Calphad,

Pergamon
PII: SO364-5916(00)00011-0

Vol. 24, No. 1, pp. l-14,2000 0 2000 Elsevier Science Ltd All rights reserved 0364-5916/00/$ - see front matter

THERMODYNAMIC

ASSESSMENT OF THE BI-ZN SYSTEM

Dmitri V. Malakhov Mc&srer

universi[y, Department of Materials Science & Engineering, 1280 Main Street West, Hamilton, Canada L8S 4MI

ABSTRACT The thermodynamic

properties and phase diagram of the system Bi-Zn were optimized using Lukas program BINGSS. Special efforts (including the pm-optimization treatment of data) were employed to reproduce the known flat, asymmetric, and convex shape of the miscibility gap. A comparison of calculated values with their experimental counterparts is given. The basic statistical characteristics of the solution obtained are presented. Previous assessments of the system are analyzed and compared with the present endeavor.
INTRODUCTION

Eutectic and near eutectic Pb-Sn alloys are in common industrial use today, but it seems likely they will be gradually replaced with Pb-free solders. The first decisive step was taken in 1990 when the US Congress issued a bill proposing the barming of Pb from some materials, including solders for electronics. That legislative act stimulated much fruitful activity aimed at the development of new Pb-free alloys [95 Gla]. Ternary, quatemary and higher order systems containing Zn, Sb, Bi, Ag, Cu, Sn have been examined, but there remain many systems deserving of furtber investigation. The so-called CALPHAD approach [98 Hil, 98 Saul can be successfully employed to reduce the time required for such investigations. The validity of conclusions of the CALPHAD technique clearly depends on the reliability and accuracy of the input information and the robustness and intelligence of the optimization program. Since the majority of multicomponent systems where promising Pb-free phases are expected to be found contain Bi and Zn, a trustworthy thermodynamic model of this binary system is important. Previous attempts [57 Ole, 77 Bal, 85 Gir, 93 Wan] to build such a model cannot be considered completely successful. In view of this fact it was decided to repeat the optimization, taking into account both previously overlooked and recently reported data, and using accepted lattice stabilities [91 Din]. The system Bi-Zn seems to be quite simple. It has only three condensed phases: liquid, Bi-based rhombohedral and Zn-based hexagonal solid solutions. Within a certain range of temperatures and concentrations, the liquid phase separates forming Bi-rich (Lt) and Zn-rich (Lz) layers. The miscibility gap is remarkably flat and asymmetric, and it could perhaps be imticipated that this peculiar shape would engender particular difftculties during optimization. Okamoto [97 Oka] hypothesized that the strong asymmetry could be explained if the observed miscibility gap was considered as a part of a two-peak miscibility gap [93 Oka]. This opinion is not shared by the author, since it has not been shown to be consistent with the array of experimental information accumulated to date.
DATA SURVEY

Although assessment results are determined by many factors, the choice of thermodynamic models for phases and the assignment of statistical weights to experimental measurements are deemed to be the most important [92 Luk]. Original version received on 23 July 1999,Revised version on 7 January 2000

D. V. MALAKHOV

The assignment of weights is a human-dependent procedure that can hardly be unified and fulfilled in an impartial manner. However, the input information upon which the optimization procedure is based must at least be explicitly and unambiguously reported.
THERMODYNAMICPROPERTIES The gap of miscibility in the system Bi-Zn is accompanied

by a gap of another kind that is between availability of thermodynamic data on the liquid phase and solid phases. While the properties of liquid have been extensively investigated, there is no reliable information about thermodynamic characteristics of the solid solutions. CHEMICALPOTENTIAL ZNINLIQUID OF PHASE The composition and temperature dependencies of the chemical potential of zinc have been scrupulously examined by means of the EMF method. A list of pertinent papers is given in Table 1 (x in this table and everywhere below denominates the mole fraction of zinc, x, .) TABLE 1. Chemical Potential of Zn in Liquid Phase: Data Sources, Investigated Regions and Errors

Since A&, = - 2FE, the absolute random error in the chemical potential caused by mistakes in voltage measurements is 2F x A,,,E. In the present assessment, it was decided to use a more realistic quantity IA,&,,1 A,,u, x

max 2F x A,E, (

) as the estimation

of experimental error.

The zinc vapor pressure was measured by means of Knudsen effusion method in the vicinity of 625K for three samples withx = 0.056, 0.100 and 0.149 [61 Yok]. Chemical potentials the assumption that the gaseous phase contains only monomers. & = RTln(P,,/P~,) were calculated under PL, (T) was taken from

The dependence

IVTANTHERMO database. A relative accuracy of 3.24% of pressure measurements was reported in [61 Yok]; errors of temperature and compositions are supposed to be equal to 2K and 0.001. ENTHALPY MIXING OF The enthalpy of mixing was measured by Kawakami [27 Kaw] and Wittig et al. [58 Wit]. The assumed absolute and section, relative errors and the regions investigated are given in Table 2. As in the previous max(A,,H,lA,,HLIxA,~,H) re Presents the experimental error.

