Multiphase Equilibria Calculation by Direct Minimization

See
discussions, stats, and author profiles for this publication at: https://www.researchgate.net/publication/223511593
Multiphase Equilibria Calculation by Direct

Minimization of Gibbs Free Energy with a
Global Optimization Method
Article in Computers & Chemical Engineering December 2002
DOI: 10.1016/S0098-1354(02)00144-8
CITATIONS
READS
98
269
3 authors:
Dan Vladimir Nichita
Susana Gmez
French National Centre for Scientific Research
Universidad Nacional Autnoma de Mxico
61 PUBLICATIONS 640 CITATIONS
94 PUBLICATIONS 1,085 CITATIONS
SEE PROFILE
SEE PROFILE
Eduardo Luna-Ortiz
Genesis Oil and Gas
13 PUBLICATIONS 228 CITATIONS
SEE PROFILE
Some of the authors of this publication are also working on these related projects:
Differentiation of noisy data with Total Variation Method View project
All content following this page was uploaded by Susana Gmez on 13 October 2015.
The user has requested enhancement of the downloaded file. All in-text references underlined in blue are added to the original document
and are linked to publications on ResearchGate, letting you access and read them immediately.
Computers and Chemical Engineering 26 (2002) 1703 /1724

www.elsevier.com/locate/compchemeng
Multiphase equilibria calculation by direct minimization of Gibbs free

energy with a global optimization method
Dan Vladimir Nichita a,1,*, Susana Gomez b, Eduardo Luna b,2
a
PETROM-Institute for Research and Technology (ICPT), Campina, Romania

b
IIMAS-UNAM, Mexico City, Mexico
Received 23 May 2001; received in revised form 19 June 2002; accepted 1 July 2002
Abstract
This paper presents a new method for multiphase equilibria calculation by direct minimization of the Gibbs free energy of
multicomponent systems. The methods for multiphase equilibria calculation based on the equality of chemical potentials cannot
guarantee the convergence to the correct solution since the problem is non-convex (with several local minima), and they can find
only one for a given initial guess. The global optimization methods currently available are generally very expensive. A global
optimization method called Tunneling, able to escape from local minima and saddle points is used here, and has shown to be able to
find efficiently the global solution for all the hypothetical and real problems tested. The Tunneling method has two phases. In phase
one, a local bounded optimization method is used to minimize the objective function. In phase two (tunnelization), either global
optimality is ascertained, or a feasible initial estimate for a new minimization is generated. For the minimization step, a limitedmemory quasi-Newton method is used. The calculation of multiphase equilibria is organized in a stepwise manner, combining phase
stability analysis by minimization of the tangent plane distance function with phase splitting calculations. The problems addressed
here are the vapor /liquid and liquid /liquid two-phase equilibria, three-phase vapor /liquid /liquid equilibria, and three-phase
vapor /liquid /solid equilibria, for a variety of representative systems. The examples show the robustness of the proposed method
even in the most difficult situations. The Tunneling method is found to be more efficient than other global optimization methods.
The results showed the efficiency and reliability of the novel method for solving the multiphase equilibria and the global stability
problems. Although we have used here a cubic equation of state model for Gibbs free energy, any other approach can be used, as the
method is model independent.
# 2002 Elsevier Science Ltd. All rights reserved.
Keywords: Phase stability; Tangent plane distance; Multiphase equilibria; Global optimization; Equation of state
Nomenclature
B
D
/
/D
E
epsmch
F
f
subspace of constraints in Rn defined by inequalities

tangent plane distance function
dimensionless tangent plane distance function
defined by Eq. (30)
machine precision
mol of feed
objective function
* Corresponding author. Present address: Div. Chimie et Physico-Chimie Appliquees, Inst. Francais du Petrole, 1 et 4 avenue du Bois-Preau RucilMalmaison, 92852 Cedex, France. Tel.: /33-1-4752-6000; fax: /33-1-4752-7025
E-mail address: d-vladimir.nichita@ifp.fr (D.V. Nichita).
1
Currently with IFP.
2
Currently with University of Manchester.
0098-1354/02/$ - see front matter # 2002 Elsevier Science Ltd. All rights reserved.
PII: S 0 0 9 8 - 1 3 5 4 ( 0 2 ) 0 0 1 4 4 - 8
1704
fji
factr
G
g
H
h
K
kij
L , L1, L2
l
M
n
ni
nji
nc
np
ns
p
pgtol
Q
R
T
T
Tp(x )
Tc(x )
Te(x )
u
V
xij
x
y
xG
xm
xs
z
Greek letters
bj
d
o
8
lm
ls
m
n
v
Subscripts
F
i, k
j
pure
r
Superscripts
S
T
0
*
D.V. Nichita et al. / Computers and Chemical Engineering 26 (2002) 1703 /1724
fugacity of component i in phase j

tolerance in Eq. (36)
Gibbs free energy
function defined by Eq. (25)
Hessian matrix
function defined by Eq. (28)
equilibrium ratios
binary interaction coefficients
mol fraction of liquid phases
lower bound on variables
molecular weight
number of independent variables
mol number of component i in feed
mol number of component i in phase j
number of components
number of phases
number of components that may precipitate
pressure
tolerance for projected gradients
objective function given by Eq. (29)
universal gas constant
temperature
Tunneling function
Tunneling function with a pole
classical Tunneling function
exponential Tunneling function
upper bound on variables
mol fraction of vapor phase
mol fractions of component i in phase j
independent variable (Section 3)
mol fractions of vapor phase
the global minimum
stationary points
stationary points of the Tunneling function
feed composition
mol of phase j per mol of feed
the Kronecker delta
tolerance in Eq. (38)
fugacity coefficient
strength of the pole at the point xm
strength of the pole at the point xs
chemical potential
iteration level
acentric factor
feed
component index
phase index
pure component
reduced
solid
transposed
initial value
at stationary points
1. Introduction
Development of robust and efficient methods for the
calculation of phase equilibria has long been a challenge,
and still it is. The purpose of this work is to develop an
efficient and reliable tool for various multiphase equilibria calculations for interest in Chemical Engineering
and Petroleum Engineering.
The difficulty of phase equilibrium calculations consists in the highly non-linear and non-convex form of the
objective function that gives no guarantee that the
global minimum will be found. Complexity of the
problem increases near critical points and phase boundaries.
The theoretical basis of multiphase equilibria has been
first stated by Gibbs (1873) pp 382 /404. However, the
spectacular progresses accomplished during the last
decades are strictly related to the development of
thermodynamic models and the increasing performances
of computers.
The formalism given by Baker, Pierce and Luks
(1982) comprises a set of theorems giving the conditions
for phase stability and multiphase equilibria. They
clearly showed why the incorrect number of phases or
phase compositions may be predicted in phase equilibria
calculations.
The classical approach for phase equilibria calculation is based on the equality of chemical potentials.
Fussel (1979) presents the extension to multiphase
equilibria of the Fussell and Yanosik (1978) Minimum
Variable Newton /Raphson (MVNR) for two-phase
calculations. Risnes and Dalen (1984) are using the
successive substitutions (SSI) method for multiphase
equilibria. Mehra, Heidemann and Aziz (1982) are using
accelerated SSI, followed by Newton iterations. Nghiem
and Li (1984) used Newton /Raphson and Quasi-Newton successive substitution iterations.
We should note the remarkable work of Michelsen
(1982a,b), whose papers represents a milestone in
the development of phase equilibria calculations.
Michelsen formulated a method for finding the
stationary points of the tangent plane distance (TPD)
function (Michelsen, 1982a), and applied second order
methods for multiphase equilibria calculation (Michelsen, 1982b).
The equality of chemical potential represents only the
stationary point conditions for the minimum of thermodynamic potentials. Any method based on these conditions may fail to find the global minimum, being
attracted by local minima. A related feature is the
fractal behavior of robust algorithms, observed by Lucia
et al. (1990), Green, Zhou and Luks (1993).
The modern trend is the treatment of the phase
equilibria problem by the direct minimization of Gibbs
free energy or of the Gibbs TDP function subject to the
linear material balance constraints.
1705
Trangenstein (1985) formulated the two-phase

