can then be used for the calculation of phase
known. Instead, it is necessary to resort to convenient mathe-
boundaries under conditions where chemical equilibration is
matical forms. The present paper describes an attempt to
hampered by various kinetic, mostly diffusional, factors.
reproduce experimental heats and entropies in a convenient
mathematical form and to apply the results to the calculation
Paper No. MS 206. Manuscript received 19 May 1970. B. Gale, Ph.D., of phase boundaries in binary metallic systems for two typical
B.Sc., and Mrs. J. M. Davis are in the Division of Inorganic and
Metallic Structure, National Physical Laboratory, Teddington, Middx. cases: (1) a solid miscibility gap Cr-Mo and (2) a solidus-
METAL SCIENCE JOURNAL 1971. Vol. 5
26 Gale and Davis: The Computer Calculation of Phase Boundaries from Thermochemical Data
liquidus gap Mg-Cd. Both have previously been evaluated The partial molar free energy of chromium was calculated
graphically by Kubaschewski and Chart.! In Mg-Cd the from the vapour-pressure measurements of activity made by
possibility of using data for the liquidus boundary in order to Lafitte and Kubaschewski.6 The coefficients b~(T') were
refine the solidus boundary is also discussed. then calculated using a least-squares analysis.
The free energy and enthalpy of the solid phase at a tem-
Thermochemical Reactions perature T' are given by
Let XA and XB denote the molar fractions of elements A and
B in a binary solution, XA -~ XB = 1; it is convenient to put GS(x,T') = xAGi(T') + xBG;(T') + ~GS(x,T')
x = XB = 1 - XA. TA and TB, with TA ~ TB, are the melting ... (7)
points. Brackets <>
and { } are used to denote the solid and
... (8)
liquid phases of the alloy enclosed by them, e.g. {Ax A BXB}
is the liquid solution containing molar fractions XA and XB.
The molar Gibbs free energies of A and B in the solid state at It should be noted that it is assumed that HS(x) is independ-
temperature T are denoted by GAS(T) and GBS(T), and in the ent of temperature. This was done since the data used in this
work did not suggest otherwise. It would not, of course, be
liquid state by GAL(T) and GBL(T). A similar notation is
used for the enthalpies and entropies. For Cr-Mo A
B= Mo; for Mg-Cd A· Mg, B= Cd. ..'
Cr, = difficult to include temperature-dependence if required.
At any other temperature T the free energy is

Cr-Mo GS(x,T) = HS(x)(1 - TIT') + TIT'GS(x,T')

The solid solution {AXABxB} can be formed at T' ~ TB by
the reaction = (1 - x)Gl(T) + xG%(T)
N

The molar free energy ~GS(x, T') of this reaction was calcu-
... (1)
+ x(1 - x) L
u=o
g~(T)(1 - 2x)u

lated from calorimetric and vapour-pressure measurements. + RT{(1 - x) In (1 - x) + x In x} ... (9)

By definition ~HS(x) is zero when x = 0 or 1, so that for the
whole range of composition it can be conveniently repre- where
sented by the function
... (10)
I
N

~HS(x) = x(1 - x) h~(1 - 2x)u ... (2)

u=o Mg-Cd
For this system the molar free energy of reaction (1) was
The coefficients h~ were calculated from the calorimetric obtained from emf measurements.7 If E is the emf, Z is the
measurements of Kubaschewski and Chart,! using a least- valence of the A ions in the cell, and F is the Faradayconstant,.
squares technique (see· Appendix A) and they are assumed then ~Gl(x,T') = -ZFE and ~Sl(x) = ZF(dEldT') From.
to be independent of T. For an ideal binary solution it is (2) and (4) it follows that the entropy for the reaction
known that ~G(x, T') is given by (1) can be written in the form
~GS(x,T') = RT' {(l - x) In (1 - x) + x In x} ... (3)
L
N

~SS(x) = x(1 - x) S~(1 - 2x)U

The deviation from the ideal case can be represented by a u=o
polynomial in x, and using Guggenheim's form5 ~GS(x, T')
can be written as
- R {(I - x) In (1 - x) + x In x} ... (11)

