Estimation of molar excess Gibbs energy and component acti

vities of K-Na, Cs-K, Al-Sn and Mg-

Si liquid alloys from pair distribution functions
Chun-Long Wang, Xiu-Min Chen, Dong-Ping Tao*
Faculty of Metallurgical and Energy Engineering, Kunming University of Science
and Technology, Kunming 650093, Yunnan Province, People’s Republic of China
Keywords:
Ab-initio molecular dynamics (AIMD); Thermodynamic models; pair distribution function;
Activities prediction;
Abstract: Based on the latest research methods, that estimating the activity and excess Gibbs
energy of liquid mixtures without use thermodynamic data, extracting g1ij(r), the local distribution
function, from the partial pair distribution function gij(r) of liquid mixtures as the unique input
parameter to solve thermodynamic model parameters. AIMD(g1ij(r))/(MIVM, RSM) methods
were proposed and extend to the Wilson and NRTL. The excess Gibbs energy and activity of four
liquid alloys, K-Na, Cs-K, Mg-Si, Al-Sn, were estimated by the AIMD(g1ij(r))/(MIVM, RSM,
Wilson, NRTL) and AIMD(gij(r))/QCM(Quasi Chemical model) proposed by Dorini. The mean
ARD(Average Relative Deviation) of excess Gibbs energy of four liquid alloys for five models are
±29.46%, ±24.82, ±66.68%, ±61.50%, ±32.86% respectively and the ARD of activity are
±10.50%, ±8.76%, ±47.07%, ±26.61%, ±11.47%, respectively, that show estimability are credible.
The estimation for Mg-Si are poor, through verify the excess Gibbs energy and activity
experimental data in literature, the Mg-Si experimental data in STGE not satisfy thermodynamic
consistency might be the mainly reason for poor estimation. The estimation results of MIVM,
Wilson and NRTL were compared, and conclude that MIVM have a good theoretical basis for
structure to compensate for the theoretical deficiencies of the other two models. The comparison
between the RSM and the QCM indicate that the both of the exchange energy parameter not
consistent and cannot interchange, and the difference in estimation mean that to explain the
relation between the two models from theoretical may not accord with the actual situation. The
correlation coefficient analysis and variation coefficient analysis between the models parameters
and liquid alloy components show that the parameters containing volume information are low
correlation, which can be regarded as constants at whole component range, while the model
parameters containing energy information have functional relation with components and the
constant treatment not suitable. Apply mathematical statistical analysis in thermodynamic model
parameters can deepen the understanding of the parameters characteristic. Comparing with the
regression method, the AIMD(g1ij(r))/thermodynamic-models can deepen the understanding of the
characteristics of the liquid system thermodynamic model based on lattice theory, and reveal the
rationality of the theoretical basis of the model construction and the defects of the model caused
by simplified. AIMD(g1ij(r))/thermodynamic-models is meaningful for excess Gibbs energy and
activity estimation.
1 Introduction
The excess Gibbs energy and activity thermodynamics models are widely used in the
calculation of gas-liquid phase equilibrium reactions. Since engineers usually obtain the model

*Corresponding author
E-mail addresses: 1176038593@qq.com (C.L. Wang), dongpingt@aliyun.com (D.P. Tao)

Electronic copy available at: https://ssrn.com/abstract=4121941

parameter by using experimental of activity and excess Gibbs energy to fit regression, it is
challenge to use the thermodynamic models to estimate gas-liquid equilibrium without any
relevant thermodynamic experimental values. In 1999, Sam used AIMD to obtain molecular pair
potential from clusters of eight atoms composed of water and alcohol(acid) as the UNIQUAC
model parameters the prediction well, while Wilson model’s not good[1]. In 2002, Raabe used
AIMD/NRTL to predict the gas-liquid equilibrium of nitrogen and ethane mixtures, and the
prediction results are in good agreement with experimental data[2]. In 2004, Neiman used
COSMO-RS and AIMD/UNIQUAC to predict the phase equilibria of 16 binary organic mixtures,
the prediction result of AIMD/UNIQUAC deviated greatly from the experiments, the same
method applied to Wilson and NRTL could not get good result. Neiman pointed out that there are
two basic issues to consider when using the AIMD/thermodynamic-models. The first one concern
is whether the potential parameters of the model are true molecular pair potentials, Sam support
UNIQUAC, Raabe prefer NRTL, and Neiman's study show that all three models perform poorly.
The second concerns is whether liquid structure interaction energy can be expressed as molecular
cluster energy[3]. The excess Gibbs energy and activity models parameters calculated by first
principles are controversial, and researchers actively explore new method. In 2015, Haghtalab
used gij(r) (The subscript ij also refers to ii, jj, and ji without special explanation, in the following)
of five binary organic mixtures obtained by MD to represent the models parameters of Wilson,
NRTL and UNIQUAC, which have anadjustable parameter, local structure radius to determine the
upper limits of integral for gij(r)[4]. Haghtalab’s method can't achieve prediction, and the local
structure radius in paper, obviously, large than range of lattice theory for liquid[5]. in 2018,
Ravichandran expressed the NRTL model parameter as the function of the molecular pair
potential, molecular diameter and local structure radius, and predicted the gas-liquid equilibrium
of 14 binary organic mixtures. Ravichandran’s study show that the completely miscible mixtures
have a good prediction result, while the partially miscible mixtures are worse[6, 7]. The
AIMD/NRTL need to calculate molecular pair potential and gij(r) at same time, and the method of
acquiring local structure radius not standardized, that may lead to different conclusions in different
studies. In 2019, Dorini proposed AIMD(gij(r))/QCM, the short-range ordering σ defined as a
bridge between the gij(r) and QCM parameters β, ω. The Bi-Pb and Bi-Li liquid alloys use
Dorini’s method has a good prediction result[8]. Dorini’s method require gij(r) fitting regression
under the multi component mixture to obtain ω. In this paper, a set of gij(r) under one component
be used to solve ω, and complex solutions for parameters are appearing, which indicate that the
AIMD(gij(r))/QCM method need to improve in theory. At present, the specific methods for to
predict excess Gibbs energy and activity has different shortcoming. The Gibbs energy and activity
thermodynamics models research mainly aim to propose prediction method with rigorous physical
significance, parameter simple, good reproducibility and good estimability.
In this context, proposing a AIMD(g1ij(r))/(MIVM, RSM) method for to estimate the excess
Gibbs energy and activity of liquid mixtures. The gaussian extrapolation method were used to
extracting g1ij(r) from gij(r) obtained by AIMD simulation, using the g1ij(r) calculate the local
coordination number Zij and local pair potential function ε1ij(r). The local structural parameters,
g1ij(r), Zij and ε1ij(r), were used to derive the models parameter expression, and were extended to
Wilson and NRTL. The excess Gibbs energy and activity of four liquid alloys, K-Na, Cs-K, Mg-Si
and Al-Sn are estimated by the AIMD(g1ij(r))/(MIVM, RSM, Wilson, NRTL) and AIMD(gij(r))
/QCM proposed by Dorini, which enriched the research about estimating excess Gibbs energy and

Electronic copy available at: https://ssrn.com/abstract=4121941

activity. The correlation analysis and discrete analysis to judge the model parameters calculated by
g1ij(r) and liquid alloy components whether both of them have function relations and the feasibility
of constant treatment. Mathematical statistical analysis in thermodynamic model parameters was
applied to deepen the understanding of the parameters characteristic.
2 Methods

2.1 AIMD obtain part pair distribution function

The successful application of simulation methods based on first principles have increased the
recognition that computer experiments as a research method. The gij(r) of liquid obtained by
AIMD are credible as long as the simulation operation standardize and correctly built the the
liquid model. The AIMD has the problems of time consuming and high computational cost, due to
require the large amount of gij(r) data in this research, limited experimental condition make it
difficult to obtain all the data through own’s experiments. AIMD be used to reproduce the
examine the existing gij(r) data in the literature, so as to prove the feasibility of AIMD to obtain
gij(r) and the reliability of published data. Fig.1 are gij(r) of four liquid alloys in different
components. Korkmaz applied the effective ion-ion pair potential function that proposed by
Fiolhais to describe the pair potential of K-Na and Cs-K alloys, and obtaining the gij(r) of nine
components corresponding to Na-K and Na-K by solving the Ornstein-Zernike Integral
equation[9, 10]. LiuDan obtained the gij(r) corresponding to Mg-Si at different temperatures and
different components by AIMD[11]. Roik used Reverse Monte Carlo Simulation to obtain the
gij(r) of Al-Sn at liquid alloy seven components[12].
11
K-K Na-Na K-Na x :0.9
K
10 0.8
0.7
9 0.6
0.5
8
0.4
7 0.3
0.2
6 0.1
g(r)

5
4
3
2
1
0
3 4 5 6 7
2 4 6 3 4 5 6 7
r/Å r/Å r/Å

Electronic copy available at: https://ssrn.com/abstract=4121941

 11
x :0.9
Cs-Cs K-K Cs-K Cs
10 0.8
0.7
9 0.6
0.5
8 0.4
7 0.3
0.2
6 0.1
g(r)