equilibria reservoir compositional simulation as a constrained optimization problem. Vickery and Taylor
(1986), DeGance (1993) used path continuation methods. The homotopy continuation method was used by
Sun and Seider (1995), for both phase stability analysis
and phase splitting. The approach of Nagarayan, Cullik
and Griewank (1991) is based on the thermodynamic
energy analysis, using a modified semi-local stability
test and a higher order initialization procedure in
the two-phase region. A radically different approach
(the Integral Area Method), presented by Eubank,
Elhassan, Barrufet and Whiting (1992) is based on
the integration of the Gibbs energy surface. Bullard
and Biegler (1993) presents a penalty-based approach,
using a linear programming strategy for solving
vapor/liquid equilibria. A version of their method is
used by Han and Rangaiah (1997) for vapor/liquid,
and liquid /liquid equilibria. Wasylkiewicz, Sridhar,
Doherty and Malone (1996) uses ideas from differential
geometry and the theory of differential equations
for solving global stability analysis and LL phase
splitting. The Interval Newton method was used by
Hua, Brennecke and Stadtherr (1996, 1998), Hua et al.
(1999) for stability analysis.
The simulated annealing (SA) algorithm (Corana,
Marchesi, Martini & Ridella, 1987) was used by Pan and
Firoozabadi (1998) for various kinds of phase equilibria,
Nichita, Goual and Firoozabadi, (2001) for wax precipitation from hydrocarbon mixtures, and Zhu (2000)
for phase stability analysis.
In a series of papers, McDonald and Floudas
solved the phase and chemical equilibrium problem
using different models for the Gibbs free energy
(McDonald & Floudas, 1995a,b) the phase global
stability problem (McDonald & Floudas, 1995c)
and presented the GLOPEQ code (McDonald & Floudas, 1997). They used the deterministic branch and
bound method for global optimization. Although
this method guarantees convergence to the global
solution, the computational cost is generally
unaffordable, as the methods used to compute
the lower bound are still inefficient. In many practical
applications, heuristic stochastic methods based on
a space search, that do not need gradient evaluations,
and that have a probability of convergence, have
been successfully used, as genetic algorithms. These
algorithms have the property to descend the
objective function value very fast at the beginning, but
as they loose diversity information of the space, they
tend to be extremely slow to actually get the
global solution (see Gomez, Gosselin & Barker, 2001).
In this paper, we propose the use of the Tunneling
global optimization method for solving the phase
stability and multiphase equilibrium problems. This
method has a hybrid nature. It is deterministic in the
1706
sense that it produces a monotonic sequence of improved local minima (avoiding all the local solutions
above). It also has a stochastic element, as the initial
points to start the search for a point in another valley
(Tunneling phase) are taken in random directions,
allowing the method to perform an intelligent space
search for more promising valleys. The Tunneling
method is able to escape from local minima and saddle
points, and has shown to be able to find efficiently the
global solution for all the academic and industrial
problems tested (Gomez and Romero, 1994; Gomez,
Keeping & Pantelides, 1998; Gomez, Gosselin and
Barker, 2001; Gomez, Solano, Castellanos & Quintana,
2001).
The Tunneling method has two phases. In phase one,
a local bounded optimization method is used to
minimize the objective function. In the second
phase (tunnelization), either global optimality is ascertained, or a feasible initial estimate for a new
minimization is generated. For the minimization
step, a limited-memory quasi-Newton method is
used (Zhu, Byrd, Lu & Nocedal, 1994; Byrd, Lu,
Nocedal & Zhu, 1995), but any favorite minimizer
may be used.
The Tunneling method has proved its efficiency and
reliability on several highly difficult academic and
industry. The method proposed here does not require
a priori the number of phases existing at equilibrium.
The calculation sequence is organized in a stepwise
manner. Results from two-phase flash and stability
analysis provide good initial guesses for the three-phase
equilibrium. Phase distribution is obtained by solving a
convex unconstrained optimization problem (Michelsen,
1994).
The code Tunneling phase equilibria (TUNPEQ)
incorporates the code TUNNEL for the global optimization method, developed at IIMAS-UNAM (Castellanos & Gomez, 1998).
A general form of cubic equations of state (EOS),
(Coats, 1985; Michelsen, 1986), including the Soave/
Redlich /Kwong (SRK) EOS (Soave, 1972), and the
Peng /Robinson (PR) EOS (Peng & Robinson, 1976) is
currently incorporate in TUNPEQ. However, the
method is general and model independent; although an
EOS formulation is used here for the Gibbs free energy,
any other thermodynamic model can be used for various
kinds of thermodynamic equilibria.
We will present the multiphase equilibrium problem
in Section 2, the description of the Tunneling
global optimization method will be given in Section 3,
and the results on a set of difficult problems will be
reported and discussed in Section 4. A comparison
with other global methods will be presented in Section 5
and some concluding remarks will be given toward the
end.
2. Statement of the phase equilibria problem
2.1. Multiphase equilibria calculation by direct

minimization of the Gibbs free energy
Consider that at given pressure and temperature
conditions, a mixture of nc, components of composition
z /(z1, z2, . . ., znc)T, may split into np phases. We
assume that we have np/1 condensed phases, and the
np phase is the vapor phase. According to the second
law of thermodynamics, at the equilibrium state, the
system has the lowest possible Gibbs free energy from all
possible states. The Gibbs free energy is expressed as:
G(n)
nc X
np
X
nji mji (nj )
(1)
i1 j1
where n/(n1, n2, . . ., nnp)T is a vector of nc /np

elements, and the nj /(nj 1, nj 2, . . ., nj nc)T; j/1, np
vector contains the component mol fractions in phase j.
It is more convenient for EOS calculations to express
the Gibbs free energy G in term of fugacities. Eq. (1) can
be rewritten as:
G(n)
G(n)
RT
nc X
np
X
nji ln fji (xj )
(2)
i1 j1
where G is the reduced Gibbs free energy.

