Korean J. of Chem. Eng.

, 10(2), 71-77 (1993)

CONTINUOUS THERMODYNAMICS OF PHASE EQUILIBRIA

USING THE BETA DISTRIBUTION FUNCTION
AND AN EQUATION OF STATE
Jiho Park and Hwayong Kim*

Hannam Chemical Corporation

*Department of Chemical Engineering, Seoul National University, Seoul 151-742. Korea
(Received 2 September 1992 9 a~eplcd 9 March 1993)

A b s t r a c t - P h a s e equilibrium calculation involving mixtures of very many components requires

significant amounts of computation time. In this paper we considered the description of such mixtures
as a collection of discrete components and a continuum distribution of species represented by the
beta distribution function. Calculations are performed using the method of continuous thermodynamics
and the Soave-Redlich-Kwong equation of state. Comparisons were made with the pseudocomponent
method and gamma distribution flmction. The results showed that continuous thermodynamics with
beta distribution function leads to better prediction of phase equilibria and shorter computing time
compared to pseudocomponents. And more reasonable results were obtained by this method because
of the flexibility of beta distribution function.

INTRODUCTION This approach is not a new one. In 1949, Bowman

Eli studied this method first, but his study was limi-
For multicomponent systems which contain very ted within Raoult's law. It was (;ualtieri et al. [2] that
many components having similar properties, there are introduced the equation of state in continuous ther-
generally two problems in predicting phase equilib- modynamics and Cotterman et al. ~3, 4j used Soave-
rmm: analysis and long computing trine. The similari- Redlich-Kwong equation of state to calculate phase
ties of physical and chemical properties of components equilibrium. They used the gamma distribution func-
do not permit them to be identified and analyzed. And tion and the molecular weight as a distribution varia-
if they were identified and analyzed, another problem ble. But they treated only a linear system in which
exists-king computing time because of very many com- the constants of the equation of state are linear func-
ponents inw~lved in the system. tions of molecular weight. Nonlinear system was trea-
Up to now two approximate methods have been ted by Shibata et al. [5] and the)' pointed out t w o
used to solve these problems-pseudocumponent and problems-physical infeasibility and mathematical incon-
key-component methods. These components, though sistency-contained in the continuous thermodynam-
different in definition, have been regarded to have ics. In addition Willman and Teja [6] used bivariate
the system properties and phase equilibrium calcula- log normal distribution function and Wang and Whir-
tions have been performed with them. But the ther- ring [7] introduced the activity model into the contin-
modynamic properties of the systems represented by uous thermodynamics.
these components differ from researcher to research- Beta distribution function was aheady used by ano-
er because they have chosen them with their expe- ther researchers ES]. Their method, however, is re-
rience and intuition. stricted in the PttC (Perturbed hard chain) theory. In
Continuous thermodynamics is an another systemat- this paper, we presented more general method to cal-
ic approach to calculate phase equilibrium. [n contin- culate phase equilibrium.
uous thermodynamics, the composition of the system The advantages of the continuous thermodynaluics
and the thermodynamic properties are expressed by are the facts that all the components involved in the
the distribution function and distribution variables. system can be considered and mathematical descrip-
tion of composition by distribution function decreases
*Author to whom correspondences should be addressed. the computing time.