5
4
3
2
1
0
4 6 8 4 6 8 4 6 8
r/Å r/Å r/Å

12
Mg-Mg Si-Si Mg-Si x :0.9
Mg
0.8
10 0.67
0.6
0.47
0.4
8 0.3
0.2
0.1
g(r)

0
2 3 4 5 2.5 3 3.5 2.5 3 3.5
r/Å r/Å r/Å

Electronic copy available at: https://ssrn.com/abstract=4121941

 9
Al-Al Sn-Sn Al-Sn x :0.97
Al
8 0.92
0.87
0.8
7
0.65
0.5
6 0.25

5
g(r)

0
2 3 4 2 2.5 3 3.5 4 4.5 2 3 4 5
r/Å r/Å r/Å
Fig. 1 Partial pair distribution function gii(r) , gjj(r) , gij(r) ,for the liquid alloys K- Na[9], Cs-
K[10],Mg-Si[11],Al- Sn[12] form literature
The AIMD method, proposed by Car and Parrinello, can calculate particle forces and
trajectory motion[13]. By using VASP(Vienna AB-Initio Simulation Packages) software to carry
AIMD simulation[14]. The generalized gradient parameterized by Perdew and Wang was used to
approximate the exchange correlation potential[15, 16]. The cutoff energy, step size has a great
influence on the accuracy of the results, and they need to be determined by the energy
convergence test, the value listed in Table 1. The K point of all liquid alloys chose 3*3*3. The
temperature is controlled by the regular ensemble (NVT) and Nose thermostat[17]. The simulation
conditional density are consistent with the literature, they are listed in Table 2. AIMD selects two
stages, the first stage is to achieve equilibrium of the system at the corresponding temperature, and
the second stage is used to obtain gij(r).
Table 1 Parameters of AIMD simulation for K0.5Na0.5, Cs0.5K0.5, Mg0.6Si0.4,
Al0.5Sn0.5
Structure VASP

Alloy T/K Atom Length Cutoff Energy t1 t2 ∆t

Number 10-8/cm (Ev) /ps /ps /fs

K0.5Na0.5 373 128 19.664 140 5 10 1

Cs0.5K0.5 373 72 15.96 320 4 6 2

Mg0.6Si0.4 1435 72 11.41 200 4 6 2

Al0.5Sn0.5 973 108 13.76 249 6 9 3

Electronic copy available at: https://ssrn.com/abstract=4121941

 Table 2 Numerical density of liquid alloy K0.5Na0.5, Cs0.5K0.5, Mg0.6Si0.4, Al0.5Sn0.5
K-Na[9] Cs-K[10] Mg-Si[11] Al-Sn[12]
ρ0/Å-3 ρ0/Å-3 ρ0/Å-3
xi xi xi T/K ρ0/Å-3 xi
T =373 K T =373 K T =973 K
0.1 0.0223 0.1 0.0123 0.1 1722 0.0452 0.25 0.0367
0.2 0.0207 0.2 0.0118 0.2 1643 0.0446 0.5 0.0404
0.3 0.0192 0.3 0.0114 0.3 1548 0.044 0.65 0.0434
0.4 0.0179 0.4 0.0109 0.4 1431 0.0434 0.8 0.0454
0.5 0.0168 0.5 0.0105 0.47 1319 0.0429 0.87 0.0472
0.6 0.0158 0.6 0.0101 0.6 1435 0.043 0.92 0.0477
0.7 0.0147 0.7 0.0096 0.67 1458 0.0426 0.97 0.0511
0.8 0.0140 0.8 0.0092 0.8 1383 0.0419
0.9 0.0132 0.9 0.0087 0.9 1254 0.0413

2.2 Thermodynamic model

The five models, MIVM, RSM, Wilson, NRTL, QCM are used to estimate the excess energy
and activity of liquid mixture. The corresponding model equation given and point out the
parameters unsolved.
2.2.1 MIVM(Molecular Interaction Volume Model)
Dong-Ping Tao rewrote the partition function of liquid based on lattice and cell theory, and
obtain the excess Gibbs energy expression of liquid mixture——MIVM(Molecular Interaction
Volume Model)[18].
GmE  Vmi   Vmj  xi x j  Bij B ji  (1)
 xi ln   x ln  ln Bij  Z j
 x V  x V B  j  x V  x V B  2  i x  x B
Z ln B ji 
RT  i mi j mi ij   j mj i mi ji   i j ij x j  xi B ji 
GmE is excess Gibbs energy; R is the molar gas constant value 8.314 J/(K.mol); xi, xj are the
component value fractions of substances i and j in the mixture respectively. Vmi,Vmj are the molar
volumes of pure substance i and j respectively; Zi, Zj are the first coordination number of pure
substances i and j; Bij and Bji are unsolved parameters. Dai Heng point out Bij and Bji represent
volume information,while lnBij and lnBij represent energy information, the meanings of Bij and Bji
parameters were not clear enough[19]. In this paper, Dai Heng’s parameter separation method
were adopted, Bij and Bji represent volume information, lnBij and lnBji are replaced by new symbols
∆εij, ∆εij and give new expression in subsequent work.
Z ij Z ji
Bij  B ji  (2)
Z ii Z jj

 ij   ii  ji   jj (3)
 ij  ln Bij   ji  ln B ji 
kT kT
Bij is the ratio of local coordination number of Zii, Zij, Bji represents the ratio of local
coordination number of Zjj and Zji. The first and second term in MIVM is volume term contain
entropy information. ∆εij, ∆εji is the pair potentials difference of molecular pairs i-i and i-j, j-j and
j-i respectively. The MIVM third term is energy term contain enthalpy information. k is
Boltzmann constant value 1.38*10-23.

Electronic copy available at: https://ssrn.com/abstract=4121941

2.2.2 RSM(Regular Solution Model)
In 1929, Hildebrand proposed the theory of regular solution that the mixing entropy
vanishing at constant temperature and constant volume[20].

GmE w (4)
 xi x j
RT kT
w, interchange energy, is the function of coordination number and pair potential.
2.2.3 Wilson
Wilson presented the following expression for the excess Gibbs energy of a binary solution,
on the basis of molecular local structure consideration[21].

gE
  xi ln  xi  Aij x j   x j ln  x j  Aji xi  (5)
RT
Aij, Aij are related to the pure components molar volumes and local structure volume. Wilson
expression can convert to the MIVM volume term.
2.2.4 NRTL(Non Random, Two Liquid)

Renon used the concept of local component in derivation of the NRTL(Nonrandom two-
liquid) equation, NRTL unlike Wilson’s, which is applicable to partially miscible as well as
completely miscible systems[22].

gE   ji exp   ij ij   ij exp   ij ij  

 xi x j   
RT  xi  x j exp   ij ij  x j  xi exp   ij ij   (6)
 
αij is related to the non-randomness in the mixture, the values range from 0.20 to 0.47, in
paper αij =0.3; τij is the molecular energy difference between i-i and i-j, τij is the difference
between the molecular energy of j-j and j-i.
2.2.5 QCM(Quasi Chemical Model)
The lattice theory assume the different kind of atoms mixing complete random in mixtures,
that can derive the RSM, but the different type atoms will non-random in actual situation.
Guggenheim used the Quasi Chemical approximation to deal with random mixing and non-
random mixing, and gave the parameter β to connect random mixing and non-random mixing,

which is the QCM[23].

GmE   2  (7)
 xi x j  
RT kT   1

   
2


  1  4 xi x j exp   1 (8)
  ZkT  
 
β represent the random mixing factor; Z is the mean coordination number of binary mixtures,

when knowing the Zij can calculated using geometric rule； ω is the exchange energy of the

Electronic copy available at: https://ssrn.com/abstract=4121941

mixture, in theoretically ω=w. Dorini proposed a method AIMD(gij(r))/QCM to calculate β and
ω, which was used in this paper.
2.3 Obtaining model parameter from gij(r)
The five model’s parameters can be divided into volume parameter and energy parameter,
volume parameter: Bij, Bji, Λij, Λji, β ; energy parameter: ∆εij, ∆εji, w, τij, τji. This section introduce
the methods obtaining volume parameters and energy parameters according to g1ij(r).
2.3.1 Volume parameter
The gij(r) of liquid can well represent the liquid structure of short-range ordering and long -
range disordering. The gij(r) have an obvious peak near the origin of coordinates, and the second
peak and the third peak will appear with the distance increase. The first peak contain important
liquid structure information, which can combine with lattice theory[5]. the g1ij(r), the pair
distribution function of the first coordination shell or called local distribution function, extract
from the first peak of gij(r). The g1ij(r) was used to characterize the local structure characteristics
of liquid alloy. I use 27 pure liquid metals to perfect the method of constructing g1ij(r). Fig.2 is the
schematic diagram of constructing g1ij(r), and Eqs.(9) is the expression of g1ij(r). After systematic
evaluation, Eqs.(9) have high accuracy and good performance in coordination number calculation,
and will continue be used in liquid alloy.