N The Ss coefficients are related to those that appear i~ the-

~GS(x,T') == x(1 - x) I
u=o
g~(T')(1 - 2x)u
correspo~ding exp~ession for the partial entropy ~Sl(x)
and component A, I.e.
+ RT' {l - x) In (1 - x) + x In x} ... (4)
l)j+u
L
N (-'-
When all the g~(T') coefficients are zero except g~(T'), S~ = j+l d~ ... (12)
equation (4) represents the formation of a regular solution. j =u

When they are not zero, (4) represents the more general non- where
regular solution. It can easily be shown that the g~(T') are
related to those that appear in the corresponding expression N
for the partial free energy ~G1 (x, T') of component A, Le. ~Si(x) = x2 I
u=o
d~(1 - 2x)u - R In (1 - x) ... (13)
N

g~ (T') = Ie~+
j=u ]
l)j+u b~(T')
1
... (5) The coefficients b~(T') in (6) and d~ in (13) were calculated,
using a least-squares analysis from data given by Trumbore,
where Wallace, and Craig7, and the entropy and enthalpy by
N

~Gl(x,T') = x2 I b~(T')(1 - 2x)u + RT' In (1 - x) SS(x) = XAS~ + XBSZ + ~SS(x) ••• (14)

u=o .•• (6) HS(x) = G(x,T') + T'SS(x) ... (15)

 Gale and Davis: The Computer Calculation of Phase Boundaries from Thermochemical Data 27
From (4), (11), and (12), the coefficients h~ of the expansion N
of t"HS(x), given in (2), are t"G~(X,TA) = x2} b~(TA)(1 - 2x)U + RTA In (1 - x)
;=0 ... (25)
h~ = g~(T') + T'S~ ... (16)
the coefficients b~(T A) being obtained by a least-squares
Hence, using (8), (11), and (12), the free energy of the solid analysis. From (23) the partial enthalpy of A is
phase at any temperature Tis
N

t"GS(x,T) = x(1 - x) L
N

(h~ - TS~)(1 - 2x)u

t"Ni(x) = x2[h~ + L {u(l - 2x)u-1
u=1
u=o + (u + 1)(1 - 2x)U}h~] ... (26)
+ (1 - x)(H~ - TSi) x(H; - TS;) +
+ RT {(1 - x) In (1 - x) + x In x) } ... (17) Then from (25) and (26) the partial entropy of A is

The enthalpy HL(X) of the liquid phase can be obtained -L 1-r.

t,,8.ACx)= TA [t"H.lx) - t"r;~(X,TA)]
from the enthalpy HS(x) of the solid phase and the latent heat
of fusion L(x) is
= x2 L N
d;(l- 2x)u - Rln(1- x) ... (27)
HL(X) = HS(x) + L(x) ... (18) u=1

Experimental values of L(x) for the Mg-Cd system are taken where
from the paper by Hirabayashi. 8 Since L(O) = L A and
L(1) = LB, the latent heats of fusion of the elements, a suit-
able form to represent L(x) is d~ = ;A {(u + l)(ht + h~+l) - b~ (TA)} for u=O,l ... N-l
1
N = TA {(N + l)h; - b~(TA)} for u =N ... (28)
L(x) = xALA + xBLB + x(1 - x) L lu(1 - 2x)u ... (19)
u=o
The integral entropy for reaction (21) is then given by
The coefficients lu are obtained by a least-squares analysis.
1
I
XA
Using (15), (18), and (19), HL(X) can be written in the form t"SL(X) = XA 0 (1 _ y)2 t"S~(y)dy ... (29)

HL(X) = x(1 - x) L h~(1 - 2x)U + xAHj + xBH~ Hence, using (27)

u=o ... (20) N

where
t,,8L(x) = x(l - x)
u=o
LS~(l - 2x)u

... (21) - R {(1 - x) In (1 - x) + x In x} ... (30)

The liquid solution can be formed at T ~ TA by the reaction where

N
XA {A} + XB {B} = {AXABxB} ... (22) = -.!...-
U
SL
u TA
{hL _
u ?
""
]=U
(-
j +l)i+1 bL(T)}
u A ...
(31)