7l
72 J. Park and H. Kim

CONTINUOUS THERMODYNAMICS THE EQUATION OF STATE

Thermodynamics that treats continuous mixture is In this paper, Soave-Redlich-Kwong (SRK) Ell]
called continuous thermodynamics. By continuous mix- equation of state was used to represent the behavior
ture. it is meant a mixture which contains very many of gases and liquids.
components having similar physical and chemical prop-
RT a
erties, and its composition changes continuously. P- (11)
v-b v(v~ b)
In continuous thermodynamics, the composition of
component can be represented by distribution func- a--as (12)
tion, e.g.
(RT,) 2
a. = 0.42747 - - - - (13)
E.
f F(I)dI 1 (i)
a E1 + m(1-Tl:~)] 2 (14)
For the more general case of the system containing
both discrete components and continuous mixture, the m = 0.480 + 1.574,:o - 0.1760-' (15)
composition relationship is
RT~
b = 0.08664 - - (16)
Ex, + q J" F(I)dI = 1 (2) P,
where F(I) is the distribution function, I is the distri- Since coefficient of Eq. (15) were developed by re-
bution variable, q is the total mole fraction of contin- gressing vapor pressure data of low molecular weight
uous mixture, and x, is the composition of i compo- hydrocarbon, it is unlikely that this equation will suf-
nent. Multivariate distribution function may be used fice for high molecular weight component. Sim and
to represent the composition of continuous mixture. Daubert E12~ determined new coefficients for the high
But in such a case the number of components to be molecular weight components as.
examined increases explosively [9], so does the com-
m - 0.315 + 1.60~o- 0.166m=' (17)
puting time. Therefore univariate distribution function
was used in this paper. It was known that equation of state modified by
When a system contains both continuous and dis- Sim and Daubert is more accurate than original SRK
crete components, general phase equilibrium conditions equation E9] when applied to the system containing
and mathematical relation were derived by Cotterman heavy fractions.
et al. [3,4] and Kehlen et al. [10].
DISTRIBUTION FUNCTIONS AND
T v - TL (3)
DISTRIBUTION VARIABLES
p~ - ff (4)
It was generally known that as the molecular weight
f~(i) ((i) (5) increases the composition of heavy fraction decreases
exponentially [5, 13-151. And components of Eea W
f~(I) = f"(l) (6)
fraction can be identified by two physical properties
where the fugacities of the discrete components in Imainly boiling point, T, and specific gravity, S.G.)
each phase are given as L16].
Based on these facts, distributkm functions invol-
f(i) = x,o(i)P (7)
ving exponential term was used and boiling poin]: and
and for the continuous components specific gravity were used as distribution variables in
previous studies E3-6].
f(I) = qF(Do(I)P (8) But Shibata et al. pointed out two problems, mathe-
And fugacity coefficients are given as follows matical inconsistency and physical infeasibility E5].
They suggested the method to overcome physical in-
RTlno(i)= ( ' [( o G ' - - R T ]dV RTlnz (9) feasibility. But their method was an approximate one,
J , L On, , i , V J and requires additional numerical procedure to calcu-
late quadrature points and weighting factors. Ancl the
RTlno(I) ('[(,~7,~,8G' RT]dv RTlnz (10) problem of mathematical inconsistency also reraains
. , L, oF,,,),~, v
because it can be solved only when the considered

A.pHI, 1993
 Continuous Thermodynamics Using the Beta Distribution Function 73

ul_
L.L

Distr;bution vorloble Distribution variable

(a) Gamma distribution function (b) Beta distribution functior,

Fig. I. Comparison of gamma and beta distribution function.

system is linear and binaw interactions are zeros.

obj. func. Y~[ s f(l)dI I /Zx, (19)
In this paper, physical infeasibility was overcome
by use of the beta distribution function. Numerical integration of beta and gamma distribu-
1 (I-a)* ~(b IY ~
F(I) B(p, q) (b_a)p-q i (18)
Table I. Experimental TBP data for hea~ oil [ 15]
(a) Pure component
The advantage of using beta distribution function
Component Mole fraction(q )
is that the method suggested by Shibata to overcome
N, 0.34
physical infeasibility is not needed because beta dis-
CO, 0.84
tribution is bounded on both sides. In addition to this,
Methane 49.23
beta distribution function is more flexible than gamma
Ethane 6.32
distribution function (Fig. 1), so it seems that beta dis-
Propane 4.46
tribution function can represent the composition of
i-Butane 0.86
continuous mixture better than gamma distribution
n-Butane 2.18
func:ion.
i-Pentane 0.93
For example, at equilibrium, light components are
n-Pentane 1.33
richer than heavier ones in the vapor phase. In this
Hexane 2.06
case gamma distribution can represent the composi-
tion of this phase [Fig. l(a)] and so can the beta dis-
(b) Component of Cr-
tribution EFig. l(b) I]. In contrast gamma distribution
cannot exhibit the composition of liquid phase [Fig. 1 No. Mole fraction SP. GR. %,(c )
(a)] in which phase the heavier components are richer 1 3.33 0.7395 -100
while beta distribution function can do because of its 2 4.06 0.7518 100-127
flexibility [Fig. l(b) III]. 3 2.76 0.7756 127-152
4 1.33 0.7930 152-176
RESULTS 5 1.79 0.7902 1'76-197
6 1.70 0.8060 197-218
1. The C o m p a r i s o n of Continuous T h e r m o d y n a m - 7 1.81 0.8203 218-237
ics and P s e u d o c o m p o n e n t Method 8 1.46 0.8311 237-255
In this paper, beta distribution function was used 9 1.49 0.8446 255-272
to express the composition of continuous mixture. The 10 1.08 0.8515 2'72-288
coefficients of distribution function were calculated by 11 1.13 0.8542 288-304
performing optimization calculation. The optimization 12 0.99 0.8561 304-318
method was Powell method [17] and used object func- 13 0.88 0.8663 318-332
tion is as follows 14 7.64 0.9350 332-