Enthalpies as measured by Kawakami are significantly less endothermic than those determined by Wittig et al. Preference should be given to [58 Wit], because in many cases it was found out that Kawakamis data which were ob-

THERMODYNAMIC

ASSESSMENT:

Bi-Zn SYSTEM

tained in the 1920s for numerous binary systems, had a systematic bias. Comparing quantities obtained at 470C and 570C, one can hardly detect a temperature dependence of A,,,,HL. This circumstance is important when a model of the liquid phase is considered. TABLE 2. Enthalpy of Mixing: Data Sources, Investigated Regions and Errors Composition range 0.166-0.806 0.0982-0.9929 0.1531-0.9482
PHASE DIAGRAM

Temperature C 450 470 570 K 723 743 843 AxL devends on xL 0.0005

Errors AT 2 1 A,H, J A,Ji 100 0.05 10 0.01

Data source [27 Kaw] [58 Wit]

The first paper where the phase diagram was examined appeared in 1861 (ref. 1 in [58 Han]) while the most recent work is dated 1996 [96 Bou, 96 Kha], i.e., the nature of phase equilibria in the system Bi-Zn has been the subject of experimental investigations for many decades. Table 3 lists the sources used for optimization and specifies the assumed accuracy of measurements. TABLE 3. Monovariant Equilibrium: Data Sources and Errors

* Chemical analysis of separated layers ** EMF measurements LIQUI&SOLIDEQUUBRIA Equilibria between the liquid phase and solid solutions were carefully explored in [1892 Hey] and [1897 Hey] by means of cooling curves. An averaging of melting points of bismuth and zinc reported in [1892 Hey, 1897 Hey] gives 266.4C and 418.9C respectively The modem recommended values are 271.4C and 419.53C [91 Din]. For this reason, the original authors estimations of AT were enlarged, especially for the Bi-rich alloys. The location of a point on the phase boundary Lt/Lt+HCP found in [55 Sei] is in an excellent agreement with [1897 Hey]. LIQUID-LIQUID EQUILIBRIUM The first quantitative results on the miscibility gap were obtained by Spring and Romanoff who employed a chemical analysis of separated layers [1897 Spr]. It was found that the phase boundary had a normal shape, with a maximum located above 8OOC. This conclusion was confirmed by Hass and Jellinek who arrived at almost the same location of the upper critical temperature T, using a similar experimental technique [33 Has]. In 1955, that view was challenged by Kleppa who sagaciously noticed that the earlier procedures did,,not provide full opportunity for the system to reach equilibrium and decreased the critical temperature by more than 200 degrees, in accordance with his precise EMiF measurements [52 Kle]. Very soon the correctness of Kleppas inference met with further experimental corroboration [55 Sei], but five years later it was caIled in question again, when Lantratov and Tsarenko observed a constan: activity of Zn in the melt in the range 0.883-0.879 at 700C [60 Lan]. The phase boundary Lr/Lt+Ll found in [62 Koz] by means of EMF technique is located below that reported by Kleppa and the difference exceeds 20C at xL = 0.8. Okajima and Sakao using both EMF technique and chemical analysis of separated layers reported the miscibility gap having several regions of concavity with T,,, k 625C and x,, o 0.85 [80 Oka]. The situation changed again, when the values T,,, = 575C and x,, = 0.815 were found by means of a high precision

D. V. MALAKHOV

acoustic technique [88 Gla]. New evidence that the critical temperature is located significantly below 6OOC was obtained by Kanai et al., who measured sound velocity in the bismuth-zinc melt and found r,, = 580C and
Xcl,, 0.822 [93 Kan]. T,, = 573C at x = 0.83 was reported by Khairulin and Stankus as the result of their study =

of the two-phase region by the y attenuation technique 196 Kha]. After the array of all available data on miscibility gap had been carefully reviewed and interrogated, it became evident that a realistic model of liquid must provide a critical temperature not exceeding 580C and must also guarantee that the miscibility gap is convex.
NONVARIANT REACTIONS

The characteristics of nonvariant transformations have been reported in many papers (see, in particular, the references in [58 Han]). They will not be discussed here, because they have been already examined in depth by the editors of [90 Mas] whose qualifications are beyond question. Thus, it is believed that a monotectic transformation occurs at T = 416C with mole fractions of zinc equal to 0.37 in the Bi-rich melt and xzn = 0.993 in the Zn-rich liquid. It is also considered established that the eutectic temperature is 254.5C and the eutectic composition is xL = 0.081. These quantities are treated as reference values and the optimization procedure will be required to reproduce them with a high accuracy by means of artificially enlarged statistical weights.
SOLID STATESOLUBILITY

Using the diffusion couple technique, Ohnuma recently found the following solubility of zinc in bismuth: 1.24 at. % at 200C and 2.39 at. % at 250C [98 Ohn], which is significantly less than the values accepted by [90 Mas]. The low temperatures at which the experiments were conducted suggest that the accuracy of data obtained should not be overestimated. Probably, the best that can be said is that the solubility of zinc in bismuth is of the order of 2 atomic percent in the region from 200C to eutectic temperature.
PREVIOUS ASSESSMENTS

Four, partly successful, previous attempts have been undertaken to develop a thermodynamic model of the system Bi-Zn. To elucidate why it was decided to m-optimize the system, let us review the results obtained by our predecessors.
FITTINGBY OLEARI, FIORANIANDVALENTI Oleari et al. [57 Ole] fitted experimentalquantities[52 Kle, 55 Sei], and derived the following expression describing

the excess Gibbs energy of the liquid Bi-Zn solution: AGL 41s4=3.8+7685-5.142Q -(3.8x(1-x)(7685-5.1422-)) 5.6RT((3.8):(1-X)+X) Although this formula provides a convex phase boundary LJLi+Lr, high.
THERMODYNAMIC MODELOF LIQUID PHASE BALE, PELTONANDBIGAUD BY

x(1-x) 3.8(1-x)+x

-7.6+2250-2.576T)