From (20) the enthalpy t"HL(X) for this reaction is

N
The free energy of the liquid phase can be written in the same
t"HL(X) = x(1 - x) L h~(1 - 2x)U ... (23)
form (9) as the solid phase, Le.
u=o
GL(x,T) = (1 - x)G;(T) + xGi(T)
N
The partial free energies and entropies of solid magnesium in
liquid alloys have been measured by Terpilowski and Barycka 9
+ x(1 - x) L g~(T)(1 - 2x)U

and by Eremenko and Lukashenko.10 These quantities cor- u=o

+ RT {(I - x) In (1 - x) + x In x} ... (32)
respond to the reaction
... (24) where

At the melting point T A the partial free energies of A for

G~(T) = Gi(T) + LA - TcrA ... (33)
reactions (22) and (24) are equal. The data for reaction (24)
were obtained for a number of different temperatures between
Gi(T) = G~(T) + LB - TcrB ... (34)

TA and TB from which the partial free energy t"GL(x,TA) for

reaction (22) could be readily calculated. As in (5) this partial cr A and cr B are the entropies of fusion of pure A and pure B
free energy can be written in the form respectively, the values of which are taken from HirabayashL 8
 28 Gale and Davis: The Computer Calculation of Phase Boundaries from Thermochen1ical Data
The Phase Diagrams
It is well known that a necessary condition for the chemical 0= ------
equilibrium of two phases in these simple cases is that the 1=-'-'-'
2=--
chemical potential of components A and B must be equal in 3~ ---
4=-"-"-
the two phases a and f3. These conditions determine the
compositions ytx and yf3 of the two phases and they are ob- 1200
tained by solving the equations

dGtx( tx T)= dGf3( f3 T) = Gtx(ytx,T) - Gf3(yf3, T) ... (35)

dytx y ,. dyf3 Y , ytx - yf3
1100

Using (9), (32), (33), and (34) it can be shown that for the
solidus-liquidus boundary (35) can be written as

U(ytx,yf3,T) = 2 g~(T){ytx - y(3)[(1 - 2ytx)2 - 2uytx(I- ytx)]

u=o
- ytx(I - ytx)(1 - 2ytx) }(I - 2ytx)u-1

+ yf3(1 - y(3) 2 N
g~(T)(1 - 2y(3)u 900

u=o

+ RT f3
[ y ln
yf3)
( ytx + (1 . (1 -
y(3)]
- y(3) In 1 _ ytx

+(1 - y(3)(LA - TaA) + yf3(LB - TaB) =0 ... (36) 800

OCr 0'25 0'50 0'75 1·00
Mo
Composition (x)

.2:
N
Fig. 1 Immiscibility boundaries 0/ Cr-Mo using different analytical
V(ytx,yf3,T) = g~(T){(ytx - y(3)[l - 2y(3)2 approximations (N = 0, 1,2, 3,4).
u=o
- 2uyf3(1 - y(3)]
+ yf3(1 - y(3)(l - 2y(3)}(1 - 2y(3)u-l
N

- ytx(1 - ytx) 2 g~(T)(1 - 2ytx)U

400

u=o

+ RT[Y« In (;:) + (l - r) InG = ;:)]

+ (1 - ytx)(LA - TaA) + yrJ.(LB - TaB) = 0 ... (37) 200

(a=S, f3=L)

Cr
The same equations pertain to the immiscibility gap if the
latent-heat terms are removed and a = f3 = S. o
The pair of non-linear equations (36) and (37) can be "0
solved numerically for as many values of T as are required to E

describe the phase boundaries adequately. The numerical "3

technique used is outlined in the Appendix.
f>-
-50

The Cr-Mo Diagram

The immiscibility curves, calculated for N = 0, 1, 2, 3, are
--.
<5
E
shown in Fig. 1. They are, apart from curve 4, similar for 2=--
3 = ---
T < 1000 K but for higher temperatures there are large -100 4= -- ---
-400
discrepancies. The dotted lines (4) represent possible immis-
cibility boundaries associated with new re-entrant parts of
the free-energy curves which appear when N is increased from
3 to 4. This behaviour is particularly evident near the tem-
perature T = 950 K (Fig. 2). Curves for N = 2, 3 are similar -150
-600
but when N = 4 two new re-entrants appear which are so
arranged that there are 3 significant tangent lines: PIQl, Fig. 2 Calculated/ree energies 0/ Cr-Mo alloys at 950 C (1323 K)
0

using different analytical approxtmations (N = 2, 3, 4).