Korean J. Ch. E.(Vol. 10, No. 2)

74 J. Park and H. Kim

'3. :J C Table 3. The binaff interaction parameters used in this

paper
N._, CO.,
r
2~35 N, 0.000 - 0.030
c~ /
CO~, - 0.030 0.000
Methane 0.028 0.093
Ethane 0.041 0.136
Propane 0.076 0.129
r i-Butane 0.094 0.128
n-Butane 0.070 0.143
,) 75 i-Pentane 0.(187 0.131
n-Pentane 0.088 0.131
Hexane 0.150 0.118

: '., -: 9 4C1(" .: ( ; ~-" ' " :- ~: ' , ~)L,(i

0.01:~ '~ . . . . . . . . . .
]c

Fig. 2. Correlation of specific gravity ~'ith boiling point.

0.0~0

c~:

Table 2. Physical properties calculated for the case of 7 o C, ( ]C',~

quadrature points
No T,,(K)-=S~G_ M[:W:[:=:TTKi:[3~(atm)=:'=o,:
1 372.4 0.749 97.4 560.2 31.8 0.278
2 403.2 0 . 7 6 6 113.6 589.2 27.7 0.330
3 453.1 0 . 7 9 3 142.0 634.0 22.5 0.432
4 513.5 0 . 8 2 6 179.6 691.8 18.5 0.590
5 573.8 0 . 8 5 9 2 2 0 . 7 766.1 17.1 0.737 C D,i.?

6 623.7 0.886 257.2 856.9 l&9 0.762

7 654.5 0.903 2 8 0 . 9 926.5 22.3 0.713

'! , r ~:rl" [,el f ~ u < , 2 !r t.Jr ~ ;P

tion was performed by use of Gaussian quadrature Fig. 3. Calculated vapor composition of ( : ~ witlh contin-
method. The detailed method was found in [5]. uous thermodynamics (circle) and pseudocompo-
Experimental data used were for hea%' oil data per- nent (filled circle).
formed by Pedersen et al. -153. The result of true
boiling point experiment is shown in "Fable 1. In this
paper the composition of C , components was rcpre- Binary interaction parameters used in this paper are
sented by beta distribution function. Calculated para- listed in "Fable 3. In this paper, the binary imeraction
meters for beta distribution functkm are a = 364.8, b:- parameters between organic components, and between
662.1, p :-0.544, and q=0.665. discrete components and continuous mixture are as-
Generally at least two physical properties are requir- sumed to be zeros.
ed to identify a component from others in continuous In continuous thermodynamic theory, as the num-
mixture [16]. Mostly these are boiling pc,int (T,)and ber of quadrature points increases, the calculLated re-
specific gravity (S.G.). But data obtained showed that sult approaches a definite value. It is known that the
specific gravity can be linearly related to boiling point number of quadrature points required to converge is
(Fig. 2). So only boiling point was used as distribution from 6 to 10 -'l~. In case of heavy oil studied in this
variable. paper, the general trend shows the results are conver-
Needed parameters (critical temperature and pres- ged for the number of the quadrature points greater
sure, and acentric factor) were calculated with the data than 7 (Fig. 3).
shown in "Fable 1 bv use of Lin-Chao correlation ~18]. Calculated results are presented in Table 4. Both
In order to apply Lin-Chao correlation, the method of the obtained results by continuous thermodynamics
of Gray et al. ~19j were used and the results are and pseudocomponent method seem to be reasonable.
shown in Table 2 for the case of 7 quadrature points. As shown in Table 4, the components with relatively

April, 1993
 Continuous Thermodynamics Using the Beta Distribution Function 75

Table4. Comparison of the results from continuous thermodynamics with the pseudocomponcnt method
Gas phase Liquid phase
Compn Expt Cont Thermodyn Pseudocompn Expt Cont Thermodyn Pseudocompn
(~ dev _ I .
5"2dev c~dev c,~ de,,,
N=, 0.0059 0.0064 9.2 0.0065 9.56 0.0000 0.0004 0.0005
CO_, 0.0133 0.0112 15,7 0.0113 14.88 0.0035 0,0057 61.7 0.0056 60.87
C~ 0.8484 0.8537 0.6 0.8492 0.10 0.1099 0.1393 26.8 0.1538 39.97
C.e-C:~ 0.1113 0.1097 1.4 0.1072 3.71 0.0911 0.1060 16.4 0.1084 18.99
C4-C~ 0.0198 0.0170 14.1 0.0165 16.81 0.1210 0.1289 6.5 (I.1278 5.60
C:. 0.0013 0.0019 49.4 0.0094 619.50 0.6745 0.6198 8,1 0.6039 10.47

~?c F. . . . . . . . . of the hea W comt)onent much better.