(I-) 7.6(1-x)+x

the critical temperature

T,, = 600C is too

Bale, Pelton, and Rigaud measured the chemical potential of zinc in liquid melts Zn-Cd-Bi-Sn using the EMF technique and fitted the values obtained by a polynomial model [77 Ball. Their Margules-type expression provides a good agreement between experimental and calculated thermodynamic properties of liquid phase, although their calculated Am,HL is slightly more endothermic. The phase diagram (reconstructed with the lattice stabilities used in [77 Pel] and [SO Lin]) presented in Fig.1 has two dubious features. First, the top of miscibility gap is located at the

THERMODYNAMIC

ASSESSMENT:

Bi-Zn SYSTEM

very high temperature exceeding 8OOC. Second, the phase boundary LJLl+Lz has a region of concavity, which is located between the two solid circles in Fig. 1. Its exact extent was established by solving an equation d*T/&* = 0 with the partial derivative taken along the phase boundary. Although this type behavior of a monovariant line is not thermodynamically prohibited, for this particular case it is unnatural. It can be supposed that the actual origin of the concavity is not the unusual properties of the melt, but that it is caused by data treatment. In other words, the true geometrical structure of miscibility gap was corrupted by approximation phantoms [95 Mall.
OPTIMIZATION

BYGIRARD

Girard optimized all data available by 1985 using the Redlich-Kister expression for the excess Gibbs energy of liquid phase. It was assumed that bismuth dissolves in zinc. The excess Gibbs energy of the HCP solution was then evaluated, but the extent of the Bi-based solid solution was neglected. Girards optimization provides an excellent agreement between calculated and experimental thermodynamic quantities. The phase diagram reconstructed with lattice stabilities taken from [85 Girl (they almost coincide with [91 Din] except for some slight differences) is presented in Fig. 2. Although both the critical temperature and the shape of miscibility gap are quite acceptable, the model employed by Girard cannot be considered completely satisfactory. First, T,, = 615C remains too high. Second, a concavity region still exists on the phase boundary Li/Li+L2. Third, the solid solubility of Zn in Bi was not taken into account.
805.07% I1081.22K0.8325 I988.26K - 1000 1100

650
615loC I888.25K - 900

800
0.7135

600 550

700

6B3.lVC

600 9 /:? 500 I-400

concavity region

-900

-800 _-I
0.4176 433.WC x ,707.UK 413.410CIB56.56K 0.37862 0.95615 ..ll.WC I684.74K

700 -600
-

600
257.18OC I530.33K 0.99892 - 550

300
251.72% 0.072173 I529.87K

500

*50- o.074768
I,,,

200 0.0 0.1 0.2 0.3 0.4 0.5 0.8 0.7 0.8 0.9 1.0

--.a
FIG. 1. Phase diagram calculated with the model by Bale, Pelton and Rigaud

- 500 2001 ,,, I ,I 0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0

FIG. 2. Phase diagram calczated

with Girards model

THERMODYNAMIC MODEL OF LIQUID PHASEBY WANG ET AL.

Wang et al. [93 Wan] optimized experimental data obtained in [55 Sei, 80 Oka, 58 Wit, 52 Kle, 81 Oka, 66 Pta, 59 Kle, 65 Chi] by the least square technique and found an optimal set of parameters in the Redlich-Kister formalism, chosen for the analytical description of the Gibbs energy of the liquid phase. In many respects, this assessment is remarkable. Firstly, the critical point is located at low temperature (equal to 583.75C i856.9K). Secondly, calculating the derivative &/ax* along the phase boundary L&+Lz, one can make certain that the monovariant line is convex. Unfortunately, the authors did not take into consideration the solid solubility of bismuth and zinc, treating the boundary rhombohedral and hexagonal solutions as pure Bi and Zn, correspondingly. While their estimation of the eutectic temperature 415.85C/689K is irreproachable, the composition of liquid in the eutectic point, x= 0.39, deviates significantly from the value x= 0.37 recommended in [90 Mas]. The maximum at the A,,HL(x) is located at x = 0.611, although the results of direct calorimetric measurements [58 Wit] imply that the mole fraction of zinc is less than 0.6.

D. V. MALAKHOV
OPTIMIZATION PROCEDURE

The program package BINGSS developed by H.L. Lukas [77 Luk, 92 Luk] was chosen to carry out the optimization. The software offers a wide choice of thermodynamic models, but for the liquid phase the preference was artlessly given to the simple and flexible Redlich-Kister formalism: AGL =?(l-&(T)(l-2x) Ml [58 Wit] imply that Am,HL is a weak function of temperature, the functions in (1)

Since calorimetric measurements

are taken in the form f;(r) = 4 + B,T . Although thermodynamic properties of neither the rhombohedral nor the hexagonal solid solution have been measured, it was possible formally to describe them within the framework of some models and to find the best values of corresponding adjustable coefftcients in the course of optimization. From the mathematical point of view, this a correct procedure, but the vulnerability of this approach from the thermodynamic point of view is evident [82 Kohl. It was decided to adopt the following simple regular expressions with known coefficients:
A% RHOMBO =xRHoMEo(l-xRHoMBo)xlOOOO, AGHcp =xHcp(l-~Hcp)x35000

(2)