Fig.2 Schematic graph of g1ij(r) constructed by the Gaussian function

  rm  r 2 
 g ij rm  exp - 2
 r0  r  rm

  2 v 
g ij1 (r )   (9)
 g r  exp - r  rm   r  r  r
 2

 ij m  2u 2 
m 1
 
r0 is the start point at gij(r) not 0; r1 is the position of the first valley value of the gij(r); v and
u are solved by fitting the data on the left side of the peak and the coordinates of point 2
[(rm+r1)/2, gij((rm+r1)/2)] in Fig.2. According to the meaning of g1ij(r), the total number of atoms
in the g1ij(r) first coordination shell defined as the local coordination number Zij, the number of
atoms j distributed around the central atom i in the adjacent sphere. Binary mixture have four type
coordination numbers.
Z ii  xi  0 4  r 2 g ii1 (r )dr； Z ij  x j  0 4  r 2 g ij1 (r )dr；
(10)
Z jj  x j  0 4  r 2 g 1jj (r )dr； Z ji  xi  0 4  r 2 g 1ji (r )dr；
ρ0, the average number density, is a function of alloy component and temperature; The Zij is an
important indicator of volume information of liquid local structure, and all volume parameters of

Electronic copy available at: https://ssrn.com/abstract=4121941

five models can be calculated based on Zij . An important concepts in two fluid theory —— local
fractions xii, xjj, xij and xji can obtained from Eqs.(10) [18, 21].
Z ii Z ij
xii  xij 
 Z ij
j 1
Z
j 1
ij
（11）
Z jj Z ji
x jj  x ji 
Z
j 1
ij Z j 1
ij

Wilson parameters Aij, Aji and MIVM’s Bij, Bji both the expressions of xij, bring Eqs.(11) into the
parameters derivation process can obtain new expression expressed by g1ij(r), or bring Eqs.(10)
into Eqs.(2) also can obtaining.

r r
2
gij1 (r )dr 2
g 1ji (r )dr
B ,A = B ji ,A =
r r
ij ij ij
2
gii1 (r )dr 2
g 1jj (r )dr （12)

The parameter form of MIVM and Wilson expressed by g1ij(r) as same as Haghtalab’s. The
Eqs.(12) differ from Haghtalab’s in that the g1ij(r) is complete closed function, which represent the
local structure information, while Haghtalab’s not have this meaning. Haghtalab’s upper limit of
integral not reasonable enough[4]. The Aij, Aji in Wilson’s work have the volume ratio of two pure
substances i and j, but this paper and Haghtalab’s work take the volume ratio out of parameter
expression, because adding volume ration make the estimation worse. The QCM parameter β was
calculated by Zij of liquid alloy in Dorini’s work, and the process was briefly described. The short-
range ordering parameter σ and the average coordination number Z can written as the expression
of the coordination number Zij.
Z ij Z ji
  xi  xj
Z ii  Z ij Z jj  Z ji （13)

Z  xi ( Z ii  Z ij )  x j ( Z ji  Z jj ) （14)

The short-range ordering σ also can be expressed by random factor β.

4 xi x j

 1 （15)
In the case of knowing Zii, Zjj, Zij, Zji, the β can be obtained by simultaneous Eqs.(13) and (15).
By substituting Eqs.(14) and β into Eqs.(8), ω can be solved. In practical application, Dorini did
not choice solving the equation, but fitting regression to get the parameter ω, the mainly purpose
is to make ω have strong adaptability in the whole range of alloy components, and also to avoid
complex value in the process of solving the equation.
2.3.2 Potential parameter
The molecular pair interaction potential function εij(r) play a decisive role in the liquid structure
study. The gij(r) is the result of dynamic equilibrium under εij(r) interaction. Johnson study the
relation between the two-body distribution function, the three-body distribution function and the
pair interaction potential function εij(r)[24]. At present, there is little understanding and
insufficient research on the three-body distribution function, in order to simplify the representation

Electronic copy available at: https://ssrn.com/abstract=4121941

of the pair potential in terms of distribution functions. using the local distribution function g1ij(r)
to calculate the local pair potential function ε1ij(r) .

 ii1  r  (16)
  ln g (r )
1
ii
kT
According to Eqs.(16), the equation ε1jj(r) and ε1ij(r)=ε1ji(r) can be obtained. In theory, the
model energy parameters can be calculated based on the ε1jj(r). The energy parameters of RSM,
MIVM and NRTL models have equations (17) and (18).
(17)
w   ij 1   ii  jj  
Z    
kT  kT 2  kT kT  

 ij - ii  ji - jj (18)
 ij , ij =  ji , ji =
kT kT
εii is the potential of atom i interacting with atom i, the εjj, εij = εji meaning similar with εii.
The energy parameters of a model is function of pair potential of atoms, but the the specific εij
values still unsolved. The pair potential εii, εjj, εij and εji in the three models are constants values,
which meaning of the mean value of the pair potential and can be solved according to the g1ij(r)
and ε1jj(r).

g ij1 r   g r ln g r dr

1 1
 ij 1 1
 g 1 r dr  r dr   kT
ij ij
 1
(19)
 ij  g r dr
ij
kT kT 1
ij

Similarly, the expressions of εii, εjj and εji can be obtained. substituting Eqs.(19) into Eqs.(17),
(18) obtain the energy parameters of the model.
3 Results and discussion

3.1 AIMD reproduces the partial pair distribution function gij(r)

The three type gij(r) of the four liquid alloys obtained by AIMD are shown in Fig.3. The
Eqs.(20) was used to calculate the SD(standard deviation) of gij(r) in literature and AIMD, in
order to evaluate the performance of AIMD, The SD are listed in Table 3. Usually, when the SD
less than 0.01, they are overlap and indistinguishable. The SD in Table 3 range from 0.04 to 0.24
show the gij(r) obtained under the small atoms model and short time simulation condition
consistent with the gij(r) in the literature. It can be known from Fig.3 there is a significant
difference between the gij(r) from AIMD and the gij(r) from literature. The difference mainly
caused by the deviation on the right side of the peak, and there is a significant difference between
the peak and valley. Actually, the subsequent study will use g1ij(r), not gij(r). The g1ij(r)
construction method in paper have the effect of weakening the right side trend and peak valley
inconsistency, and the consistent g1ij(r) can be constructed from two incompletely coincident
g1ij(r).

Electronic copy available at: https://ssrn.com/abstract=4121941

 10
i-i j-j i-j Present work
Korkmaz
Korkmaz
8 Liu Dan
Roik
K0.5Na0.5
Cs0.5K0.5
6
Mg0.6Si 0.4
Al0.5Sn0.5

0
2 4 6 8 2 4 6 8 2 4 6 8
r/Å r/Å r/Å
Fig.3 Partial pair distribution function gii(r), gjj(r), gij(r) for the liquid alloys K0.5Na0.5, Al0.5Sn0.5,

Cs0.5K0.5, Mg0.6Si0.4
Table 3 The standard deviation of gij(r) calculated by AIMD are compared with gij(r)
in literature for four alloys
K0.5Na0.5 Cs0.5K0.5
r0 /Å r1 /Å SD r0/ Å r1 /Å SD
i-i 3.365 4.431 0.239 3.849 5.162 0.144
j-j 2.755 3.613 0.168 3.426 4.503 0.108
i-j 3.117 4.043 0.146 3.698 4.842 0.175
Mg0.6Si0.4 Al0.5Sn0.5
r0 /Å r1 /Å SD r0 /Å r1 /Å SD
i-i 2.368 3.16 0.044 2.053 2.495 0.146
j-j 2.011 2.434 0.077 2.465 3.136 0.098
i-j 2.162 2.743 0.103 2.193 2.822 0.222
r1
1
g  r   gijliterature  r  
2
SD  AIMD
ij
(20)
N r  r0

3.2 Calculation of thermodynamic model parameters

Solving the five models parameters only need a set of g1ij(r) in arbitrary alloys component. In
order to make the AIMD(g1ij(r))/(MIVM, RSM, Wilson, NRTL) has strong adaptability and
representativeness in the whole liquid alloy component, the average parameters values under each
component as the final input parameters. Dorini calculated parameters ω and β by regression,
without the average value[8]. Table 4 show the corresponding alloy model parameters of each
model.