Fig. 4 shows the relation between calculation time
i
I
and the number of quadrature points. Calculation time
by continuous thermodynamics is shorter than that
' O0 t
"3 o Q by pseudocomponent method. In particular calculation
- i time by pseudocomponent is twice as long as that by
u I
<,
if}
i
I ;Z
continuous thermodynamics when the number of quad-
,.'; R : i
.b i rature point is 7 where the calculation result conver-
! ? 7, : ges.
It is generally known that computing time increases
3 6;) ;2 0 2'
as the number of quadrature points considered and
the difference of properties of each component in-
99999 P5r ~,i,,r?~[ ,~'n~r'l crease. In Fig. 4, calculaion time increases stepwise.
04.'.' ........ : l . ~. . . . 1 ..... This is because the increase of the calculalion time
6 S ~(: Ib f." 2'., caused by the increment of the number of quadrature
The I,L.F'OF~@r ,'f "; i.Zl~l: T]) . , ' ' : " i'~ "-:', points is canceled a little bit by the decrement of the
Fig. 4. Elapsed time to calculate phase equilibrium accord- difference of the physical property between the com-
ing to the number of quadralure point. ponents.
2. C o m p a r i s o n of B e t a and G a m m a D i s t r i b u t i o n
Functions
small quantity in each phase show large relative er- In Table 5, beta and gamma distribution function
rors. Particularly pseudocomponent method fails to were compared. As shown in Table 5 both beta and
predict the composition of h e a w component in gas gamma distribution functions predicts liquid phase com-
phase though absolute error is very. small On the position reasonably. But beta distribution function
other hand. continuous thermodynamics can predicts represents the composition of vapor phase than gam-
the composition of the components in both phases bet- ma distribution function, and original gamma dislri-
ter than the pseudocomponent method. Especially con- bution function and bounded gamma distribution func-
tinuous thermodynamics can describe the behavior tion suggested by Shibata et al. make no difference

Table 5. Comparison of the beta and gamma distribution functions

(a) for vapor phase
Beta distribution Gamma distribution Gamma distribution
Expt
dev(%) Cotterman* dev(C/c) Shibata** dev((:,~)
N~ 0.0059 0.0064 9,2 0.0062 5.8 0.0063 7,8
CO., 0.0133 0.0112 15.7 0.0111 16.5 0.0110 17.1
C1 0.8484 0.8537 0.6 0.8518 0.4 0.8581 1.1
C:-C:~ 0.1113 0.1097 1.4 0.1145 2.9 0.1100 1.2
C4-C~ 0.0198 0,0170 14.1 0.0163 17.7 0.0146 26.2
Cr- 0.0013 0,0019 49.4 0.0000 100.0 0.0000 100,0
AARD 15.1 23.9 25.6

Korean J. Ch. E.(Vol. IlL No. 2)

76 J. Park and H. Kim

Table 5. Continued
(b) for liquid phase
Beta distribution Gamma distribution Gamma distribution
Expt
dev(%) Cotterman* dev(%) Shibata** dev(C~)
N2 0.0000 0.0004 0.0002 0.0002
CO~ 0.0035 0.0057 61.7 0.0053 53.7 0.0056 60.9
Cl 0.1099 0.1393 26.8 0.0090 17.3 0.1075 20.2
C~-C:~ 0.0911 0,1060 16.4 0.1003 10.1 0.1055 15.8
C4-C~ 0,1210 0,1289 6.5 0.1376 13.7 0.1357 12.1
C7+ 0.6745 0.6198 8.1 0.6656 1.3 0.6454 4.3
AARD 23.9 19.2 19.1

(c) for K-value

Beta distribution Gamma distribution Gamma distribution
Expt
dev(%) Cotterman* dev(%) Shibata** dev(%)
N~ 15.000 27.629 22.056
CO2 3.800 1.980 47.9 2.064 45.7 1.959 48.5
Cl 7.720 6.129 20.6 9.372 21.4 7.981 3.4
C~-C:~ 1.222 1.035 15.3 1.142 6.5 1.042 14.7
C4-C~ 0.164 0.132 19.3 0.119 27.6 0.108 34.2
C7, 0.002 0.003 62.5 0.000 100.0 0.000 100.0
AARD 33.1 40.2 40.1
*Continuous thermodynamic method by Cotterman
**Continuous thermodynamic method by Shibata

in both phase. As a whole gamma distribution function distribution function is bounded on both sides. And
predicts only the composition of h e a w oil rich phase, the flexibility of beta distribution function leads to
but beta distribution function can predicts the compo- more accurate results for vapor phase in which phase
sitions of poth phase. This is because beta distribution the heavy fraction is nearly zero so gamma distribu-
function is more flexible than gamma distribution, so tion function cannot predict its composition.
it can describe the composition of h e a w oil of both
phase. ACKNOWLEDGEMENT