P2Q2, and P 3Q 3· The horizontal dotted lines in Fig. 1 cor-
respond to temperatures at which the pairs of tangents PIQl,
P2Q2 and P2Q2, P 3Q3 are coincident. The anomalous be- by the errors in the original data. The optimum value of N,
haviour when N = 4 is almost certainly spurious and is a as determined by 'Forsythe's criterion, is 2 and the corres-
consequence of using a curve of higher order than is justified ponding immiscibility curve is shown as a full line in Fig. 1.
 Gale and Davis: The Computer Calculation of Phase Boundaries from Thermochemical Data 29
1000 1000

--- Points calculat¢d by fixing

--- Points calculated using

3 points on th¢ liquidus (N = 2)
shown by 0
equations (36 ') & ( 37) N: 2
M¢on ~xp¢rim¢ntally d¢t¢rmin¢d
900
" \" "
\
\ '\
Mean experimentally
boundaries
determinlEd 900 \~~
\ ~.~
\ o
boundari¢s

" .~. Giv¢n points on th¢ liquidus

T~mp.,
\\ \
\
K \
\
\
\
~\.
\
, .~
\ ~.
BOO
""
" ""
"" " " " ,
,, ",
"l "

" " ~
~ """~ 700
"" '\

'- " '- ",,"

:-...
~

600 600 I-

500!
I
Mg X- Cd
I sJ Mg X Cd
!

Fig. 3 Calculated liquidus-solidus curves for Mg-Cd.

Fig. 4 Liquidus and solidus curves for Mg-Cd calculated using
3 given points on the liquidus.

The Mg-Cd Diagram

The optimum values of N for the different data sets are not
necessarily equal, e.g. for (5) N = 2 and for (12) N = 4. For
convenience in the numerical calculations equal values of N
for different sets of data are used. If, for a particular data set,
1000
N is chosen to be larger than the optimum value then false
detail will, as shown in the Cr-Mo system, appear in the least-
Points CQlculQt~d by fixing
squares representation. It is therefore reasonable to choose
4 points on th¢ liquidus (N=3)
N to be the least of the optimum values and for this system it shown by 0
was 2. M~an czxp~rim¢ntally d~t¢rmln¢d
The points calculated using the method described in the boundariu

Appendix are shown in Fig. 3. The dotted lines represent o Giv¢n points on thcz liquidus
mean boundaries estimated using measurements made by
Hume-Rothery and Raynor,l1 Bruni and Sandonnini,12
Grube,13 and Hirabayashi. 8 A comparative plot of all the
BOO
experimentally determined boundary points has been given
by Kubaschewski and Chart.1 It is evident in Fig. 3 that the
boundaries calculated from the thermochemical reaction data
do not agree closely with the measured ones.
The phase boundary, the emf, and the latent-heat measure- I
700 t-
ments should form a self-consistent set of thermochemical
data for the system. It would therefore seem to be preferable
to select the most accurate data, which are sufficient to define
the system, and to use this set to derive other thermodynamic
quantities. The relations between the reaction and the
boundary data can be most conveniently described using 600

relations (33) and (34).

The 2M relations (36) and (37) which hold for temperatures
Tt, T2 ... TM can be regarded as equations from which any
2M of the 2M + 4(N + 1) parameters yf.... y~;
yf .... yi; h~ .... h1v; S~ ••.• S;; h~ .... h;; 500 '-------------.--.-.---.-- ~ ..J
S~ •••• sj may be determined as roots. Since the kinetics
Mg " Cd

in the liquid phase is much faster than in the solid phase it is Fig. 5 Liquidus and solidus curves for Mg-Cd calculated using 4
likely that the measured liquidus curve is more reliable than given points on the liquidus.
30 Gale and Davis: The Computer Calculation of Phase Boundaries from Thermochemical Data
the solidus curve. It is then possible to fix M values yf .... where the P u(x) are polynomials defined by the orthogonality
yf; on the liquidus curve and to use the 2M equations to relations
obtain M points on the solidus curve y~ ... y 1 and a set
of adjusted entropy coefficients sg ... S.:.
These particular
coefficients are chosen here since it is considered that they are
likely to be most in error. Also, since all the coefficients are
in error it is necessary to put M = N + 1. It is interesting to
note that it would have been possible to use the fusion entropy
LQ

i=:1
B2(Xi)PfJ.(Xi)Pu(Xi) = 0 ... (40)