The numerical values of coefftcients in (2) were not chosen arbitrarily, but a detailed explanation how they were fixed is not relevant here, because the calculated phase diagram is not sensitive to their moderate variation. It is probably sufticient to say that these large positive values lead to the required levels of solid state solubility of components. B INGS S provides a great deal of flexibility when statistical weights of different data sets (groups of measurements) are to be changed. In particular, a specially introduced parameter I FAC is used to redetermine an initially assigned statistical weight 0: making it equal I FAC x up/l 00 . The author strongly believes that this flexibility must be used with extreme care. Let us recall that the optimization of data on thermodynamic properties and phase diagram is essentially a statistical problem and that its solution must be controlled using statistical criteria such as variance of approximation, asymmetry and excess of weighted residuals, presence of trends etc. In order to get good sample statistics, one can experiment with I FAC, obtaining more and more promising statistical characteristics of a solution but forgetting that such improvements can be also achieved for a faulty phase model. It is considered that all statistical weights should be estimated and fixed before the optimization procedure is invoked. Certainly, it may be disclosed later that some of them should be refined, but one must have more serious reasons to change the weights than intermediate and immediate optimization results. An outcome of considered analysis, comparison and cross-checking of available data is presented in Table 4 Taken together with Tables 1-3, it gives the comprehensive understanding of what statistical weights were actually used. Let us emphasize again that I FAC values reflecting our confidence in reliability of different data sources had been evaluated before invoking B I NGS S . TABLE 4. Levels of Confidence Granted to the Data Sources Used in Optimization

aEMFdata
data on liquid-solid equilibria

4data on Ii&d-liquid equilibria chemical analysis of separated layers

THERMODYNAMIC

ASSESSMENT:

Bi-Zn SYSTEM

THERMODYNAMICPROPERTIESOFRHOMBOHEDRALZN The lattice stability of rhombohedral phase of zinc, which is not included in [91 Din], was estimated by Dr. A.T. Dinsdale in the following fashion [97 Din]. First, the free energy of fusion of Zn was set, in approximation, as G; -Gr = 732:2 - 10.5706T . After that, the fusion data for rhombohedral As, Bi and Sb were analyzed and it was guessed that the entropy of melting of rhombohedral Bi should be high. A certain value, 22 J/(molxK), was postulated. Since Zn does not transform to another modification on applying external pressure, it could be assumed that Zn/RHOMBO/ is never stable. To reflect this, any negative temperature of transition Zn/HCP/ + Zn/RHOMBO/ can be adopted. After the temperature -2OOK had been selected, the relation GI,HOUBGr = 2285.88 + 11.4294T was obtained and rounded to G~,HoMBoGF = 2300 + 11.5T

OPTIMIZATION:INITIALSTEPS It can be assumed that the huge array of accumulated data on thermodynamic properties of the liquid phase and solid-liquid equilibria is doubtlessly suflicient to build a reliable phase diagram even without using data on liquidliquid equilibria and that an addition of forgotten data on the miscibility gap will not change the description drastically. To check this, all data on miscibility gap and monotectic equilibrium were temporarily excluded and optimization was carried out with different number of terms n (summands) in the model (l), n = N + 1. The accuracy of the description thus obtained is shown in Table 5. TABLE 5. Accuracy of Descriptions Obtained with Different Numbers of Terms Number of terms 3 4 5 Variance of approximation 15.2 2.73 1.81 95% critical point -1.08 99% critical point -1.12 Number of terms 6 7 Variance of approximation 1.11 0.955 95% critical point -1.08 99% critical point -1.12

Since special efforts have been undertaken to assign reasonable statistical weights to experimental measurements, the variance can be considered as a realistic measure of the quality of fitting and thus can be directly compared with Fishers crincal points. When n < 4, the accuracy of description is bad. When n > 6, the accuracy is too good and the calculated Gibbs energies of liquid phase have multiple inflection points. These approximation phantoms lead in turn to unacceptable phase diagrams as shown in Fig. 3. It is worth noting that in all cases 7&, is located much above 600C and it is becoming clear that temporarily forgotten data must be invoked. At the next step all data on monovariant liquid-liquid equilibria were still disregarded, but information on nonvariant monotectic equilibrium was utilized. The monotectic temperature 416C and the compositions of coexisting liquids x4 = 0.37 and xL? = 0.993 were assigned artificially enlarged statistical weights. In some sense, this approach can be thought as optimization with constraints, because the monotectic equilibrium is forced to take place at the predetermined conditions. Figuratively speaking, an introduction of such constraints can be considered as reducing the effective number of coefficients. As follows from the comparison of Table 5 and Table 6, the accuracy of the description is significantly degraded by the inclusion of data on the monotectic equilibrium. Since the mean square of error is a generalized characteristic only, it is interesting to identify which data sets are most important in the approximation variance growth. The majority of residuals has reasonable values and only a few points corresponding to the enthalpies of mixing do not conform well to their experimental counterparts [58 Wit]. It is worth noticing that the shape of AmUHL(xL) is signiflcantly affected by the number of terms used. As Fig. 4 suggests, this number must be kept large enough for a simultaneous description of data on nonvariant transformation and enthalpy of mixing with an acceptable accuracy. Fig. 5 demonstrates the number of terms has also a pronounced effect upon the shape of miscibility gap. Remarkably, in all cases T,,,, is still located above 6OOC.