The constrained optimization problem is:
G(n) 0 min
(3)
such that the component material balance constraints:

ni
np
X
nji ;
i 1; nc
(4)
j1
and bounds of independent variables:

0 5nji 5ni ;
i 1; nc;
j 1; np
(5)
where ni /Fzi are the feed component mol numbers for

F mol of feed.
Fugacities depends on mol numbers by means of mol
fractions xj /(xj 1, xj 2, . . ., xj nc); j/1, np:
xji
nji
nc
X
(6)
njk
k1
In Eq. (2), one can eliminate the mol numbers of one

phase:
n1i ni
np
X
nji ;
i 1; nc
(7)
j2
and the optimization problem (3) becomes an unconstrained one, with only bounds on the remaining
nc(np/1) independent variables, n2, . . ., nnp.
The gradient of the objective function is:

X

nc
@G
f
@ ln fjk
@ ln f1k
ln ji
n1k
njk
;
@nji
f1i
@nji
@n1i
k1
j 2; np;
The expression of the reduced TPD in terms of

fugacities is:
(8)
i1
nji
@ ln fji
@njk;k"i
D(x)
i 1; nc
0;
j 1; np
D(x)
RT
nc
X
xi [ln fi (x)ln fi (z)]
(14)
i1
If one of the mol fractions is eliminated:
According to the Gibbs /Duhem equation:

nc
X
1707
(9)
xnc 1
nc1
X
xi
(15)
i1
the sums in Eq. (8) vanish. Note that Eq. (9) is satisfied
for any mol fractions, not only for the equilibrium mol
fractions. Therefore, it is not necessary to compute the
fugacity derivatives in order to compute the gradient of
the Gibbs free energy:

@G
f
ln ji ; j 2; np; i 1; nc
(10)
@nji
f1i
In this work a Limited Memory Quasi-Newton
minimization technique is used (the L-BFGS-B routine,
Zhu et al., 1994), that requires only the values of the
objective function and of the gradients. Any solver may
be used for minimization; if the elements of Hessian
matrix are required, there are given by:
2
@ G
@ ln fji @ ln f1i
Hik
dik

;
@nji @nlk
@nli
@n1k
(11)
the optimization problem to be solved is:

D(x) 0 min
(16)
for nc/1 mol fractions as independent variables

bounded by:
0 5xi 5 1;
i 1; nc1
(17)
This is an unconstrained optimization problem. The

elements of the gradient vector are:

@D
fi (x)
fnc (x)
ln
; i 1; nc1
(18)
ln
@xi
fi (z)
fnc (z)
and elements of the Hessian matrix are given by:
Hik
@2D
@xi @xk
@ ln fi (x)
@xk
@ ln fnc (x)
@xk
(19)
i; k 1; nc1
i; k 1; nc(np1)
where d is the Kronecker delta.
2.3. Stepwise organization of multiphase equilibria
calculation
2.2. Phase stability analysis by minimization of the

tangent plane distance function
Phase stability testing is not only an important and
intricate problem per se, but the stability analysis
routine is a key sequence in the stepwise organization
of multiphase equilibria calculation program. Stability
analysis provides equilibrium phase status and good
initial estimates for flash calculations.
The TDP function is defined as:
D(x)G(x)T(x; z)
nc
X
xi [mi (x)mi (z)]
(12)
i1
The TPD function is given by the difference between

the Gibbs free energy surface in x, and the tangent plane
to this surface in z, where, x is the composition of a trial
phase, and z is the composition of the phase whose
stability is to be asserted.
The necessary and sufficient condition for phase
stability is given by Theorems 3 and 4 of Baker et al.
(1982). The mixture is stable if and only if the TPD
function is nonnegative at all stationary points, or:
D(x)]0()x
(13)
Convergence to the correct solution of the problem

(global minimum) can be obtained irrespective to initial
guesses, provided a feasible initial estimate is given. This
is one of the strengths of the Tunneling global optimization method. However, for computational efficiency,
improved initial estimates are required.
The sequential approach for stability/flash calculation
has been used by Michelsen (1982b), Nghiem and Li
(1984).
Basically, the idea is that one can take advantage in
the np-phase flash on the results of (np/1)-flash and of
the stability testing for one of the equilibrium phases.
The sequence of calculations is as follows:
1) stability analysis is tested for a phase with overall
composition z ;
2) results of stability testing are used to initiate the
two-phase flash;
3) two-phase flash calculations are performed;
4) stability analysis of one of the equilibrium phases
(say, the liquid phase) from two-phase flash is
tested;
1708
5) results from step 3 and 4 are used to initiate the

three-phase flash;
6) three-phase flash calculations are performed;
7) one can check the results of three-phase flash by
testing the stability of equilibrium phases. There is
no need to test the stability of all phases, because
the points representing equilibrium phases must
have a common tangent plane, according to the
necessary condition of equilibrium.
The TUNPEQ code is currently developed for threephase; extension to four-phase equilibria VLLL is being
investigated.
zi
xnpi yi
1
np1
X
bj ((1=Kji ) 1)
(24)
j1
i 1; nc
and the mol fractions of the remaining phases, xji ,
i/1, nc, j/1, np/1 are obtained from Eq. (23).
The gj functions are defined:
gj (b)
nc
X
[xji (b)yi (b)];
j 1; np1
(25)
i1
Taking into account that:

2.4. Initialization
nc
X
xji 1;
j 1; np
(26)
i1
2.4.1. Initialization of phase splitting calculations

For the initialization of two-phase flash, the procedure of Michelsen (1982b) is used here. Let us discuss
initiation of three-phase flash (Step 5).
The independent variables are the component mol
numbers for np/1 phases. From the stepwise approach
we have component mol fractions from flash and
stability. We need the molar fractions of phases (mol
of phase j per mol of feed) b/(b1, . . ., bnp)T to calculate
the mol numbers of component i in phase j:
nji bj xji ;
i 1; nc;
j 1; np
(20)
The problem is to find the phase distribution. This

can be achieved either by: (a) solving a non-linear
system of np/1 equations (the extension of the
Rachford /Rice equation to the multiphase case), or
(b) by solving an optimization problem.
a) Extension of Rachford /Rice equation to the multiphase case.
The component material balance, Eq. (4), can be
written:
nc
X
bj xji ni ;
j 1; np
(21)
i1
with the overall material balance:

np
X
(22)
bj 1
i1
The definition relationship for the equilibrium

constants is:
Kji
yi
8
ji ;
xji 8 npi
i 1; nc;
j 1; np1
(23)
we obtain a set of np/1 equations with the

unknowns b1, b2, . . ., bnp1.
gj (b)
i1
(1=Kji 1)zi
0;
np1
X
1
bj (1=Kji 1)
(27)
j1
j 1; np1
If the vapor phase disappears Eq. (27) cannot be
applied; in this case, in a similar manner the
following set of equations is obtained:
hj (b)
nc
X
[xji (b)x1i (b)];
j 2; np
(28)
i1
and solved for the unknowns b2, . . ., bnp.