Electronic copy available at: https://ssrn.com/abstract=4121941

 Table 4 the model parameter for five models

model parameter K-Na Cs-K Mg-Si Al-Sn

𝑤
RSM 0.510 0.498 1.966 2.115
𝑘𝑇
Zc 0.005 0.202 5.650 0.855
Z0 9.485 10.244 10.228 8.104
QCM
𝜔
0.460 0.251 1.403 0.577
𝑘𝑇
Bij(Λij) 0.761 0.854 0.802 1.154
Bji(Λji) 1.174 1.125 2.041 0.616
∆εij(τij) 0.020 0.013 0.372 0.003
MIVM ∆εji(τji) 0.186 0.110 0.012 0.413
NRTL Zi[25] 10.500 10.500 10.900 11.500
Wilson Zj[25] 10.400 10.400 6.400 10.900
Vi[26] 47.211 73.059 18.428 11.388
Vj[26]
24.920 47.211 11.911 17.692
cm3/mol

Haghtalab and Flemer respectively pointed out that the models parameters and components
of liquid alloys has function relation. Haghtalab fitted the parameters of NRTL, Wilson,
UNIQUAC with the components of mixtures by using third order polynomials. Fig.4 is the
relation between liquid alloys components and the models parameters, the parameters fluctuate
significantly. It is difficult to obtain the accurate relation only through to observe Fig.4, thus
correlation analysis is necessary. The correlation values that calculated by Eqs.(21) are listed in
Table 5. The change of correlation values in Table 5 are complex, the same parameters are
significantly different in different liquid alloys, only describe the overall characteristics of each
model parameter, in next analysis. The w and ω are analyzed together due to both they are
interchange energy, among the 8 values, only the ρw of Cs-K is 0.132 belong to weak correlation,
while the other seven’s are greater than 0.6 belong to strong correlation. Basically, it can be
conclude that the exchange energy parameter calculated based on g1ij(r) is function of
components, which consists with Haghtalab's work. That the ρZ of K-Na, Cs-K and Al-Sn are
0.007, 0.284, 0.497 respectively, indicates ρZ is weak correlation. The ρZ of Mg-Si is 0.99 belong
to strong correlation, because the gij(r) of Mg-Si affected by both temperature and components,
and the ρZ tend to reflect the relation of Z and T. There are a weak correlation between the Z and
the components of alloys, and the Z can be considered is characterize of differ type alloy. The ρBij,
ρBji of K-Na, Cs-K and Mg-Si are moderate correlation, while the value of Al-Sn is weakly
correlated. correlation analysis combine with e, f in Fig.4 can known that Al-Sn Bij relative to the
other three alloy, and obvious fluctuations in the components of regularity, that could be gij(r) and
the accuracy of numerical calculation method of swings, in a follow up study is required to ensure
that obtaining more accurate g1ij(r) cases for further study. The correlation of ∆εij and ∆εji of
MIVM is obvious, the correlation coefficient values are generally greater than 0.8, but in Mg-Si
and Al-Sn, one value is strongly correlated, the other value is weak correlation. For example in
Mg-Si, ∆εji, ∆εij are 0.779, 0.019 respectively. theoretically, it is unlikely that indexes with the

Electronic copy available at: https://ssrn.com/abstract=4121941

same meaning are extreme in the same system. Combined with g and h in Fig.4, it can be seen that
the ∆εij, ∆εji of Mg-Si and Al-Sn fluctuates obviously, the ∆εij fluctuation decreases and then
increases and ∆εji drops monotonically in Mg-Si, the ∆εij drops monotonically and ∆εji drops and
increases again in Al-Sn,which results show that ∆εji and ∆εij are affected by the components in
Mg-Si and Al-Sn, the correlation value is low due to non-monotone variation. By analyzing the
correlation of each model parameter in four alloys, it can be found that each parameter has
abnormal value in a certain alloy, which makes the correlation of description parameters
uncertain. The exact value of the correlation is affected by numerical calculations methods, but the
overall trend should be related to the gij(r) of the different components of the alloy when the
numerical calculation methods are stable enough, the Mg-Si ρZ is 0.99 obviously different from
that of other alloys due to the influence of temperature. Similarly, the correlation between the
MIVM parameters of Al-Sn and the other four alloys is obviously different. It can be used to
check whether the variation trend of Al-Sn alloy g1ij(r) is consistent with the other three alloys
under the influence of components which can be used to verify the accuracy of numerical
calculation. The correlation analysis of parameters and components provide very useful
information for us to realize the characteristic changes between parameters and components and
the possible problems of the proposed method, which should be pay attention to in our work.
11 0.6 1.6 2.5
(a):Z K-Na K-Na
(b):w Cs-K (e):B ij (f):Bji
10 Cs-K
0.4 Mg-Si 1.4 2 Mg-Si
Al-Sn Al-Sn
9
0.2 1.2 1.5
8
Bij

Bji
w
Z

0 1 1
7

6 -0.2 0.8 0.5

1.4
5 -0.45 0.6
0.4 0.8
0
(d):ω (g):Δε ij (h):Δε
(c):β ji
1.2 0.2 0.6
0 0.4
1 0
Δεij

Δεji

0.2
ω
β

0.8 -5 -0.2
0
0.6 -0.4 -0.2
-10
0.4 -0.6 -0.4
0 0.5 1 0 0.5 1 0 0.5 1 0 0.5 1
x2 x2 x2 x2

Fig.4 The relation between Model parameters and alloys component

Table 5 Model parameter correlation for five models
Mode Parameter K-Na Cs-K Mg-Si Al-Sn
𝑤
RSM 0.651 0.132 0.675 0.706
𝑘𝑇

Z 0.007 0.284 0.990 0.497

QCM
𝜔
0.868 0.891 0.917 0.736
𝑘𝑇
Bij(Λij) 0.909 0.567 0.619 0.265
MIVM Bji(Λji) 0.618 0.613 0.737 0.411
(Wilson
∆εij(τij) 0.967 0.923 0.019 0.885
NRTL)
∆εji(τji) 0.959 0.845 0.779 0.241
weak correlation 0~0.4;medium correlation 0.4~0.6; strong correlation 0.6~0.8; high strong

Electronic copy available at: https://ssrn.com/abstract=4121941

correlation 0.8~1.
n

 X i  X Yi  Y 
Y  i 1
（21）
n n

 X  X  . Yi  Y 
2 2
i
i 1 i 1

Although the relation between the models parameters and the components of mixture are
analyzed on above, constructing specific function expression for parameter not mature enough,
and the models parameters as constant value are convenience. It is necessary to do discrete
analysis when treating a variable values as constant values, treating as constant value is
appropriate, if the lower discrete degree, otherwise, the effect is poor. It is inappropriate to use SD
as the index of discrete degree when there are orders of magnitude differences between sets of
data. using coefficient of variation CV represent discrete degree. Coefficient of variation is the
ratio of standard deviation to the mean, CV= STD(Y)/mean(Y), the CV values of models
parameters listed in table 6. That the CVw values of the four alloys in the range from 0.20 to 0.9
belong to strong variation. It can known that the w changed significantly with the component,
which as constant value are not a beast choice. The CVZ values of K-Na, Cs-K and Al-Sn were
2%, 1.9% and 5.8%, respectively, belong to weak variation, the CVZ values of Mg-Si is 19% at
medium variation. The correlation coefficient analysis show that the ρZ were weakly correlated,
and the CV of Z was also low, which further proved that the Z treatment as constant values is

appropriate，Z have the potential to be an essential characteristic of the liquid alloy. That the CVω

range of the four alloys range from 2 to 104 are strong variation, the ω treatment as constant value
not conform the actual situation, also contrary to the constant treatment of Dorini, the Dorini’s
method need further study. The values of CV∆εij and CV∆εji of the four alloys all have a significant

difference，one value in the range of moderate variation and another value is strong variation. The

universality of this result are difficult to explain at present, and the basic information shown by it
can be judged whether the treatment as a constant is the best choice under the condition that the
components of ∆εij and ∆εij are obviously affected by the components and the dispersion is large.
The coefficient of variation shows that the discrete of Z,

Table 6 The Coefficient of variation for five model parameter

Mode Parameter K-Na Cs-K Mg-Si Al-Sn

𝑤
RSM 0.204 0.321 0.466 0.888
𝑘𝑇
Z 0.021 0.019 0.205 0.058
QCM
ω 103.721 15.282 2.065 4.75
Bij(Aij) 0.036 0.027 0.054 0.096
MIVM Bji(Aji) 0.045 0.023 0.109 0.197
(Wilson NRTL） ∆εij(τij) 2.866 3.101 0.208 70.663
∆εji(τji) 0.412 0.452 12.2 0.383

Electronic copy available at: https://ssrn.com/abstract=4121941

 Coefficient of variation division： <0.1 weak variation; 0.1~1 Medium variation; Strong

variation >1.
Bij, Bji are small, and the constant treatment is appropriate for the whole component. The model
parameters, ω, ∆εij, ∆εji, belong to strong variation, and the constant treatment does not conform to
the actual situation. The coefficient of variation table in Table 6 is given to realize that the
constant treatment needs mathematical basis in the calculation of thermodynamic model. The
larger the CV, the more problems will be caused by constant treatment.