measurements of Hirabayashi 8 to calculate the liquid-phase

entropy from the solid-phase entropy. If that had been done, (Q being the number of points) then, using the least-squares
then any errors in the S~ would also occur in the S~. criterion, the Yu are given by
Evidently many other possible calculations can be con-
sidered, e.g. Q
(a) Determine M values y~
h~, h~, S~, S~ (u = 0 N):
Y1 and 4(N 1) values
here M would have
+
Yu =
L
_i=_I
AiB(Xi)Pu(Xi)
_ ... (41)
to equal 4(N + 1) and there would be 8(N +
1) equa- Q
tions. ') B2(Xi)P~(Xi)
(b) Both phase boundaries would be given and all the ~
4(N + 1) hand S parameters could be calculated: here
M would have to equal 2(N +1). Apart from the introduction of the weighting function
Some numerical details of the calculation of the solidus B(x), this procedure is the same as that described by Clen-
boundary and the S~ parameters are given in the Appendix. shaw.l4 The functions Pu(x) can be generated by recurrence
Fig. 4 shows an " adjusted" phase diagram calculated as relations similar to those given by Forsythe,15 which may also
follows: be used to obtain the Cu from the Yu.
(1) Using N = 2, the optimum value determined by the The decision as to which is the order of the polynomial
Forsythe statistical test, 3 points on the measured liquidus, at giving the best fit can usually be decided from the behaviour
T = 650, 750, and 850 K, were used to calculate new values of the mean square of the residuals a. Forsythe15 has sug-
of the entropy parameters sg, Sf, s~, and 3 points on the gested that the optimum order is that for which a ceases to
solidus curve. decrease significantly. This value is usually well defined and
(2) The modified S~ values were used with the original is used for all the data representations in this paper.
values of the other thermochemical reaction parameters
h~, S~, and h~ to calculate a sufficient number of other
points on the phase boundaries. B. Calculation of the Phase Boundaries
The new calculated boundaries agree closely with the esti- The phase-boundary points (yrx,T) and yC3,T) are roots of
mated experimental ones. equations (36) and (37). However, not all the roots corres-
If N is greater than the optimum value and the above pond to thermochemical equilibrium, for a further condition
operations (1) and (2) are performed then new spurious feat- must be satisfied: the common tangent line, defined by a pair
ures occur (see Fig. 5). Here the case N = 3 is illustrated; of roots, must not intersect either of the free-energy curves
N = 4 is similar. Such behaviour is to be expected in these of the phases. This latter condition is used in th'e initial search
calculations where, if the original dat a are overdetermined, for roots before calculating them accurately by iteration.
additional extremum and inflection po ints will be introduced, A search range (ymin, ymax) and a resolution ~y are speci-
so that, when U = V = 0 are solved, spurious roots (common fied. ~y determines the least significant root separation
tangents) appear. This behaviour is like that noted in the required. If there are two free-energy functions, as in the
Cr-Mo calculations and points to the need for considerable Mg-Cd system, then it is necessary to test whether or not
care in stat istically representing the data. GC3(Ymin,T) < Grx(Ymin,T). If the inequality does not hold,
The methods discussed here can be readily adapted to other, then U is replaced by V in the following discussion. For the
more complex, bi nary thermochemical systems, provided that immiscibility gap this test is not required. The function
a sufficient set of data is known, and the authors would be =
U(yrx,yC3,T) U(m,n) is calculated for yrx = Ymin (m - t) +
prepared to analyse systems of special interest. ~y; yC3 = Ymin + n~y, where m = 0, 1, 2 ... , n = m 1, +
m + 2, and yrx < Ymax - t~y; yC3 < Ymax t~y. If +
ApPENDIX sgnU(ml, nl) #- sgnU(ml, nl +
1) and sgnU(ml, n2 - 1)
A. The Least-Squares Calculations #- sgn U(m!, nz) (Le. U has two sign changes for a particular
value of m), and if the number of sign changes is 0 or 1 for
Inspection of relations (5), (12), (19), and (25) shows that
m = ml - 1, then a pair of roots yS,yL corresponding to a
the functions whose parameters are determined by least
non-intersecting common tangent exists in the range
sauares have the form