CONCLUSION The authors gratefully acknowledge the financial

support of Yukong Ltd.
Compared to pseudocomponent method, continuous
thermodynamics can predict phase equilibrium more NOMENCLATURE
accurately because it can count the h e a w fraction by
using the distribution function. And in this paper, we attraction energy parameter in SRK equation of
showed that only one distribution variable is sufficient state
to characterize the h e a w fossil fuel when its physical attraction e n e r ~ parameter in SRK equation of
or chemical properties are interrelated each other like state at critical point
in this case. So it can reduce the computing time great- a,b parameters used in beta distribution function
ly. b volume parameter in SRK equation of state [cm :~]
in this paper, beta distribution function was used. B beta function
The advantages of using beta distribution function are F : distribution function
the facts that the physical infeasibility problem which G :Gibbs free e n e r ~
arose in the case of gamma distribution tunction or I : distribution variable
other distribution functions containing exponential te- m : parameter defined in Eq. (15)
rms was overcome and therefore the method suggest- n :mole n u m b e r [moll
ed by Shibata et al. [5] is not needed because beta i1, : t h e mole number of component i [mol[]

April, 1993
 Continuous Thermodynamics Using the Beta Distribution Function 77

p, q :parameter used in beta distribution function REFERENCES

P : pressure Eatml]
P, :critical pressure l~atm] 1. Bowman, J~ R.: InK Eng. Chem., 41, 2004 (1949).
R : universal gas constant [atm-cm:)/Kmol] 2. Gualtieri, J.A., Kicaid, J.M. and Morrison, G.: J.
S.G. : specific gravity [60~176 Chem. Phys., 77, 521 (1982).
T :temperature ~K] 3. Cotterman, R.L., Bender, R. and Prausnitz, J. M.:
T~ : boiling temperature ~K] Ind. Eng. Chem. Process Des. Dev., 24, 194 (1985).
T, :critical temperature EK~ 4. Cotterman, R, L. and Prausnitz, J.M.: Ind. Eng.
T, :reduced temperature (T/T,) Chem. Process Des. Dev., 24, 343 (1985).
v :molar volume [cm:Tmol] 5. Sbibata, S.K., Sandler. S.t. and Beherens, R.A.:
V : total volume [cm:~] Chem. Eng Sei., 42, 1977 (1987).
x, :mole fraction of i-component in liquid phase 6. Willman, B.T. and Teja, A. S.: AIChEf. 32, 2067
z : compressibility factor (1986).
z~ :mole fraction of i-component 7. Wang, S.H. and Whitting, W.B.: Can. ]. Chem.
Eng., 65, 651 (1987).
Greek Letters 8. Radosz, M., Cotterman, R. L. and Prausnitz, J. M.:
cL :temperature dependent parameter defined in Ind. Eng. Chem. Res., 26, 731 (1987).
Eq. (28) 9. Park, J.: M. S. Thesis, Seoul National Univ., Seoul,
bl chemical potential Korea (1992).
q~ fugacity coefficient 10. Kehlen, H., Ratzsch, M.T. and Prausnitz, J.M.:
w Pitzer's acentric factor AIChE ]., 31, 1136(1985).
rl mole fraction of continuous mixture 11. Soave. G,: Chem. Eng. Sci., 27, 1197 (lC)72).
12. Sire, W.J. and Daubert, T. E.: I&EC Process Des.
Superscripts Dev, 19, 386 (1980).
L :liquid phase 13. Pedersen, K.S., Thomassen, P. and Fredenslund.
V :vapor phase A.: t&EC Process Des. Dev., 23, 163 (1984).
14. Pedersen, K. S, Thomassen, P. and Fredenslund,
Subscripts A.: I&EC Procvss l)es. Dev., 23, 566 (1984).
b : boiling point 15. Pedersen, K.S., Tbomassen, P. and Fredenstund
i, j : any component A.: t&EC Process Des. Dev.. 24, 948 (1985).
N :total number of component 16. Riazi, M. R and Daubert, T. E.: I&EC Res., 26, 755
c : critical property (1987).
r : reduced property

Korean J. Ch. E.(Vol. 10. No. 2)