3.3 Estimate thermodynamic values

The excess Gibbs energy of K-Na, Cs-K, Mg-Si, Al-Sn are estimated by to substitute the
parameters in Table 4 into the excess Gibbs energy modes. Fig.5 show the estimation and
experimental of four liquid alloys, the estimation trend of MIVM, RSM and experimental value
are relatively close, while the QCM, Wilson and NRTL can only estimate the general trend. The
RSM ,MIVM, NRTL use the same method calculating pair potential, that the estimation of RSM
and MIVM are better than NRTL indicate that pair potential method is reasonable and the problem
of NRTL theory and the parameter calculation method not fully reasonable. Prausnitz think that
the parameters of NRTL were similar to Wilson's and in Haghtalab's thesis ,the parameter of
Wilson and NRTL are consistent[4, 27]. The preliminary experimental of this work is showed that
when both NRTL and Wilson used volume parameters, NRTL estimation is worse. The Wilson,
NRTL, MIVM compare with together, the MIVM contains two term, the volume term form as
same as Wilson and the energy term form similar to NRTL, while MIVM estimation are best,
because the structure theory of MIVM is more prefect than the Wilson and NRTL, MIVM
naturally has two contributions of excess entropy and excess enthalpy in the process of model
derivation. The RSM and QCM compare with together, theoretically,the QCM derived from RSM
more realistic physical structure, but the RSM estimation better than QCM. One of reasons is
obtaining parameter method different. In table 3, the parameter values of two model are different,
and the preliminary experiments show that the two model parameter methods cannot be used
interchangeably, which means that two models relation need deep research. Calculating model

Electronic copy available at: https://ssrn.com/abstract=4121941

 800
300 quasi-chemical
(a):K-Na (b):Cs-K MIVM
250 regular solution
600 Wilson
200 NRTL
experiment
400
G
150
K-Na:Hultgren
100 Cs-K:SGTE
200 Al-Sn:Hultgren
50 Mg-Si:SGTE

4000
0 0
(c):Al-Sn
-2000
3000
-4000

2000 -6000
G

-8000
1000
-10000
(d):Mg-Si
0 -12000
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
x2 x2

Fig.5 Excess Gibbs energy in the estimated and experimental contrast diagram for K-Na[28],Cs-
K[29],Al-Sn[28],Mg-Si[29]
parameters from structural parameters enlarge the advantages and disadvantages of the model,
which the regression method can not do it.
The excess Gibbs energy describe the energy changing follow with components mixture, the
fluid phase balance also needs to know the activity. The activity of each alloy is calculated
according to Eqs.(22). It should be mentioned that the parameters β and Z of the QCM are
components function, and require do partial derivatives. In order to keep the formula simple in the
practical application, the technicians treat the variable coefficients in the model as constants, the
methods will make the γi and γi cannot keep the thermodynamic consistency with GE. In the case
of known analytic formula of coefficient and component, the calculation should be carried out in
strict accordance with the rule of derivation. This paper does not suggest simplifying the variable
coefficient in theoretical research. Fig.5 is the activity of experiment and estimationfor four liquid
alloys. The estimation of MIVM and RSM are reasonable in all system, QCM is poor performance
in Al-Sn, Wilson ans NRTL estimation worse than the first three models, and have reasonable
estimation only in K-Na and Cs-K system. The activity estimation used Wilson have the
characteristic,that estimation value of i and j is significantly different, the aj was within a
reasonable range in Al-Sn and Mg-Si system, while the ai is abnormal, and In K-Na and Cs-K, the
ai estimation result is better than aj.

ln  i 
GiE GmE  GmE RT
 

 xi
 GmE RT 
；ai  x i  i
 
RT RT xi xi
（22）
G E G E  GmE RT
ln  j  2  m   xj
 GmE RT 
；a j  x j  j
 
RT RT x j x j

Electronic copy available at: https://ssrn.com/abstract=4121941

 1
quasi-chemical
0.8 MIVM
Quasi-lattice
Activity 0.6 Wilson
NRTL
0.4 Experiment
K-Na:Hultgren
0.2 (a):K-Na (b):Cs-K Cs-K:SGTE
10
Al-Sn:Hultgren
Mg-Si:SGTE
0.8
Activity

0.6
(d):Mg-Si
0.4

0.2 (c):Al-Sn

0
0 0.2 0.4 0.6 0.8 01 0.2 0.4 0.6 0.8 1
x2 x2

Fig.6 activity in the estimated and experimental contrast diagram for AlSn,CsK,KNa,MgSi
In order to further analyze the estimation performance for four liquid alloys,Table 7 lists the
average relative deviation of activity and excess Gibbs energy. The QCM, MIVM and RSM
estimability at the same level and, the ARD of Wilson and NRTL for excess Gibbs energy are
larger. The ARD of excess Gibbs energy generally bigger than the ARD of activity
relative,because the free energy and activity are different orders of magnitude cause the ARD
amplifier. The ARD indicates that the excess Gibbs energy and activity of QCM, MIVM and
RSM all have estimability, and the follows are: RSM, MIVM, QCM. When the g1ij(r)/model
parameters methods are extended to Wilson and NRTL, both of them perform worse than the first
three methods, but the RAD of the K-Na, Cs-K and Al-Sn is 5.5~22 mean that both of them also
have the estimability. RSM, MIVM, Wilson and NRTL can jointly demonstrate the rationality of
the g1ij(r)/model parameters presented in this paper.
Table7 The RAD for five model parameter
QCM MIVM RSM Wilson NRTL
Alloy
Energy Activity Energy Activity Energy Activity Energy Activity Energy Activity

K-Na 43.37 11.64 23.18 6.57 15.16 4.44 74.98 18.48 67.65 16.87
Cs-K 22.87 2.93 34.03 3.40 50.99 4.93 72.35 11.20 55.52 5.55
Mg-Si 16.14 13.42 39.65 21.88 21.09 18.92 54.34 136.71 65.21 63.85
Al-Sn 49.06 17.88 20.99 10.14 12.04 6.77 65.04 21.93 57.60 20.18
average 32.86 11.47 29.46 10.50 24.82 8.76 66.68 47.08 61.49 26.61
(24)
1 Ypre  Yexp
ARD% 
N
 Yexp
*100%

Except for QCM, the excess energy and activity RAD of the others four models in Mg-Si
system are abnormal compared with other alloys’s, the Wilson RAD of activity is 136.7%,
obviously incomprehensible, needs to be explained. Examining the experimental data of excess
Gibbs energy and activity coefficient of the four alloys and the results find out the experimental
data in Mg-Si and Al-Sn literatures did not fully accord with each other. Fig.7 and 8 are activity

Electronic copy available at: https://ssrn.com/abstract=4121941

and excess energy data verification of Mg-Si and Al-Sn respectively. In Fig.7, the experimental
activity coefficient of Mg-Si and excess Gibbs energy do not meet the relation of molar and partial
molar. In Fig.8, the activity coefficient does not meet the Gibbs-Duhem equation. We have the
reason to suspect that the estimation performance of Mg-Si and Al-Sn are affected by the accuracy
of experimental values, and the inconsistency between Mg-Si activity and free energy is quite
significant. From finding abnormal estimation of Mg-Si to checking problems in experimental
data of Mg-Si, the scientific of AIMD(g1ij(r))/(MIVM, RSM, Wilson, NRTL,QCM) and it’s
ability to provide guidance for thermodynamic research are proved.
Mg-Si: Consistency test of molar and partial molar Mg-Si:Activity coefficient consistency test
0 2

1.5
-5000

-10000
0.5

-15000 0
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1

Fig.7 Consistency test of Thermodynamic data of Mg-Si

Al-Sn: Consistency test of molar and partial molar Al-Sn:Activity coefficient consistency test
3000 0.4

2500 0.2

2000 0

1500 -0.2

1000 -0.4

500 -0.6

0 -0.8
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1

 Fig.8 Consistency test of Thermodynamic data of Al-Sn

4 Conclusions
In this study, the model parameters of MIVM and RSM are calculated by using the only input
parameter gij(r) to predict the excess Gibbs energy and activity of of liquid alloys，the ARD of
excess Gibbs energy and activity is less than ± 30%, and the ARD of activity is less than ±
20%. The g1ij(r)/parameter methods extended to Wilson and NRTL estimability are not good as
MIVM and RSM, but the RAD still at reasonable range. Dorini’s gij(r)/QCM as contrast methods
compare with our work, the RAD of excess energy and activity of four QCM systems is ±32.862%

Electronic copy available at: https://ssrn.com/abstract=4121941

and ±11.468%, respectively. gij(r)/QCM estimability at same level with gij(r)/(MIVM, RSM).
It is also necessary to further improve the calculation performance and reduce the prediction
deviation. At present, the reasons for the large RAD of estimation is that research have point out
models parameters is the function of liquid alloys components, but still consider as constant
values, because the accurate function relation cannot be accurately described. The order of
magnitude of model parameters are small, and the gij(r) of alloy and the numerical fluctuation
caused by the numerical processing method have a significant influence on the parameter values.
The gij(r) obtained by AIMD should enough accurate and the numerical processing methods
should enough stable, only satisfy the above two requirements can to reflect the (gij(r))/(QCM,
MIVM, RSM) real estimability. The inconsistency between thermodynamic model theory and the
g1ij(r) structure theory also is an important reason to influence estimability. The comparison
between the RSM and QCM, MIVM and Wilson, NRTL, both of them clearly show that the
theoretical deficiencies of the models can make estimation worse, Changing the models theory.
Modifying model appropriately based on model theory and g1ij(r) may be one the research
direction to improve the estimability In liquid computer simulation experiments, the suitable
molecular pair potential function is an important issues. Using the g1ij(r) to express ε1ij(r) are
reasonable in practical application, but from the theoretical point of view that the expression is
quite rough, which possible a the important reason affecting the model estimability. In general, the
estimation accuracy of AIMD(g1ij(r))/(QCM, MIVM, RSM) can be further improved, and it is
possible to achieve the consistent accuracy of thermodynamic data obtained in laboratory.

Acknowledgements
This work was financially supported by the National Natural Science Foundation of China
under Grant (No. 51464022).