A(x) = B(x)C(x) ... (38) Ymin + (ml - :/2)~y < yrx < Ymin + (ml - t)~y} ..• (42)
Ymin + (nl - 2)~y < yC3 < Ymin + nz - t)~y

where B(x) = x2 or x(1 - x) and C(x) = LN

Cu(1 - 2x)U Otherwise there are either no roots or I y" - yf3/ < ~y.
u=o If a root pair is detected their values are refined using an
If C(x) is written in the form iterative steepest-descents algorithm which is outlined below.
When the roots are calculated to the required accuracy a new
N
+
C(x) = L
u=o
YuPu(x) ... (39)
search-range minimum is defined by Ymin = yC3 ~y and
the whole process repeated until the original search range is
exhausted.
 Gale and Davis: The Computer Calculation of Phase Boundaries from Thermochemical Data 31
The steepest-descent calculation is based on the algorithms o ~!X/ox tJ., but the evaluation of the reciprocal Hessian requires
described by Davidon16 and Fletcher and Powell.I7 Let the not only rather complex analytical expressions for the second
non-linear equations be written formally as derivatives of the ~r but also the time-consuming inversion
process. Following Fletcher and Powell,16 Q is calculated
~r(Xl ••• xP) = 0 (r = 1,2 ... P) ... (43) iteratively using the relation

where

(1) For the common tangent calculation

where

(2) For the fixed liquidus calculation

All the calculations described above were programmed in
P = 2(N + l)xu = yS(Tu); XN+u+1 = S~-l ALGOL for a K.D.F.9 computer.

~r = U(Xr,XN+2 ••• X2N+2) (~= 1,2 ... N + 1) Acknowledgement

The authors wish to thank Dr. O. Kubaschewski for sug-
~N+l+r = V(Xr,XN+2 ••• X2N+2) gesting the problem and for giving advice and encouragement.

Define a function cp by References

1. O. Kubaschewski and T. G. Chart, J. Inst. Metals, 1964-65,93,
P 329.
cp (x 1 .... x P) _- ~
~ I 2( x 1 ...
9r X P) ... (44) 2. L. Elford, F. Muller, and O. Kubaschewski, Ber. Bunsenges.,
1969, 73, 601.
r=l 3. F. H. Hayes and O. Kubaschewski, J. Inst. Metals, 1969, 97,
381.
so that <p is zero and a minimum when ~r = o. 4. F. Muller and o. Kubaschewski, High Temp.-High Press.,
Let x(~) denote the initial values of the roots. For the 1970,2.
5. E. A. Guggenheim, "Thermodynamics", 5th edn. 1967:
tangent calculation they are defined by the centroid of the Amsterdam (North Holland).
range (42). For the liquidus calculation x(~) = yS(Tu) as 6. M. Lafitte and O. Kubaschewski, Trans. Faraday Soc., 1961,
calculated from the common tangent and the original reaction 57,932. .
7. F. A. Trumbore, W. A. Wallace, and R. S. Craig, J. Amer. Chem.
data; and the x~tU+1 values are the original solid-phase Soc., 1952, 74, 132.
entropy parameters. The function cp decreases and, in general, 8. M. Hirabayashi, J. Japan Inst. Metals, 1952, 16, (6), 295.
9. J. Terpilowski and 1. Barycka, Bull. A cad. Polon. Sci., 1961,
has a minimum at x(~) = x(~) + tnu, where, using the 9,175.
summation convention 10. V. N. Eremenko and G. M. Lukashenko, Ukrain. Khim. Zhur.,
1963, 29, 1048.
11. W. Hume-Rothery and G. V. Raynor, Proc. Roy. Soc., 1940,
[A], 174,47.
12. G. Bruni and C. Sandonnini, Z. anorg. Chem., 1912,78,277.
13. G. Grube, ibid., 1906, 49, 72.
The value of t corresponding to a minimum is obtained by 14. C. W. Clenshaw, Computer J., 1960, 2, 170.
15. G. E. Forsythe, J. Soc. Indust. Appl. Math., 1957, 5, 74.
cubic interpolation. 16. W. C. Davidon, Atomic Energy Commission Rand D Rep.
The matrix Q is the reciprocal of the Hessian of cp. The gtJ. (ANL-5990), 1959.
can be readily calculated using analytical expressions for 17. R. Fletcher and M. N. D. Powell, Computer J., 1963, 6, 163.

© The Metals and Metallurgy Trust. 1971.