D. V. MALAKHOV

750 1000 700 650600550;;;;;

4 terms
6 terms

a.,.
*

------5terms

;,*

tss0

5;500I-450400350300-

,I,. ,, : , ,: ,.. ,, .. :i /# :; .I .;

^C ,,,!.( ,: />
, , , 0.4 0.5 0.6 X
I I

1
850 800 750 _-I 700 -650 -600 -550
I

900

3000

.? 0 ,:? o

is

; "; .E 2000 0

0
l

[27 Kaw] [58 Wit]4 terms

1000

------5terms .._. 6 terms _._._._._ terms 7

25ovI,, 0.0 0.1 0.2 0.3

0.7 0.6 0.6

1.0

0/ 0.0 0.1

0.2

0.3

0.4

0.5 X al

0.6

0.7

0.6

0.9

FIG. 3. Liquidus caIculate?with different numbers of terms (only the thermodynamic properties were used)

FIG. 4. Dependence of the calculated enthalpy of mixing on the number of terms

TABLE 6. Accuracy of Descriptions Obtained with Different Numbers of Terms 1Number of terms 4 5
1 Variance of approximation 7.31 5.00

195%critical 199%critical 1Number 1 Variance of I 95% critical I 99% critical 1

point -1.08 point -1.12 of terms 6 7 approximation 2.06 2.46 point -1.08 point -1.12

Using the derivative a*T/dx* taken along the phase boundary, curvatures of the monovariant lines L&+Lr

were

calculated for the model (1) with different numbers of terms. Concavity regions are absent if n < 5, but they appear for the greater number of summands required for a good fitting quality. Since an appearance of such artifacts has been declared a shortcoming, it was clear something should be done to make the miscibility gap convex. Initially it was expected that the simplest way to eliminate the false inflection points and decrease the critical temperature is to include the most reliable data on the miscibility gap, namely [55 Sei, 62 Koz, 88 Gla, 93 Kan, 96 Kha], in the set of experimental quantities used by BINGSS and simultaneously increase, if necessary, their statistical weights. It was found that this way led to the desirable lowering of T,,,,, but the regions of concavity could not be removed, in spite of utilizing various numbers of terms, different values of I FAC, etc. This situation is understandable: deviations between series (each of the aforementioned five papers represents one data series) are significantly larger than within the series; the combined data array can be qualified as noisy. BINGSS attempts to minimize the target function, a sum of weighted residuals, but nothing restrains the program from constructing a Gibbs energy, which generates a liquidus with manifold false inflection points. Something further is required to help B IN GS S to generate a convex phase boundary.
700650600550-

,...
4terms ------5terms ...... 6 terms

~_._._._ terms 7

8 500' c_450_ - 650 - 600

-550 0.0 0.1 0.2 0.3 0.4 0.5 X 0.6 0.7 0.6 0.9 1.0

FIG. 5. Liquidus calculatet&itb different numbers of terms (the thermodynamic properties and information on the monotectic equilibrium were utilized)

THERMODYNAMIC

ASSESSMENT:

Bi-Zn SYSTEM

The main idea is simple. Let us approximate experimental data on Lt/Li+Lz and Li+L& by a function for which convexity is guaranteed, and replace actual observations with function values calculated on a dense mesh. To realize this approach it is first necessary to find a proper class of functions satisfying the following conditions: 1. 2. 3. 4.

It contains flexible approximants which do not introduce systematic errors (trends);

The class consists of functions with a controllable smoothness; It is suitable fcr isogeometric approximation when the slope and curvature are maintainable; There are effective algorithms to build functions belonging to this class. [78 Boo] match all these requirements. It was decided in favor of B-splines of the fourth order, thus proseparation f was taken as (3) Since a B-spline is defined on a knot seproblem

B-splines

viding a continuity of the second derivative. The temperature of liquid-liquid

ncor f=

where a, are coefficients of B-spline,

t=, ncoef is a number of coefficients.

2%4,, (XL)

quence t , numerilcal values of coefftcients can be formulated as

a in (3) depend on the placement of knots. The minimization

zw,(q-i;)+min(a,t), a2f/ld(xL)* <OV xL E [0.37,0.994], f(0.37) = f(0.994) = 689.15 (4) ,=I where m is a number of observations, w, is a statistical weight of the i-th measurement. The last condition in (4) was added in order to guarantee that the B-spline is bounded with monotectic points. Usually, when B-splines are utilized for fitting, a number of required coefficients ncoef is selected to satisfy one or other statistical criterion based on accuracy. Due to the high noise level of input data this method could not be utilized and a simple visualization-based approach was employed. Looking at the experimental points in Fig. 6, one can roughly point at four regions the miscibility gap consists of the slanting and almost rectilinear region of increase, the region of growth with moderate curvature, the region of decay with increasing curvature and the extremely steep and almost rectilinear region of decline. The heuristic postulate that 3 internal knots would be sufficient resulted in seven required coefficients. The resulting isogeometric B-spline, i.e., the solution of the problem (4) is presented in Fig. 6. A set of temperatures f(x) calculated along the phase boundary on a dense 600
575 550

so0 y 775 +-

525 500.-I

is
475 450 425

0.4

0.5

0.6
X

0.7

0.6

0.9

1 0

Zn FIG. 6. Isogeometric B-spline approximation of the miscibility gap

mesh (with the array pitch 0.01) was used instead of data from [55 Sei, 62 Koz, 88 Gla, 93 Kan, 96 Kha]. AT = 1 and AX.= 0.0001 were used as errors for this new data set, but actual weights can be controlled via I FAC.
OPTIMIZATION:COMPLETION