The nonlinear systems of Eqs. (27) and (28) can
be easily solved by a globally convergent Newton
method (Press, Flannery, Teukolski & Vetterling,
1989).
However, some problems may arise with vanishing phases during iterations. Bunz, Dohrn and
Prausnitz (1991) suggests that a three-phase flash
should be performed only if the outcome of prior
single-phase tests and two-phase flashes indicates a
three-phase distribution.
In order to avoid problems of disappearing
phases during iterations for solving Eqs. (27) and
(28), we have chosen to be included in TUNPEQ the
procedure described next.
b) The technique suggested by Michelsen (1994) for
calculation of phase distribution. The objective
function Q is defined by Michelsen:
Q(b)
By combining Eqs. (22) /(24), the mol fraction of

the gas phase can be expressed as:
nc
X
np
X
j1
where
bj
nc
X
i1
ni ln Ei
(29)
Ei
np
X
bj
j1
8 ij
2.5. Convergence criteria

;
i 1; nc
(30)
and the fugacity coefficients are calculated using

mol fractions provided by two-phase flash, and the
mol fraction of the trial phase from stability analysis
of two-phase equilibrium liquid.
We have to solve an unconstrained optimization
problem:
Q(b) 0 min
s:t: 05 bj 5 1 ;
j 1; np
(31)
(32)
The gradient vector is given by:

@Q
@bj
1
1709
nc
X
ni
i1
8 ij Ei
j 1; np
There are two stopping criteria for the minimization

routine. The first is given by:
(f n f n1 )
Bfactrepsmch
max(f n1 ; f n ; 1)
where epsmch is the machine precision (automatically

generated by the code), factr is provided by the user to
get desired tolerance; here f denotes the objective
function.
The second criteria checks on the norm of projected
gradients:
maxfprojgri ;
(33)
The objective function is a convex function

defined on the convex set b , therefore the minimum
of Q is unique. One can eliminate one variable from
Eq. (22) and use np/1 independent variables in Eq.
(29), but it is more convenient to maintain the full
set of equations (Michelsen, 1994).
We use the same solver (L-BFGS-B) to minimize
the objective function Q . Usually, few iterations are
required for convergence.
(36)
i 1; ng 5pgtol
(37)
We have set a high restrictive value for the tolerance

pgtol (1.d-8) for all examples.
The Euclidean norm on fugacities is also recorded,
and a third additional criteria is set for flash calculations
only
S
X
np X
nc
(fji f1i )2
0:5
Bo
(38)
j2 i1
3. The Tunneling global optimization method

2.4.2. Initialization of stability analysis calculations
The initialization procedure for stability analysis, as
suggested by Michelsen (1982a), and by Sun and Seider
(1995), is based on the equations:
Yi zi Ki
Yi
zi
Ki
(34a)
(34b)
where estimates for equilibrium constants are obtained

from the Wilson (1969) empirical relationship:
Ki pri exp[5:37(1vi )(1Tri1 )]
(35)
A slightly modified initialization is used in this work.

The composition of the trial phase is first estimated
from Eqs. (34a) and (34b). The normalized values of mol
fractions are then used to calculate the fugacities, then
with K -values from fugacities, corrected values of the
mol fractions follows from Eqs. (34a) and (34b).
Basically, this is equivalent to performing one SSI.
This initialization procedure brings the equilibrium
constants at more realistic (close) values. We designate
the initialization of being of Type L (for liquid stability
testing), or type V (for vapor stability testing), whether
Eqs. (34a) and (34b) is used.
The aim of the global optimization method is to

track down the global minimum skipping every unwanted local minimum. In this work we test the benefits
of using a global optimization method called the
Tunneling method (Barron & Gomez, 1991; Levy &
Gomez, 1985; Levy & Montalvo, 1985) for the calculation of multiphase equilibria, and improving results
obtained using standard local methods.
This method has been designed to find the global
optimum of a general non-convex smooth function f,
and in this work the objective is to find the global
solution xG of:
min f (x)
xB
(39)
with B /{x jl 5/x 5/u }, f:Rn 0/R , x /Rn , f /C2, n being

the number of independent variables.
In this study the objective function f is either the TDP
function (for phase stability analysis), or the Gibbs free
energy (for phase splitting calculations).
It is necessary to use a global optimization method,
whose computational cost is affordable. The cost in this
context is determined by the number of times the
objective function has to be calculated, including its
gradient.
1710
Fig. 1. The basic idea of the Tunneling method.
3.1. The global optimization method

The code that we are using in this work has recently
been developed (Castellanos & Gomez, 1998), and
incorporates the numerical experience obtained in recent
years (Gomez, Levy, & Calderon, 1987; Barron &
Gomez, 1991; Gomez & Romero, 1994). It has already
proven to be very efficient to solve highly difficult
problems, both academic and real (see for instance
Gomez et al., 1998; Gomez et al., 1999; Gomez et al.,
2001).
The basic idea of these methods (see Fig. 1) is to
tunnel from one valley of the objective function to
another, to be able to find a sequence of local minima
with decreasing function values:
f (x1 )]f (x2 )]. . .]f (xG )
(40)
where xG is the global minimum of f(x ), ignoring all the

local minima with larger objective function values than
the best already found. This characteristic of ignoring
minima makes the algorithm faster and more efficient
than other general purpose methods, like SA, random
search, clustering, and genetic algorithms (see Gomez et
al., 2001).
3.2. The Tunneling method
To perform the minimization phase any algorithm

designed to solve local optimization problems with
bounds on the variables can be used. We use a
limited-memory Quasi-Newton method (Zhu et al.,
1994; Byrd et al., 1995). Any favorite optimizer can
also be used.
3.2.2. Tunneling phase

Once a local minimum has been obtained, we want to
find a point in another valley, that is we need to solve
the following inequality problem:
Find
f (xm )
xm B such that f (x)5
m
(42)
To accomplish this goal using gradient-type methods

(Newton-like iterative algorithms), it is necessary to
destroy the minimum, placing a pole at the minimum
point xm, and then to start Tunneling (solving inequality
Eq. (42)), from an initial point near the minimum.
As we are trying to find a point xm in another valley
with less or equal value than fm , we first create the
Tunneling function Tp(x ), such that, if f(x )5/fm then
Tp(x )5/0, for x "/xm.

Tp (x) (f (x)f (xm ))P(x; lm ; xm ) if xxm 51
Tp (x) f (x)f (xm ) if xxm ]1
(43)
The Tunneling method has two phases. In the

minimization phase, starting from an initial point x0, a
local minimum x1 is found with f1 /f (x1), using any
local bounded optimization method. From x1 in phase 2
it obtains a feasible point x1 in another valley with
f(x1 )5/f1, which will be taken as the initial point for the
next minimization phase.
3.2.1. Minimization phase
The problem here is to find a local bounded minimum:
Find
xm arg(min f (x))
xB
(41)
where, we have placed a pole P at xm using one of the

following alternatives:
1) Classical Tunneling function (Levy & Montalvo,
1985),
P(x; lm ; xm )xxm lm
(44)
2) Exponential Tunneling function (Barron & Gomez,

1991),

lm
P(x; lm ; xm )exp
(45)
x xm
where lm is the strength of the pole.

Solving problem Eq. (44) now consists of finding xm
such that:
Tp (x)50
m
(46)
The role of lm is crucial for the algorithm. In the

classical method its value is an integer that depends on
the multiplicity of the minimum. In the exponential
method a small real value is enough to place a pole
destroying the minimum. However, if the minimum was
not computed exactly, for both methods in the direction
towards the exact minimum, a small value would not
create a pole, and a larger lm has to be used. Also,
increasing lm has the effect of smoothing the Tunneling
function, as it inherits the multimodality of the original
function f. Although in principle, it would be desirable
to take a large lm so that T does not have stationary
points, a large lm gives larger values of the Tunneling
function T further away from the solution to problem
Eq. (46). Thus, in practice, lm is computed just to
destroy the minimum, that is the strength of the pole lm
is computed so that a descent direction can be generated, and mobile poles are used whenever a stationary
point of T is encountered, as will be described in Section
3.4. The code TUNEL manages poles and mobile poles
automatically. For further details in the actual implementation, the reader is referred to Castellanos and
Gomez (1998).
We need to take descent directions to solve the
inequality Eq. (46), and thus we use the same algorithm
used to solve Eq. (41), with appropriate stopping
conditions to check convergence for Eq. (46).
As the original objective function f is a general
nonlinear function only assumed to belong to C2 for
x /B , it could have many local and global minima and
convergence to minima at the same level is possible, that
is f (x1m*) /f(x2m*) /. . ./f (xtm*) as xtm* would be acceptable solutions of Eq. (46). In order to avoid cycling
and going back to these minima at the same level
already found, it is important to preserve the poles used
to destroy them, until a better minimum xt1
m * with
lower value of the objective function is found. To
achieve this goal it is necessary to modify the definition
of the Tunneling function:

t
Y
lt
m
Te (x)(f (x)f (xm
(47)
))
exp
x xt
i1
m
and making t/1 as soon as an xm is found with smaller
value than fm.
3.3. Initial point for the Tunneling phase
Once a local minimum xm has been found, we
generate an initial point x0m to start the Tunneling
process. This point is generated along a random
direction in a neighborhood of xm, to preserve local
1711
information. Then from this point, the local minimization code will generate descent directions and step
lengths to force descend of the Tunneling function, until
it finds a point where inequality Eq. (46) is satisfied. It is
important to recall, that the random direction is only
used to produce the initial point in the neighborhood of
the local minimum xm, to be able to start from there, the
local descent method.
If Tunneling is not successful from there, we generate
a new initial point, until 2n (n being the dimension of
the problem) initial points have been tried. We then
generate initial points at random in the entire feasible
region, until the maximum number of function evaluations (FE) given by the user, has been used. The
Tunneling search is not successful when:
. A corner of admissible region has been reached.
. The strength of the pole is greater than a maximum
value given, without having obtained a descent
direction.
. No further precision for x is possible.
. The maximum number of FE allowed for this phase
has been reached.
3.4. Stationary points of T (x )
The Tunneling function T inherits the non-convex
nature of the original objective function f. It is then
possible that during Tunneling phase we reach stationary points xs of the Tunneling function T (x ), that is
points where the gradient of T with respect to x is zero
(up to a tolerance). As we use a gradient type method to
find xm that satisfies Tp(xm)5/0, to be able to continue
the search we place mobile poles that we turn on and off,
and the Tunneling function is modified multiplying by
the mobile pole:
Tp (x) (f (x)

ls
exp
x xm
x xs
f (xm ))exp
lm
(48)
These poles are placed where the stationary point is

detected and while needed and we turn it off when the
iterand is out of the zone of attraction of the stationary
point (making ls /0).
3.5. General stopping conditions
To check for global optimality, either a lower bound
for the objective function is known and has been
attained or the search for the solution T (x )5/0 has to
be exhaustive.
In general the value of the objective function at the
global solution is not known, but when the global
solution has been found, there is not a point xm in
another valley with f(xm)5/f(xm). Consequently, the
1712
algorithm has to spend some time checking for global

optimality, by performing unsuccessful Tunneling
searches, starting from all the initial points allowed.
The user decides then, how much computer effort he
wants to spend to check for global optimality through
the maximum number of initial points to start the
Tunneling phase, or through the maximum number of
FE.
The algorithm stops when any of the following global
criteria is satisfied:
. In the Tunneling phase the given maximum number
of initial points to start the search for xm has been
reached. The last minimum found is the putative
global minimum.
. The given maximum number of FE has been reached.
. The lower bound of the objective function given by
the user has been attained. The last minimum found
is the putative global minimum.
. All the global minima at the same level required by
the user have been found.
3.6. Features of the code
We have developed an in-house FORTRAN prototype
to perform the minimization iteration with calls to the
TUNPEQ code providing gradient information, calls to
the local optimizer and to the Tunneling module, in a
way giving to the user flexibility to design the numerical
experiments. Some of the options are:
1) The user can incorporate the local optimization
method more suitable for his problem. However, if
not decided otherwise, the Tunneling code uses a
Limited Memory Quasi-Newton Method.
2) The number of cycles min /tun /min, and/or the
maximum number of FE the user wants to spend in
the process.
3) The tolerances to control:
. The precision for the local optimization (see Eqs.
(36) /(38)).
. the desired decrease in the objective function
imposed to the Tunneling phase. This can help
the user to force the code to search for another
valley with sufficiently lower objective function.
. The precision to find minima at the same level.
This can be interesting if a sufficiently good value
of the objective function has already been
attained, and the user is interested in finding
alternative solutions within certain range (/ or
/) of this objective value.
4) The number of tries from different initial points in
the Tunneling phase.
5) The code is programmed to save all the files
containing partial results at every local minimum
the method finds along the way. Reverse commu-
nication allows the user even to save every iteration

points within the local minimization (if the local
minimizer has also reverse communication).
4. Results
This section presents the results of phase stability
analysis and multiphase equilibria calculation for several representative systems. The problems addressed
here are vapor /liquid, liquid/liquid, vapor /liquid/
liquid, vapor /solid, liquid/solid, and vapor /liquid/
solid calculations.
4.1. Stability analysis
The reliability of the phase stability testing module of
TUNPEQ is first investigated separately. We have
chosen a set of five examples from the literature, for
binary and ternary mixtures. These simple mixtures
exhibit several stationary points of the TPD function,
among them local minima and saddle points that may
attract the minimizer. TUNPEQ finds the global
minima of the TPD for all examples.
For all examples considered, both kinds of initial
estimates (V and L) are used. For some cases, initial
guesses other than those given by V or L initialization
are provided, in order to test the ability of our approach
to escape certain local minima that are not found with
the V or L initializations.
The following test examples (Eqs. (1) /(5)) are addressed by Hua, Brennecke and Stadtherr (1998). For all
cases results are consistent with their results. The
problem of stability testing will be also addressed in
the next sections in a multiphase context.
4.1.1. Problem 1. Hydrogen sulfide /methane binary
mixture
This is a mixture of methane and hydrogen sulfide at
T /190 K and p /40.53 bar. Calculations are performed using the SRK EOS, with the BIC 0.08. This
example is taken from Michelsen (1982a), and it
represents a very difficult problem. Sun and Seider
(1995); Hua, Brennecke and Stadtherr (1998) have also
tested their proposed methods for stability analysis on
this example.
The TPD function is plotted in Fig. 2a versus the trial
phase composition for an equimolar mixture. Fig. 2b
provide a detail of the region between z1 /0.9 and 1. A
third branch corresponding to the intermediate Z -factor
root of the EOS, with a local minimum at z1 /0.968 is
disregarded here.
There are four stationary points, three minima and a
maximum corresponding to the trivial solution. The
global minimum corresponds to 0.9254 methane, and
has a negative value, indicating instability.
Fig. 2. (a) TDP function for the methane /hydrogen sulfide binary
mixture. (b) TPD function for the methane /hydrogen sulfide binary
mixture (detail).
1713
For any initial guess comprised between z1 /0.5 and