Appendix
The equation parameter of g1ii(r), g1jj(r), g1ij(r) in four alloys Table 1, 2, 3, 4
Table 1 K-Na
component xk
parameter
0.1 0.2 0.3 0.4 0.5 0.5(cal) 0.6 0.7 0.8 0.9
i-i 4.423 4.448 4.464 4.458 4.442 4.567 4.437 4.453 4.577 4.528
rm/Å i-j 4.047 4.030 4.025 4.075 4.065 4.133 4.047 4.039 4.077 4.117
j-j 3.650 3.644 3.647 3.656 3.644 3.604 3.644 3.601 3.641 3.611
i-i 2.716 2.676 2.617 2.638 2.639 2.376 2.630 2.612 2.706 2.533
g(rm) i-j 2.503 2.554 2.608 2.707 2.673 2.241 2.737 2.768 2.961 2.769
j-j 2.730 2.846 2.961 3.263 3.173 3.591 3.307 3.491 3.737 3.884
i-i 0.264 0.306 0.326 0.297 0.294 0.368 0.267 0.273 0.330 0.333
v i-j 0.230 0.213 0.205 0.261 0.258 0.302 0.223 0.212 0.208 0.270
j-j 0.188 0.192 0.197 0.215 0.214 0.188 0.208 0.170 0.201 0.183
i-i 0.401 0.356 0.341 0.393 0.361 0.644 0.430 0.440 0.360 0.460
u i-j 0.338 0.349 0.363 0.283 0.289 0.416 0.347 0.361 0.357 0.343
j-j 0.291 0.275 0.258 0.236 0.233 0.368 0.234 0.265 0.195 0.270

Electronic copy available at: https://ssrn.com/abstract=4121941

 Table 2 Cs-K
component xCs
parameter
0.1 0.2 0.3 0.4 0.5 0.5(cal) 0.6 0.7 0.8 0.9
i-i 5.155 5.176 5.152 5.167 5.206 5.299 5.179 5.182 5.226 5.199
rm/Å i-j 4.840 4.870 4.864 4.873 4.886 4.955 4.858 4.892 4.910 4.877
j-j 4.525 4.551 4.565 4.555 4.522 4.561 4.531 4.535 4.567 4.518
i-i 2.497 2.410 2.392 2.379 2.368 2.373 2.373 2.350 2.313 2.317
g(rm) i-j 2.320 2.366 2.387 2.404 2.418 2.476 2.463 2.464 2.417 2.440
j-j 2.450 2.572 2.548 2.575 2.747 2.813 2.810 2.881 2.686 3.004
i-i 0.465 0.444 0.425 0.449 0.502 0.456 0.445 0.454 0.502 0.447
v i-j 0.370 0.401 0.387 0.368 0.405 0.365 0.382 0.419 0.421 0.410
j-j 0.322 0.356 0.350 0.342 0.328 0.327 0.342 0.332 0.354 0.345
i-i 0.587 0.628 0.701 0.693 0.638 0.508 0.722 0.764 0.818 0.839
u i-j 0.667 0.551 0.659 0.681 0.616 0.378 0.626 0.591 0.717 0.746
j-j 0.584 0.457 0.569 0.594 0.504 0.437 0.488 0.509 0.681 0.510

Table 3 Mg-Si
component xMg
parameter
0.1 0.2 0.3 0.4 0.47 0.6 0.6(cal) 0.67 0.8 0.9
i-i 3.243 3.178 3.190 3.214 3.216 3.162 3.291 3.143 3.084 5.155
rm/Å i-j 2.769 2.785 2.797 2.764 2.785 2.735 2.861 2.741 2.713 4.840
j-j 2.448 2.454 2.461 2.431 2.449 2.461 2.468 2.466 2.444 4.525
i-i 1.498 1.420 1.459 1.639 1.811 1.910 1.897 1.892 2.204 2.497
g(rm) i-j 1.851 2.143 2.311 2.556 2.713 2.708 2.581 2.889 3.091 2.320
j-j 2.239 2.328 2.517 2.641 2.713 2.696 2.195 2.507 2.172 2.450
i-i 0.253 0.213 0.228 0.220 0.206 0.185 0.271 0.196 0.151 0.465
v i-j 0.099 0.108 0.115 0.088 0.099 0.077 0.146 0.082 0.076 0.370
j-j 0.060 0.057 0.058 0.038 0.050 0.056 0.057 0.058 0.048 0.322
i-i 0.494 0.552 0.516 0.488 0.370 0.368 0.372 0.402 0.328 0.587
u i-j 0.426 0.293 0.244 0.254 0.223 0.226 0.241 0.194 0.158 0.667
j-j 0.125 0.129 0.088 0.093 0.072 0.085 0.126 0.057 0.074 0.584

Table 4 Al-Sn
parameter component xAl

Electronic copy available at: https://ssrn.com/abstract=4121941

 0.25 0.5 0.5(cal) 0.65 0.8 0.87 0.92 0.97
i-i 2.509 2.517 2.539 2.501 2.535 2.658 2.707 2.767
rm/Å i-j 2.781 2.805 2.903 2.941 2.898 2.957 2.898 2.903
j-j 3.13 3.13 3.194 3.103 3.136 3.117 3.119 3.094
i-i 2.177 1.998 2.107 2.011 1.876 1.921 2.015 2.371
g(rm) i-j 1.614 1.644 1.808 1.567 1.805 2.399 2.695 3.056
j-j 2.268 2.603 2.509 2.893 3.343 3.539 3.526 3.785
i-i 0.079 0.083 0.082 0.079 0.093 0.172 0.15 0.139
v i-j 0.133 0.148 0.155 0.198 0.121 0.09 0.083 0.058
j-j 0.131 0.1 0.126 0.084 0.108 0.077 0.102 0.092
i-i 0.273 0.292 0.178 0.428 0.488 0.269 0.287 0.189
u i-j 0.433 0.393 0.416 0.326 0.297 0.232 0.188 0.174
j-j 0.36 0.327 0.272 0.275 0.351 0.3 0.154 0.109

The models parameters solved by g1ij(r) under the multicomponent for four alloys Table 5, 6,
7, 8
Table 5 K-Na
component xK
parameter
0.1 0.2 0.3 0.4 0.5 0.5(cal) 0.6 0.7 0.8 0.9
Zii 1.572 2.86 4.011 5.1 5.709 6.917 6.685 7.563 8.598 8.777
local
Zij 10.03 8.392 6.918 5.874 4.475 4.428 3.444 2.494 1.694 0.786
Coordiantion
Zji 1.114 2.098 2.965 3.916 4.475 4.428 5.166 5.818 6.775 7.07
number
Zjj 8.171 7.083 5.783 5.039 3.817 4.849 2.947 2.203 1.39 0.72
ε 𝒌𝑻
ii -0.5 -0.485 -0.464 -0.471 -0.474 -0.366 -0.468 -0.461 -0.55 -0.5
𝜀
ε 𝒌𝑻
ij -0.406 -0.439 -0.452 -0.507 -0.493 -0.308 -0.508 -0.519 -0.586 -0.519
𝑘𝑇
ε𝒋j 𝒌𝑻 -0.488 -0.55 -0.586 -0.683 -0.655 -0.779 -0.697 -0.751 -0.834 -0.858

Z 9.517 9.595 9.403 9.762 9.323 9.524 9.323 9.446 9.867 9.385
RSM
𝒘 𝒌𝑻 0.84 0.756 0.687 0.686 0.671 0.265 0.693 0.827 1.041 1.497

σ 0.194 0.332 0.427 0.476 0.486 0.434 0.459 0.391 0.298 0.165
QCM β 0.851 0.928 0.967 1.015 1.056 1.305 1.093 1.147 1.151 1.186
𝝎 𝒌𝑻 -6.889 -1.177 -0.384 0.149 0.511 2.744 0.859 1.511 2.019 3.543

Bij 0.709 0.734 0.739 0.768 0.758 0.641 0.773 0.769 0.788 0.806
Bji 1.227 1.185 1.196 1.166 1.181 0.9132 1.169 1.132 1.218 1.092
MIVM
∆εij 0.094 0.046 0.012 -0.036 -0.019 0.058 -0.04 -0.057 -0.036 -0.02

∆εji 0.082 0.111 0.134 0.176 0.175 0.471 0.189 0.233 0.247 0.339

Table 6 Cs-K
parameter component xCs

Electronic copy available at: https://ssrn.com/abstract=4121941

 0.1 0.2 0.3 0.4 0.5 0.5(cal) 0.6 0.7 0.8 0.9
Zii 1.351 2.619 3.791 4.881 8.011 5.524 6.545 7.752 8.778 9.149
local
Zij 10.056 8.446 7.602 6.336 2.791 4.389 3.867 2.793 0.965 0.894
Coordiantion
Zji 1.117 2.112 3.258 4.224 6.512 4.389 5.800 6.517 3.859 8.044
number
Zjj 8.745 7.387 6.588 5.624 2.563 4.56 3.527 2.563 1.226 0.832

ε 𝒌𝑻
ii -0.416 -0.381 -0.373 -0.368 -0.267 -0.365 -0.299 -0.302 -0.340 -0.339
𝜀
ε 𝒌𝑻
ij -0.338 -0.362 -0.371 -0.379 -0.385 -0.407 -0.402 -0.403 -0.221 -0.465
𝑘𝑇
ε𝒋j 𝒌𝑻 -0.397 -0.445 -0.426 -0.447 -0.511 -0.548 -0.534 -0.560 -0.479 -0.605