Two degrees of freedom can be used to finalize the assessment. First, a different number of terms in the model (1) can be tested. Second, I FAC can be varied to eliminate concavity regions and reach a reasonable critical tempera-

10

D. V. MALAKHOV

ture. As it follows from the section Optimization: initial steps Redlich-Kister models of the liquid phase with 5, 6 or 7 terms (i.e., with 10, 12 or 14 coefficients) can be considered as potentially leading to a success. Through a series of numerical experiments, it was determined that models with 5 and 6 terms were inadequate whatever the value of I FAC. Acceptable results of optimization were obtained only with seven summands. Phase diagrams calculated with different weights are presented in Fig. 7. When IFAC is small, the shape of gap is incorrect, but when the weights are enlarged, concavity regions disappear. The influence of the I FAC parameter on the enthalpy of mixing is shown in Fig 8. It was also found that A,&, is quite stiff against moderate variations of I FAC, which result in variations much less than the experimental errors. After it had become clear that the model AGL =(1-x)x&A,+~,~)(l-2x) r=O can be used to represent the thermodynamic properties of liquid phase, it remained only to find a proper I FAC and finalize the computations. It was thought logical to take the smallest value of I FAC that could be reconciled with the desired convex shape of the phase boundary and provide the critical temperature below 58OC. A solution of this problem resulted in the optimal coefficients given in Table 7. Their covariance matrix can be found elsewhere [97 Mall.

___
600 850 550 600500 750450 Y 6 700650-1FAc=010 _--- IFAC=OBO IFACIOO - 400

4500400035003000250020001500IOOO-

600SO-

-.-.-.IFJ)C= 200 -"-' IFAC = 500 -

- 350

-300 -250 500 0.0 0.1 0.2 0.3 0.4 0.5 X al I 0.6 I 0.7 I 0.6 I 0.9 1.0

500 0, , , , , , , , , , 0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 0
X

FIG. 7. Liquidus calculated with 7 terms and different values of I FAC

FIG. 8. Enthalpy of mixing calculated with 7 terms and different values of I FAC TABLE 7. Coefficients of the Thermodynamic Model of the Liquid Phase B, A, 18265.09 -8.67630 i 3 A. 7227.44 21123.07 B, -9.29050 -27.14705 i A. 5 -20747.56 6 -7600.36 B, 22.01759 13.15957

i 0

-l -606121 2 -6422.60

o79581 I 4 11.71966

Although the results of original experiments on liquid-liquid separation were not used during optimization (their Bspline representation was used instead), they were taken into account for the calculation of a mean square of weighted residuals. The variance of approximation 2.67 (compare with 2.46 in Table 6) is reasonable, especially if it is recalled that all weights had been estimated before optimization and that they were not changed to conform with the statistical characteristics of the solution. The bell-shaped distribution of weighted deviations shown in Fig. 9. The magnitude of residuals is reasonable except a couple of points corresponding to the enthalpy of mixing. The calculated phase diagram is matched in Fig. 10 against experimental data. The coordinates of the critical point are x = 0.8702 and T=579.80C / 852.95K. Confidence intervals of calculated phase boundaries are published in

THERMODYNAMIC

ASSESSMENT:

Bi-Zn SYSTEM

[97 Mall. Characteristics of nonvariant transformations 9.

are specified in Tables 8 and

TABLE 8 Calculated Characteristics of the Monotectic Equilibrium Temperature C 1 K 416.51 1689.66 Bi-rich liquid at. % wt. % 37.083 115.571 Composition Zn-rich liquid at. % wt. % 99.253 197.651

-20

TABLE 9. Calculated Characteristics of the Eutectic Equilibrium Rhombohedral solution at. % 1 wt.% 1.6360 IO.51772 Composition Licuid at. % wt. % 8.1000 2.6839 Hexagonal solution at. % 1 wt. % 99.985 1 99.689

-I 0 5 we~htsd deyiatio

FIG. 9. Weighted residuals

600-

: 0 0

g I--

500-

: 0 . . 600

400-

0.0 0.1 0.2

0.3 0.4 0.5 0.6 0.7 0.6 0.9 X

1.0

FIG. 11. Enthalpy of mixing calculated with the proposed model Experimental enthalpies of mixing are compared with computed ones in Fig. 11. Excess chemical potentials of Zn deduced from experimental data are confronted with calculated quantities in Fig. 12. The method used for data visualization is based on the trivial consideration that a 3D picture of A,&, (T,xL) is inconvenient, To avoid this difflculty, the 3D picture was represented as a set of 2D slices. All experimental quantities were subdivided into groups. Each group corresponds to a temperature region from r,,,, to T,, Concentration dependencies A&, (xL) were calculated fox the lower and upper temperatures and plotted together with experimental values measured within the same interval. For example, in the left upper plot of Fig. 12 the solid line is A,&, (x) at T = 400C, while the dashed line is AU&,, (xL) at T = 440.85C. urements at 400 c :r 5 440.85C. Some of the A& (xL) dependencies demonstrate whimsical behavior, especially noticeable at high temperatures Solid circles represent excess chemical potentials deduced from meas-

FIG. 10. Phase diagram czculated with the proposed model in comparison with the experimental data

when regions of convexity and concavity (and even intervals of growth and diminution!) change many times. Fortunately, these oscillations in excess properties do not have any dramatic or even visible effect on the Gibbs energy of mixing as Fig. 13 on the next page proves.