0.954, and between z1 /0.982 and 1, the global minimum is found by the first minimization step. This is true
for the minimizer we use; for example, the Newton/
Raphson algorithm will find the local minimum for
initial guesses z1 /0.98. The local minimum at 0.98 is
first found by the minimizer only for initial guesses
falling between z1 /0.954 and 0.982. For z1 B/0.5, the
local minimum at z1 /0.118 is first found. This is an
illustration of the fractal behavior of solution techniques
for solving the stationary point problem observed by
Green et al. (1993); Lucia et al. (1990). For the latter two
cases, after a tunnelization phase the second minimization leads to the global minimum.
Two-phase VL split gives y1 /0.98 and x1 /0.11.
Stability analysis of these two feeds indicates again
instability. A liquid/liquid phase split gives the correct
equilibrium state, consisting in two liquid phases with
x11 /0.92 and x21 /0.11, respectively. Stability analysis
of these two feeds gives positive or zero values of
minimum TPD.
Table 1 provides for different feed compositions: the
stationary points of the TPD function found by the
minimizer, the corresponding values of the objective
function, the number of FE, norms of projected
gradients at the solution, number of minimization
phases, and initial guesses. Note that in column 4
from Tables 1/6 the number of FE required in the
minimization step is given in parenthesis. Type of
initialization is also given in the last column.
For this binary mixture, for all feeds considered, the
number of FE is less than half than the number of root
inclusion tests for FCSM, and one order of magnitude
less than F versions of the interval analysis method of
Hua et al. (1998).
Table 1
Problem 1: C1/H2S SRK-EOS at p 40.53 bar and T 190 K
Feed composition
Stationary points of the TPD function
Objective function
FE
NT
Initial guess
(0.50, 0.50)
/
/
/
/
(0.9885, 0.0115)
/
(0.9813, 0.0187)
/
/
/
/
(0.112, 0.888)
/
/
(0.11, 0.89)
/
/
(0.925382, 0.074618)
(0.118133, 0.881867)
(0.925382, 0.074618)
(0.981559, 0.018441)
(0.925382, 0.074618)
(0.9885, 0.0115)
(0.9885, 0.0115)
(0.923314, 0.076686)
(0.115208, 0.884792)
(0.923314, 0.076686)
(0.9813, 0.0187)
(0.923314, 0.076686)
(0.920821, 0.079179)
(0.112, 0.888)
(0.920821, 0.079179)
(0.919132, 0.080868)
(0.11, 0.89)
(0.11, 0.89)
8.2521e2
5.6893e2
8.2521e2
7.9352e2
8.2521e2
0
0
3.9565e3
1.09098e2
3.9565e3
0
3.9565e3
2.4623e3
0
2.4623e3
5.6076e3
0
0
696 (38)
647 (87)
/
853 (832)
/
512 (6)
581 (17)
645 (8)
693 (237)
/
690 (4755)
/
630 (7)
632 (85)
/
681 (76)
1
2
/
2
/
1
1
1
2
/
2
/
1
2
/
2
/
1
L(0.982223, 0.017777)
V(0.070334, 0.929666)
/
*(0.955, 0.045)
/
L(0.988860, 0.011140)
V(0.432088, 0.567912)
L(0.982297, 0.017703)
V(0.214960, 0.785040)
/
*(0.97, 0.03)
/
L(0.937174, 0.062826)
V(0.066633, 0.933367)
/
L(0.936020, 0.063980)
/
V(0.066010, 0.933990)
667 (8)
1714
Table 2
Problem 2: C1/C3 SRKEOS at p 50 bar and T 277.6 K
Feed composition
Objective function
FE
NT
Initial guess
(0.10,
/
(0.40,
/
(0.60,
/
(0.90,
/
(0.10, 0.90)
/
(0.865478, 0.134522)
/
(0.192783, 0.807217)
/
(0.90, 0.10)
/
0
0
0.152987
/
0.2226
/
0
0
692 (9)
637 (6)
864 (6)
987 (13)
2369 (9)
2417 (6)
1060 (6)
1023 (13)
1
1
1
1
1
1
1
1
L(0.606896,
V(0.093568,
L(0.879067,
V(0.294238,
L(0.784372,
V(0.171736,
L(0.909399,
V(0.387085,
0.90)
0.60)
0.40)
0.10)
4.1.2. Problem 2. Methane/propane binary mixture

This is a mixture of methane and propane at T /
277.6 K. Calculations are performed with the SRK
EOS at p/50 bar (Table 2) and p /100 bar (Table 3).
The BIC is set at 0.029.
Feeds 2 and 3 at 100 bar are difficult problems
because of the near-criticality. Note the low absolute
values of the TPD function at the global minimum. The
method correctly predicts instability in both cases.
4.1.3. Problem 3. Ethane /nitrogen binary mixture
This is a mixture of ethane and nitrogen at p /76 bar
and T /270 K. The PR EOS with k12 /0.08 is used.
Results are given in Table 4 for four feeds. Feed 2 is near
the bubble point locus and feed 3 is near the dewpoint
locus. With both L and V initial estimates, convergence
to the global minimum is observed. For the feed with
z1 /0.56, an extra initial guess is provided; for this case,
two minimization steps are required (the first finds the
trivial solution; after tunnelization, the second one
converge to the global minimum).
4.1.4. Problem 4. Carbon dioxide/methane binary
mixture
This is a mixture of methane and carbon dioxide at
T /220 K and p/60.8 bar. The PR EOS is used with
k12 /0.095. Calculations are performed for five feeds.
For z1 /0.90 and 0.40 (see Table 5), the mixture is stable
vapor and liquid, respectively.
0.393104)
0.996106)
0.120933)
0.705762)
0.215628)
0.828264)
0.090601)
0.612915)
For z1 /0.80, the mixture is stable and close to the

dewpoint locus. The minimizer avoids the trivial solution for both initial estimates. For z1 /0.57 the mixture
is unstable and near the bubble point locus. For one of
initial estimates, two minimization phases are required.
4.1.5. Problem 5. Methane, ethane, and nitrogen ternary
mixture
This is a mixture of methane, ethane, and nitrogen at
T /270 K and p /76 bar. The PR EOS is used, with
k12 /0.021, k13 /0.038, k23 /0.08. Four feeds are
considered. Results are given in Table 6.
Feed 1 is unstable, and close to the dewpoint locus.
With V initialization the method converge to global
minimum. With L initialization it converge first to a
stationary point (close to the trivial solution), then the
second minimization converge to the global minimum.
Feed 2 is in the two-phase region and close to the critical
point. The correct answer is given by the method for
both initial estimates. Feed 3 is at near-critical conditions, in the single-phase vapor region. Convergence of
the method to the trivial solution indicates stability.
Feed 4 is also stable; convergence to the trivial solution
is obtained.
For most of the problems considered in this section,
the number of FE is less then the number of root
inclusion tests (Hua et al., 1998) for the constrained
space plus monotonic interval extensions case (FCSM),
and up to one order of magnitude less than their F case.
Table 3
Problem 2: C1/C3 SRKEOS at p 100 bar and T 277.6 K
Feed composition
Objective function
FE
NT
Initial guess
(0.40,
/
(0.68,
/
/
(0.73,
/
/
(0.90,
/
(0.40, 0.60)
(0.40, 0.60)
(0.68, 0.32)
(0.772465, 0.227535)
(0.772465, 0.227535)
(0.650287, 0.349713)
(0.757048, 0.242952)
(0.650287, 0.349713)
(0.90, 0.10)
(0.90, 0.10)
0
0
0
3.346e4
/
2.948e4
2.081e5
2.948e4
0
0
592 (6)
586 (6)
1196 (137)
/
1113 (7)
1539 (9)
1473 (915)
/
1214 (7)
1259 (11)
1
1
2
/
1
1
2
/
1
1
L(0.426354,
V(0.093997,
L(0.881459,
/
V(0.410751,
L(0.881402,
V(0.411153,
/
L(0.921832,
V(0.530911,
0.60)
0.32)
0.27)
0.10)
0.573646)
0.906003)
0.118541)
0.589249)
0.118598)
0.588847)
0.078168)
0.469089)
1715
Table 4
Problem 3: C2/N2 PREOS at P 76 bar and T 270 K
Feed composition
Objective function
FE
NT
Initial guess
(0.90,
/
(0.82,
/
(0.70,
/
(0.56,
/
/
/
(0.40,
/
(0.90, 0.10)
/
(0.505590, 0.494410)
/
(0.510582, 0.489418)
/
(0.845281, 0.154719)
/
(0.56, 0.44)
(0.845281, 0.154719)
(0.40, 0.60)
/
0
0
9.746e3
/
1.379e2
/
1.553e2
/
0
1.553e2
0
0
1033 (7)
1167 (7)
1048 (10)
1092 (8)
1247 (12)
1282 (8)
1502 (7)
1318 (7)
1329 (86)
/
1469 (6)
1524 (9)
1
1
1
1
1
1
1
1
2
/
1
1
L(0.505716,
V(0.920178,
L(0.381899,
V(0.884377,
L(0.370731,
V(0.883679,
L(0.381497,
V(0.888463,
*(0.5, 0.5)
/
L(0.305497,
V(0.845406,
0.10)
0.18)
0.30)
0.44)
0.60)
The number of initial points in the Tunneling phase was