Z 10.016 9.812 10.310 10.395 9.938 9.64 9.978 6.132 8.811 9.926
RSM
𝒘 𝒌𝑻 0.685 0.502 0.296 0.291 0.041 0.049 0.144 0.170 1.665 0.074

σ 0.190 0.331 0.432 0.483 0.488 0.471 0.472 0.616 0.231 0.171
QCM β 0.894 0.936 0.945 0.986 1.049 1.122 1.036 0.363 1.770 1.109
𝝎 𝒌𝑻 -4.113 -1.048 -0.699 -0.149 0.478 1.079 0.365 1.087 1.567 1.866

Bij 0.827 0.806 0.860 0.865 0.835 0.7945 0.886 0.813 0.440 0.879

Bji 1.150 1.143 1.154 1.127 1.112 0.9621 1.096 1.089 0.787 1.075
MIVM
∆εij 0.078 0.019 0.002 -0.012 -0.016 -0.042 -0.103 -0.044 0.120 -0.126

∆εji 0.059 0.083 0.055 0.068 0.126 -0.141 0.132 0.157 0.258 0.141

Table 7 Mg-Si
component xMg
parameter
0.1 0.2 0.3 0.4 0.47 0.6 0.6(cal) 0.67 0.8 0.9

local Zii 1.054 1.857 2.759 4.070 4.722 6.071 6.955 7.099 8.119 8.909

Coordia- Zij 7.343 6.153 5.600 4.918 4.561 3.261 3.727 2.659 1.510 0.758

ntion Zji 0.816 1.538 2.400 3.279 4.044 4.891 5.591 5.399 6.039 6.825
number Zjj 3.931 3.453 2.883 2.384 2.061 1.696 1.613 1.175 0.612 0.358

εii 𝒌𝑻 0.089 0.148 0.134 0.001 -0.097 -0.149 -0.141 -0.142 -0.291 -0.345
𝜀
εij 𝒌𝑻 0.029 -0.263 -0.342 -0.439 -0.503 -0.494 -0.448 -0.554 -0.619 -0.665
𝑘𝑇
ε𝒋j 𝒌𝑻 -0.003 -0.345 -0.411 -0.472 -0.509 -0.497 -0.287 -0.432 -0.276 -0.638

Z 5.112 5.595 6.206 6.993 7.599 8.233 9.292 8.708 9.033 9.419
RSM
𝒘 𝒌𝑻 -0.067 -0.917 -1.264 -1.426 -1.524 -1.405 -0.234 -2.329 -3.032 -1.635

σ 0.242 0.400 0.519 0.566 0.582 0.507 0.52 0.454 0.307 0.166

QCM β 0.487 0.599 0.618 0.695 0.712 0.895 0.947 0.950 1.084 1.174
𝝎 𝒌𝑻 —— —— -4.122 -2.700 -2.595 -0.959 -1.621 -0.511 1.097 3.379

Bij 0.774 0.833 0.870 0.806 0.856 0.806 0.803 0.761 0.744 0.766

Bji 1.868 1.778 1.943 2.063 2.213 1.923 2.310 2.263 2.467 2.117
MIVM
∆εij -0.206 -0.410 -0.476 -0.440 -0.407 -0.345 -0.307 -0.413 -0.328 -0.320

∆εji 0.190 0.083 0.069 0.032 0.005 0.003 -0.161 -0.122 -0.343 -0.027

Table 8 Al-Sn
component xAl
parameter
0.25 0.5 0.5(cal) 0.65 0.8 0.87 0.92 0.97
local Zii 1.286 2.673 2.322 4.438 5.76 6.119 7.266 8.321

Electronic copy available at: https://ssrn.com/abstract=4121941

 Coordia- Zij 4.336 3.429 3.971 2.209 1.452 1.252 0.74 0.308

ntion Zji 1.445 3.429 3.971 4.103 5.806 8.378 8.51 9.967

number Zjj 7.188 5.708 5.397 4.198 3.289 2.102 0.965 0.397

εii 𝒌𝑻 -0.279 -0.192 -0.246 -0.2 -0.13 -0.156 -0.201 -0.368

𝜀
εij 𝒌𝑻 0.018 0.001 -0.093 -0.076 -0.094 -0.371 -0.492 -0.613
𝑘𝑇
ε𝒋j 𝒌𝑻 -0.326 -0.458 -0.421 -0.563 -0.708 -0.767 -0.767 -0.832

Z 7.881 7.62 7.954 7.226 7.588 7.776 8.124 8.682

RSM
𝒘 𝒌𝑻 0.321 0.272 0.268 0.426 0.334 0.083 -0.012 -0.019

σ 0.318 0.469 0.524 0.389 0.289 0.252 0.157 0.064

QCM β 1.356 1.134 0.909 1.339 1.217 0.798 0.877 0.833
𝝎 𝒌𝑻 2.956 0.958 0.762 2.264 2.125 -6.344 -6.275 ——

Bij 1.124 1.283 1.709 0.925 1.008 1.369 1.171 1.198

Bji 0.603 0.601 0.735 0.526 0.441 0.596 0.767 0.777
MIVM
∆εij 0.297 0.193 0.153 0.124 0.037 -0.215 -0.291 -0.245

∆εji 0.344 0.459 0.327 0.487 0.614 0.396 0.275 0.22

The consistency test of thermodynamic data in literature for four alloys Table 9, 10, 11, 12

Table 9 K-Na, T=773K

GmE RTΣxilnγi d ln  i d ln  j
xi xj ai aj γi γj xi xj
(J/mol) (J/mol) dxi dx j

1 0 1 0 1 2.438 0 0 -0.004 0
0.9 0.1 0.907 0.211 1.008 2.107 293 295 -0.143 -0.142
0.8 0.2 0.826 0.367 1.033 1.835 514 515 -0.266 -0.266
0.7 0.3 0.755 0.484 1.078 1.615 665 664 -0.368 -0.368
0.6 0.4 0.689 0.575 1.148 1.437 749 750 -0.442 -0.442
0.5 0.5 0.625 0.648 1.25 1.295 770 769 -0.482 -0.482
0.4 0.6 0.557 0.711 1.393 1.186 728 727 -0.483 -0.483
0.3 0.7 0.478 0.772 1.592 1.103 628 627 -0.441 -0.44
0.2 0.8 0.374 0.837 1.87 1.046 469 471 -0.35 -0.348
0.1 0.9 0.226 0.91 2.26 1.012 260 261 -0.204 -0.204
0 1 0 1 2.816 1 0 0 0 -0.008

Table 10 Cs-K, T=773K

GmE RTΣxilnγi d ln  i d ln  j
xi xj ai aj γi γj xi xj
(J/mol) (J/mol) dxi dx j

Electronic copy available at: https://ssrn.com/abstract=4121941

 1 0 1 0 1 1.595 0 0 -0.001 0
0.9 0.1 0.906 0.139 1.007 1.394 111 112 -0.124 -0.12
0.8 0.2 0.821 0.251 1.026 1.253 185 186 -0.188 -0.186
0.7 0.3 0.738 0.347 1.056 1.156 226 231 -0.208 -0.208
0.6 0.4 0.652 0.436 1.087 1.09 239 239 -0.195 -0.197
0.5 0.5 0.562 0.523 1.124 1.046 229 229 -0.16 -0.163
0.4 0.6 0.464 0.612 1.159 1.02 201 200 -0.113 -0.115
0.3 0.7 0.357 0.704 1.189 1.006 159 159 -0.064 -0.063
0.2 0.8 0.242 0.8 1.208 1 108 107 -0.022 -0.020
0.1 0.9 0.121 0.9 1.213 1 53 55 0.003 0.004
0 1 0 1 1.199 1 0 0 0 -0.005

Table 11 Mg-Si, T=2200K

GmE RTΣxilnγi d ln  i d ln  j
xi xj ai aj γi γj xi xj
(J/mol) (J/mol) dxi dx j

1 0 1 0 1 0.059 0 0 0.001 0
0.9 0.1 0.983 0.021 0.984 0.078 -3031 -4931 0.353 0.354
0.8 0.2 0.869 0.054 0.914 0.117 -5628 -9164 0.898 0.893
0.7 0.3 0.69 0.11 0.781 0.186 -7599 -12394 1.429 1.423
0.6 0.4 0.496 0.2 0.607 0.297 -8810 -14361 1.796 1.794
0.5 0.5 0.324 0.334 0.431 0.452 -9185 -14959 1.910 1.915
0.4 0.6 0.195 0.509 0.285 0.633 -8708 -14202 1.746 1.751
0.3 0.7 0.108 0.704 0.183 0.805 -7421 -12096 1.339 1.338
0.2 0.8 0.055 0.877 0.119 0.93 -5425 -8848 0.791 0.779
0.1 0.9 0.022 0.986 0.085 0.989 -2875 -4691 0.267 0.261
0 1 0 1 0.072 1 0 0 0 0.052