D. V. MALAKHOV

FIG. 12. Concentration dependency of the chemical potentials of Zn at selected temperatures

ADDENDUM The author acknowledges with thanks the helpful comments of a referee, who pointed out that several pertinent references had not been taken into account in the original manuscript. As noted below, these previously overlooked results can serve as a useful independent test of the thermodynamic model.

Liquid immiscibility in the Bi-Zn system was studied in [70 Asa] by means of DTA and a hot separation method. In addition, the vapor pressure of zinc was measured by flow method. The critical temperature was found equal to 583C at x = 0.83, which is in a good agreement with the
-4.5 -5.0 , 0.0 , 0.1 , 0.2 k.,, , 0.3 % _ , 0.4 ,, ,, ,... , 0.5 X Zn ,... I 0.6 , 0.7 I 0.6 I 0.9 1.0
620 .

t-

FIG. 13. Concentration dependencies of the Gibbs energy of mixing at selected temperatures proposed model. The miscibility gap was studied by Budde and Sauerwald [65 Bud], who measured the viscosity of the Bi-Zn melt at various temperatures and compositions. To identify the boundaries of immiscibility, one has to determine the concentration region within which the dependency q(x) is linear. The values resulting from such a determination are very close to those obtained in [52 Kle], although their ac-

this wak
. .

-.---

(93Wanl l70Asal [96BOU]

[97Kat]

%. FIG. 14. Evaluated miscibility gaps and some recent experimental data

THERMODYNAMIC

ASSESSMENT:

Bi-Zn SYSTEM

13

curacy is deemed to be low due to the scarcity of experimental data and a significant level of random errors. The isothermal solubility method was utilized in [96 Bou] to study the region of immiscibility. As follows from Fig. 14, the reliability of experimental results can be put into question, since the eutectic composition found in [96 Bou] is approximately rive atomic percent less than the adopted value [90 Mas]. The authors of [97 Kat] used the position of breakpoints on the voltage vs. composition dependencies to localize the phase boundary Li/Li+Lz. As demonstrated in Fig. 14, the critical temperature exceeds 6OOC. This might indicate that equilibrium was not achieved in their experiments. The EMF method was employed in [73 Mar, 76 Hay, 97 Kat] to measure the composition and temperature dependencies of the chemical potential of zinc in the liquid phase. As seen from Fig. 15 and 16, the calculated properties are consistent with the experimental quantities.

FIG. 15. Concentration dependencies of the chemical potential of zinc in the liquid phase at selected temperatures in comparison with the experimental data

FIG. 16. Concentration dependencies of the zinc activity in the liquid phase at selected temperatures in comparison with the experimental data

In [96 Bou] enthalpies of eutectic and monotectic transformations Liquid(x = 0.07575) + Pure Zn + Pure Bi at T = 256k 1C Liquid(x = 0.990417) -+ Pure Zn+Liquid(x = 0.301554) at T = 416.8f0.2C

were found equal 1.0 -53 k 1 and -90 k 4 J/g, respectively. Calculating the enthalpies of these transformations using the thermodynamic models of phases proposed in the present paper and with the equilibrium concentrations and temperatures specified in Table 8 and 9, i.e., the enthalpies of reactions Liquid(x = 0.081) + HCP(x = 0.99985) + RHOMBO(x = 1.6360.10m2) at T = 254.5C Liquid(x = 0.99253) -+ HCP(x = 0.99930) + Liquid(x = 0.370873) at T = 416.51C one arrives to -59.1 and -112.7 J/g, respectively. Taking into account a significant difference in compositions, it can be concluded that the deviation between the measured and computed enthalpies is within the reasonable limits.
REFERENCES

1892 Hey CT. Heycock, F.H. Nevillehurnulofthe ChemicalSociety, 1892, Vol. 61, P. 888-914 1897 Hey CT. Heycock, F.H. Neville Journal ofthe Chemical Society, 1897, Vol. 71, P. 383-422 1897 Spr W. Spring, L.Romanoff Zeitschriflfir anorganische und allgemeine Chemie, 1897, Vol. 13, P. 29-35 27 Kaw M. Kawakami The Science Reports ofthe TGhokuImperial Universit), 1927, Vol. 116, P. 9 1.5-935