set to 20 for all examples; same results would be
obtained with much less computer time for less initial
points. Note that for Eqs. (2) /(5) results are also
consistent with results from Prausnitz, Lichtenthaler
and Gomez de Azevedo (1986).
0.494284)
0.079822)
0.618101)
0.115623)
0.629269)
0.116321)
0.618503)
0.111537)
0.694503)
0.154594)
using PX iterations for dewpoints, and PY iterations

for bubblepoints (Fussell & Yanosik, 1978). The critical
point temperature (293.78 K) and pressure (210.668 bar)
were calculated based on Heidemann and Khalil (1980)
criteria, using a different solution procedure (Nichita,
Minescu, & Cretu, 1996; Nichita, 1996).
To decide whether one or two phases exists at
equilibrium, one can perform stability analysis. If
TPD B/0 results from stability are used to initiate flash.
If flash calculations are carried out in the single-phase
region, the two-phase code returns the trivial solution.
Intensive testing has shown that both approaches give
the correct answer, even for the most difficult points.
Equilibrium constants obtained from flash calculations on the critical isotherm are presented in Fig. 4. In
the immediate vicinity of the critical point (for pressures
between about 210.66 bar and the critical pressure), our
method has no problems in finding the correct solution.
For these conditions, the convergence radius of the
classical MVNR method is extremely small, therefore,
the switch from SSI to MVNR iterations has to be
performed at very small values of the Euclidean norm of
the error vector, otherwise divergence occurs.
4.2. Two-phase vapor /liquid equilibria

Two-phase vapor/liquid equilibria calculations were
carried out for two hydrocarbon mixtures, with emphasis to the near-critical region.
4.2.1. The Yarborough 8 mixture
Two-phase vapor/liquid equilibria calculation were
performed with the Tunneling method for a synthetic
mixture of six normal-alkanes (feed mol fractions:
0.8097 C1, 0.0566 C2, 0.0306 C3, 0.0457 nC5, 0.033
nC7, 0.0244 nC10), referred in the literature as the Y8
mixture (Yarborough, 1972). The Peng /Robinson EOS
is used, with all BICs zero.
The phase envelope is given in Fig. 3. Saturation
pressures were calculated with the MVNR method,
Table 5
Problem 4: C1/CO2 PREOS at p 60.8 bar and T 220 K
Feed composition
Objective function
FE
NT
Initial guess
(0.90,
/
(0.80,
/
(0.70,
/
/
(0.57,
/
/
(0.40,
/
(0.90, 0.10)
/
(0.502644, 0.497356)
/
(0.642304, 0.357696)
(0.815263, 0.184737)
(0.815263, 0.184737)
(0.57, 0.43)
(0.808873, 0.191127)
(0.808873, 0.191127)
(0.40, 0.60)
(0.40, 0.60)
0
0
7.427e3
/
1.876e4
7.003e3
7.003e3
0
1.3098e3
1063 (6)
1125 (10)
1739 (8)
1831 (6)
1353 (85)
/
1301 (7)
1284 (75)
/
1364 (7)
1836 (9)
1711 (8)
1
1
1
1
2
/
1
2
/
1
1
1
L(0.908073,
V(0.541701,
L(0.838722,
V(0.270924,
V(0.237605,
/
L(0.845278,
V(0.205001,
/
L(0.839429,
V(0.171840,
L(0.819257,
0.10)
0.20)
0.30)
0.43)
0.60)
0
0
0.091927)
0.458299)
0.161278)
0.729076)
0.762395)
0.154722)
0.794999)
0.160571)
0.828160)
0.180743)
1716
Table 6
Problem 5 C1/C2/N2 PREOS at p 76 bar and T 270 K
Feed composition
Stationary points of the TPD functiona
Objective function
FE
NT
Initial guessa
(0.10,
/
/
(0.30,
/
(0.38,
/
(0.05,
/
(0.101617, 0.586727)
(0.067797, 0.799168)
/
(0.245094, 0.658130)
/
(0.38, 0.54)
/
(0.05, 0.90)
/
5.88659e6
1.48288e2
/
0.011746
/
0
0
0
0
3079 (1814)
/
3474 (12)
4831 (16)
4779 (13)
2929 (18)
3006 (39)
4717 (19)
4727 (11)
2
/
1
1
1
1
1
1
1
L(0.118613,
/
V(0.055818,
L(0.373190,
V(0.181211,
L(0.491972,
V(0.242737,
L(0.121698,
V(0.045620,
0.60, 0.30)
0.55, 0.15)
0.54, 0.08)
0.90, 0.05)
0.380627)
0.853433)
0.360265)
0.766340)
0.356744)
0.725904)
0.611101)
0.912190)
x1, x2.
unstable for these conditions (TPD //6.914e/5).

Using L initialization, only one minimization is required. For V initialization, a local minimum with
TPD //4.008e/5 is first found. A second minimization finds the global solution. Compositions for the two
minima are close to the equilibrium liquid and vapor
phases. The two-phase flash gives L /0.534; the solution (Euclidean norm on fugacities of 4.3e/7) is found
with only 43 FE. The results are consistent with those
obtained for this mixture by Nagarajan et al. (1991);
Sun and Seider (1995). PR EOS has been used in
calculations, with BICs and properties of the two heavy
pseudocomponents as given by Nagarajan et al. (1991).
Fig. 3. Phase envelope of Y8 system.
4.3. Three-phase vapor/liquid /liquid equilibria
4.3.1. Ternary mixture CO2/C1/n C16