Table 12 Al-Sn, T=973K

GmE RTΣxilnγi d ln  i d ln  j
xi xj ai aj γi γj xi xj
(J/mol) (J/mol) dxi dx j

1 0 1 0.001 1 6.639 0 0 0.352 0

Electronic copy available at: https://ssrn.com/abstract=4121941

0.9 0.1 0.925 0.384 1.028 3.845 12932 1290 -0.527 -0.471
0.8 0.2 0.883 0.511 1.104 2.553 21598 2156 -0.769 -0.697
0.7 0.3 0.853 0.569 1.219 1.894 2670 2671 -0.757 -0.765
0.6 0.4 0.821 0.61 1.368 1.526 2888 2888 -0.689 -0.738
0.5 0.5 0.779 0.65 1.558 1.301 2858 2857 -0.633 -0.659
0.4 0.6 0.714 0.698 1.784 1.163 2607 2606 -0.575 -0.553
0.3 0.7 0.615 0.756 2.049 1.080 2176 2176 -0.469 -0.430
0.2 0.8 0.473 0.823 2.365 1.029 1578 1577 -0.289 -0.289
0.1 0.9 0.274 0.902 2.744 1.002 832 831 -0.077 -0.124
0 1 0.0001 1 2.706 1.000 0 0 0 0.077

Estimating the Excess Gibbs energy and component activities by own’s AIMD g1ij(r)
800 600 K-Na:Hultgren
(a):K-Na (b):Cs-K
Cs-K:SGTE
500
600 Al-Sn:Hultgren
400 Mg-Si:SGTE

400 quasi-chemical
G

300
MIVM
200 Quasi-lattice
200 Wilson
100 NRTL
experiment
30000 0
(c):Al-Sn
2000
-5000
1000
G

0
-10000
-1000
(d):Mg-Si
-2000 -15000
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
x2 x2

Fig.1 Excess Gibbs energy in the estimated and experimental contrast diagram for K-Na,Cs-
K,Al-Sn,Mg-Si

Electronic copy available at: https://ssrn.com/abstract=4121941

 1 K-Na:Hultgren
Cs-K:SGTE
Al-Sn:Hultgren

Activity
Mg-Si:SGTE
0.5 quasi-chemical
MIVM
Quasi-lattice
(a):K-Na (b):Cs-K
Wilson
0
1 NRTL
Experiment
Activity

0.5 (d):Mg-Si

(c):Al-Sn
0
0 0.2 0.4 0.6 0.8 1
0 0.2 0.4 0.6 0.8 1
x2 x2

Fig.2 activity in the estimated and experimental contrast diagram for AlSn, CsK, KNa, MgSi

Table13 The RAD of Energy and Activity for five model parameter
QCM MIVM RSM Wilson NRTL
Alloy
Energy Activity Energy Activity Energy Activity Energy Activity Energy Activity

Mg-Si 12.556 11.317 64.577 29.427 13.974 13.795 46.766 150.13 61.163 58.229
K-Na 35.183 9.707 16.838 4.903 11.617 3.458 71.130 19.025 66.940 16.731
Cs-K 53.531 5.06 34.033 7.793 49.242 4.762 24.885 7.7304 55.52 5.539
Al-Sn 66.544 22.489 3.748 3.603 8.301 5.035 96.565 27.746 57.6 21.001
average 41.9535 12.1433 29.799 11.431 20.783 6.762 59.157 51.157 60.305 25.375

Reference

[1]A. K. Sum, S. I. Sandler, Use of Ab Initio Methods to Make Phase Equilibria Predictions Using
Activity Coefficient Models, Fluid Phase Equilibr., 158 (1999) 375-380.
[2]G. Raabe, J. Koehler, Use of Ab Initio Interaction Energies for the Prediction of Phase Equilibria
in the System Nitrogen-Ethane, Phys. Chem. Chem. Phys., 4 (2002) 926-930.
[3]M. Neiman, H. Cheng, A Critical Assessment On Two Predictive Models of Binary Vapor Iquid
Equilibrium, Phys. Chem. Chem. Phys., 6 (2004) 3474-3483.
[4]A. Haghtalab, J. B. R, Y. Seyf, A New Insight to Validation of Local Composition Models in
Binary Mixtures Using Molecular Dynamic Simulation, AIChE J., 1 (2015) 275-286.
https://doi.org/10.1002.
[5]A. Guggenheim E., The Theory of the Equilibrium Properties of some Simple Classes of Mixtures,
Solutions and Alloys., 1979.
[6]A. Ravichandran, R. Khare, C. Chen, Predicting NRTL Binary Interaction Parameters From
Molecular Simulations, AIChE J., 64 (2018) 2758-2769.
https://doi.org/https://doi.org/10.1002/aic.16117.
[7]A. Ravichandran, H. Tun, R. Khare, C. C. Chen, Prediction of Thermodynamic Properties of
Organic Mixtures: Combining Molecular Simulations with Classical Thermodynamics, Fluid
Phase Equilibr., 523 (2020) 1-10.

Electronic copy available at: https://ssrn.com/abstract=4121941

 [8]T. T. Dorini, L. Eleno, Liquid Bi–Pb and Bi–Li Alloys: Mining Thermodynamic Properties From

Ab-Initio Molecular Dynamics Calculations Using Thermodynamic Models, Calphad, 67 (2019)

1-9.
[9]Korkmaz, S. D. Korkmaz, Structure and Inter-Diffusion Coefficients of Liquid NaxK1x Alloys,
Journal of Phase Equilibria & Diffusion, 31 (2010) 15-21.

[10]S. D. Korkmaz, Korkmaz, A Study for Structure and Inter-Diffusion Coefficient of Liquid K1−X

Cs X Metal Alloys, Phys. Chem. Liq., 49 (2011) 801-810.

[11]D. Liu, X. Zhu, First-Principles Study of Chemical and Topological Short-Range Orders in the
MgSi Liquid Alloys, Metals - Open Access Metallurgy Journal, 6 (2016) 1-10.
[12]O. S. Roik, O. M. Yakovenko, V. P. Kazimirov, V. E. Sokol'Skii, N. V. Golovataya, Y. O.
Kashirina, Structure of Liquid AlSn Alloys, J. Mol. Liq., 330 (2021) 1-9.
[13]M. Parrinello, R. Car, Unified Approach for Molecular Dynamics and Density-Functional Theory,
Phys. Rev. Lett., 55 (1985) 2471-2474. https://doi.org/10.1103/PhysRevLett.55.2471.
[14]A. Gk, B. Jf, Efficiency of Ab-Initio Total Energy Calculations for Metals and Semiconductors
Using a Plane-Wave Basis Set, Comp. Mater. Sci., 6 (1996) 15-50.
[15]G. Kresse, D. Joubert, From Ultrasoft Pseudopotentials to the Projector Augmented-Wave Method,
Physical Review B, 59 (1999) 1758-1775.
[16]John, P., Perdew, Kieron, Burke, Matthias, Ernzerhof, Generalized Gradient Approximation Made
Simple, Phys. Rev. Lett., 77 (1997) 3865-3868.
[17]N. Shuichi, A Unified Formulation of the Constant Temperature Molecular Dynamics Methods,
The Journal of chemical physics, 81 (1984) 511-519.
[18]P. T. Dong, A New Model of Thermodynamics of Liquid Mixtures and its Application to Liquid
Alloys, Thermochim. Acta, 363 (2000) 105-113.
[19]H. Dai, T. D. Ping, A Statistical Thermodynamic Model with Strong Adaptability for Liquid
Mixtures, Fluid Phase Equilibr., 0378 (2018) 154-165.
[20]J. H. Hildebrand, Solubility. Xii. Regular Solutions1, J.am.chem.soc, 51 (1929) 66-80.
[21]Wilson, M. Grant, Vapor-Liquid Equilibrium. XI. A New Expression for the Excess Free Energy
of Mixing, J.am.chem.soc, 86 (2012) 127-130.
[22]H. Renon, J. M. Prausnitz, Local Composition Thermodynamic Excess Functions for Liquid
Mixtures, AIChE J., 14 (1968) 135-144.
[23]Guggenheim, A. E., Statistical Thermodynamics of Co-Operative Systems (A Generalization of the
Quasi-Chemical Method), Transactions of the Faraday Society, 44 (1948) 263-271.
[24]M. D. Johnson, P. Hutchinson, N. H. March, Ion-Ion Oscillatory Potentials in Liquid Metals,
Proceedings of the Royal Society A, 282 (1964) 283-302.
[25]Y. Waseda, The Structure of Non-Crystalline Materials : Liquids and Amorphous Solids, McGraw-
Hill International Book Co., (1980).
[26]T. Iida, R. Guthrie, The Physical Properties of Liquid Metals, Oxford University Press (1988).
[27]J. M. Prausnitz, R. N. Lichtenthaler, E. Azevedo, Molecular Thermodynamics of Fluid-Phase
Equilibria, 1999.
[28]R. R. Hultgren, Selected Values of the Thermodynamic Properties of Binary Alloys, American
Society for Met (1973).
[29]P. Franke, Neusch, Binary Systems. Part 3: Binary Systems From CsK to Mg-Zr, Binary systems.

Electronic copy available at: https://ssrn.com/abstract=4121941

Part 3: Binary Systems from CsK to Mg-Zr, 2005.

Electronic copy available at: https://ssrn.com/abstract=4121941