14

D. V. MALAKHOV

33 Has K. Hasq K. Jellinek Zeifschriftfir unorganische undailgemeine Chemie, 1933, Vol. 212, P. 356-361 52 Kle O.J. Kleppa J. Am. Chem. Sot., 1952, Vol. 74, P. 6052-6056 55 Sei W. Seith, H. Johnen, J. Wagner 2.Metallkde, 1955. Vol. 46, P. 773-779 57 Ole L. Oleari, M. Fiorani, V. Valenti La Metollurgia Ztalianu, 1957, No. 9, P 677-683 58 Han M. Hansen, K. Anderko Constitution of binary alloys McGraw-Hill, New York, 1958 58 Wit F.E. Wittig, E. Miiller, W. Schilling Z. Metallkde, 1958, Vol. 62, P. 529-544 59 Kle O.J. Kleppa, C.R. Thalmayer Am. Chem. Sot., 1959, Vol. 63, P. 1953-1958 J. 60 Lan M.F. Lantratov, E.V. Tsarenko Zhurnal Prikladnoi Khimii, 1960, Vol. 33, P. 1116-I125 61 Yok T. Yokokawa, A. Doi, K. NiwaJ. Am. Chem. Sot., 1961, Vol. 65, P. 202-205 62 Koz Z. Kozuka, J. Moriyama, I. KushimaDenki Kaguku, 1962, Vol. 30, P. 591-596 65 Bud J. Budde, F. Sauerwald Zeirschrififiir Physikalische Chemie (Leipzig), 1965, Vol. 230, P. 42-47 65 Chi Y. Chiba, T. Matsushima, K. Ono Bull. of rhe Research Insfitute of Mineral Dressing and Metallurgy Tohoku University, 1965, Vol. 21, P. 15-22 65 GIu J.V. Gluck, R.D. Pehlke Trans. of the MetallurgicalSoc. ofAZME, 1965, Vol. 233, P. 223-240 66 Pta W. Ptak, Z. Moser Archiwum Hutnictwa, 1966, Vol. 11, P. 207-234 67 Glu J.V. Gluck, R.D. Pehlke Trans. of the Metallurgical Sot. ofAIME, 1967, Vol. 239, P. 36-47 69 Mos 2. Moser Archiwum Hutnictwa, 1969, Vol. 14, P. 371-420 70 Asa N. Asano, K. Wase, T. Kaw,azawa Suiyokai-shi, 1970, Vol. 17, P. 115-l 18 73 Mar L. Martin-Garin, P. Bedon, P. Desr6 Bulletin de la SociU Chimique de France, 1973, No. 5, P. 1533-1537 76 Hay E. Hayer, K.L. Komarek, A. Mikula Monatshelfe$r Chemie, 1976, Vol. 107, P. 1437-1448 77 Bal C.W. Bale, A.D. Pelton, M. Rigaud Z. Metalikde, 1977, Vol. 68, P. 69-74 77 Luk H.L. Lukas, E.Th. Henig, B. Zimmermann CALPHAD, 1977, Vol. I, P. 225-236 77 Pel A.D. Pelton, C.W. Bale, M. Rigaud 2. Metallkde, 1977, Vol. 68, P. 135-140 78 Boo C. de Boor A Practical Guide to Splines Springer-Verlag, New York, 1978 80 Lin P.-L. Lin, A.D. Pelton, C.W. Bale, W.T. Thompson CALPHAD, 1980, Vol. 4, P. 47-60 80 Oka K. Okajima, H. Sakao Trans. ofthe Japan Inst. of Metals, 1980, Vol. 21, P. 226-236 81 Oka K. Okajima, H. Sakao Trans. oftheJapan Inst. of Merals, 1981, Vol. 22, P. 347-356 82 Koh H. Kohler, A.D. Pelton CALPHAD, 1982, Vol. 6, P. 39-47

85 Gir C. Girard Ph. D. Thesis, Universite de Provence, Marseille, 1985 85 Mat V.N.S. Mathur, M.L. Kapoor 2. Melallkde, 1985, Vol. 76, P. 16-23 88 Gla V.M. Glazov, S.G. Kim, V.I. Timoshenko, K.B. Nurov Russ. J. Phys. Chem., 1988, Vol. 62, P. 1300-1302 90 Mas Binary Alloy Phase Diagram edited by T.B. Massalsky, H. Okamoto, P.R. Subramanian, L. Kacprzak, ASM, 1990 91 Din A.T. Dinsdale CALPHAD, 1991, Vol. 15, P. 317-425 92 Luk H.L. Lukas, S.G. Fries Journal ofphase Equilibria, 1992, Vol. 13, P. 532-541 93 Kan S. Kanai, Y. TsuchiyaJ ofrhe Phys. Sot. ofJapan, 1993, Vol. 62, P. 2388-2394 93 Oka H. Okamoto Journal ofphase Equilibria, 1993, Vol. 14, P. 336-339 93 Wan Z.C. Wang, S.K. Yu, F. Sommer Journal de Chimie Physique et de Physico-Chimie Biologique, 1993, Vol. 90, P. 3 79-385 95 Gla J. Glazer International Material Reviews, 1995, Vol. 40, P. 65-93 95 Mal D.V. Malakhov Abstracts of rhe XXIV CALPHAD meeting in Kyoto, 1995, P. 3 96 Bou F. Bouharkat, J.J. Counioux, J.R. Vignalou, J. Said Journal ofAlloys andCompounds, 1996, Vol. 238, P. 149-154 96 Kha R.A. Khairullin, S.V. Stankus The study of two-melt phase separation in the Bi-Zn system by the y attenuation technique I. of AlIoys and Compound, 1996, Vol. 234, p. 260-263 97 Din A.T. Dinsdale, private communication, 1997 97 Kat 1. Katayama, F. Sommer, Y. Hattori, T. Lida Journal de Chimie Physique el de Physico-Chimie Biologique, 1997, Vol. 94, P. 1094-l 100 97 Mal D.V. Malakhov CALPHAD, 1997, Vol. 21, P. 391-400 97 Oka H. Okamoto Journal ofphase Equilibria, 1997, Vol. 18, P. 218 98 Hil M. Hillert Phase equilibria, phase diagrams, and phase transformations: their thermodynamic basis University Press, Cambridge, 1998. 98 Ohn I. Ohnuma, private communication, 1997 98 Sau N. Saunders, A.P. Miodownik CALPHAD (calculation of phase diagrams): a comprehensive guide Pergamon, New York, 1998