The first example for VLL calculations is for a ternary
mixture containing carbon dioxide, with methane and
normal-hexadecane in equimolar participation. At low
temperatures this mixture presents three-phase VLL
equilibria, and exhibit a Type II phase behavior
according to Orr, Yu and Lien (1980).
The PR EOS is used with BICs: kC1 C16 /0.078, kC1
CO2 /0.100, and kCO2 C16 /0.125 (from Pan & Firoozabadi, 1998). Fig. 5 depicts the calculated amount of
phases versus pressure for a 90% carbon dioxide mixture
at T /294.3 K. A three-phase region exists at this
temperature between a lower consolute pressure of 64.07
bar and an upper consolute pressure of 69.45 bar.
Results are in excellent agreement with results presented
by Risnes and Dalen (1984), using SSI and by Pan and
Firoozabadi (1998), using SA.
Table 8 presents the results (mol fractions) of VLLE
calculations for these conditions. Results of stability of
the feed (TPD//0.1663), two-phase flash, and stability of the equilibrium liquid from two-phase flash
(TPD//0.0097) are also given in Table 8. One can
notice that good initial estimates were available for the
VLL flash.
/
Fig. 4. Equilibrium constants for the Y8 system on the critical

isotherm.
4.2.2. Eight-component mixture

Calculations have been performed for the eightcomponent mixture of Nagarayan, Cullik and Griewank
(1991) at near critical conditions (T /353 K, p /385
bar). Results are given in Table 7.
For phase stability analysis of the feed, using both
initial estimates, the method finds the mixture to be
1717
Table 7
Stability testing and two-phase flash for the eight component mixture
Component
Feed composition
Stability
Two-phase flash
Mol fractions of trial phase

Type V initialization
C1
C2
C3
n C4
n C5
n C6
C7-16
C17
a
0.6883
0.0914
0.0460
0.0333
0.0139
0.0152
0.0896
0.0222
Vapor mol fraction
Liquid mol fraction
0.7177
0.0920
0.0447
0.0315
0.0128
0.0138
0.0741
0.0133
0.6628
0.0909
0.0471
0.0349
0.0148
0.0164
0.1032
0.0299
Type L initialization
0.6532
0.0902
0.0471
0.0351
0.0150
0.0170
0.1077
0.0349
0.7298
0.0917
0.0438
0.0304
0.0123
0.0131
0.0677
0.0112
Local minimum found by first minimization.
Fig. 5. Phase distribution of the ternary system CO2/C1/n C16 at T /

294.3 K.
Fig. 6. Detail of the p /z phase diagram of the ternary system CO2/C1/
n C16 at T/294.3 K.
Three-phase distribution calculated with Michelsen

(1994) procedure, using fugacities calculated for mol
fractions from stability, is: V /0.2466, L1 /0.6105,
L2 /0.1429. Phase distribution from three-phase VLL
flash is: V /0.2962, L1 /0.5645, L2 /0.1393.
A detail of the p /z phase envelope at T /294.3 K is
presented in Fig. 6, showing lower and upper consolute
pressures and critical points. Critical points are located
at 73.5 and 99.5% CO2. Construction of Fig. 6 does not
require three-phase VLLE calculations. For a given CO2
mol fraction, once a phase boundary is located, extra-
polation of TPD (for stability testing of liquid phase

resulting from two-phase flash) from three-phase region
gives a good approximation, then the exact pressure is
refined to the desired accuracy.
4.3.2. A sour gas system
The second example for VLLE is for the six component synthetic mixture of Robinson et al. (1978), which
exhibits a three-phase region at T /178.8 K for a wide
pressure range. Feed composition is given in Table 9.
Peng /Robinson EOS is used, with BICs from Pan and
Table 8
Results for CO2/C1/n C16 system at p 67 bar and T 294.3 K
Component
C1
n C16
CO2
a
Feed stability
a
0.05
0.05
0.90
0.040511
0.466839
0.492650
Trial phase composition.
Two-phase flash stability
Three-phase flash
a
0.050832
0.004166
0.945002
0.044778
0.337496
0.617726
0.099791
0.000048
0.900161
x1
x2
0.078112
0.000069
0.921819
0.036181
0.340224
0.623595
0.038707
0.004609
0.956683
1718
Table 9
Calculated mol fractions for the sour gas system at T 178.8 K,
p 21.9 bar
Component Feed composition
C1
C2
C3
H2S
CO2
N2
Phase mol fractions
0.70592
0.06860
0.02967
0.10559
0.01996
0.07026
HC-rich liquid
Vapor H2S-rich liquid
0.5282
0.1598
0.0760
0.1879
0.0371
0.0110
0.8577
0.0113
0.0006
0.0094
0.0082
0.1129
0.0666
0.0261
0.0075
0.8688
0.0304
0.0006
Table 10
BIC for the sour gas system
Component
C2
C3
H2S
CO2
N2
kC j
1
kC j
2
kC j
3
kH Sj
2
KCO j
0.000
/
/
/
/
0.000
0.000
/
/
/
0.100
0.080
0.080
/
/
0.125
0.135
0.150
0.120
/
0.031
0.042
0.091
0.200
0.020
Firoozabadi (1998) shown in Table 10. This is a difficult

problem due to the very small amounts in the hydrogen
sulfide-rich phase.
Calculated phase mol fractions at p /21.9 bar are
given in Table 9. The mol percent of different phases
versus pressure are plotted in Fig. 7.
For most of the points, the global minimum is found
by the first minimization. However, local minima are
found for both stability and flash calculations. For flash
calculations, local mimima are found only in the
immediate vicinity of the consolute pressures.
For example, stability of the liquid phase from twophase flash at p /30 bar: there are two local minima,
one nontrivial with positive TPD, and one trivial (both
wrongly indicating a stable phase). The convergence to
the global minimum corresponding to a negative value
of TPD is achieved by the third minimization phase, and
then the system is correctly asserted as being in threephase VLL state (third tunnelization).
Calculation results at p/35 bar are given in Table 11.
The TPD for two-phase flash liquid composition is /
0.0185, indicating a three-phase state. The phase distribution resulting from three-phase VLLE flash is V /
0.0407, L1 /0.9447, and L2 /0.0146.
4.3.3. Quaternary mixture C1/n C6/H2S/CO2
The third example for VLLE presents the results for
the quaternary mixture studied by Kohse and Heidemann (1992). Same input data and version of the SRK
EOS (Graboski & Daubert, 1978) were used. BICs are
given in Table 12. Three-phase flash calculations were
performed at T /175 K and p /15 bar (Table 13). For
these conditions, our results correspond to the global
Table 12
BIC for the quaternary system
Fig. 7. Phase distribution for the sour-gas system at T /178.8 K.
Component
C6
H2S
CO2
kC1-j
kC6-j
kH2S-j
0.036
/
/
0.080
0.050
/
0.093
0.118
0.099
Table 11
Results for the sour gas system at p 35 bar and T 178.8 K
Component
C1
C2
C3
H2S
CO2
N2
Feed
Stability
0.70592
0.06860
0.02967
0.10559
0.01996
0.07026
0.060303
0.011263
0.003438
0.904991
0.018229
0.001776
Two-phase flash
Stability
Three-phase flash
x1
x2
0.715787
0.069469
0.030069
0.093383
0.019979
0.071312
0.064426
0.012099
0.003697
0.899205
0.018734
0.001839
0.705964
0.006233
0.000489
0.008382
0.004765
0.274167
0.726129
0.006352
0.000466
0.008596
0.004933
0.253524
0.714973
0.072154
0.031333
0.097494
0.020621
0.063425
0.065262
0.012393
0.003738
0.897742
0.019166
0.001699
1719
Table 13
Results for the quaternary system at p 15 bar and T 175 K
Component
C1
n C6
H2S
CO2
Feed
Stability
0.50000
0.02627
0.41633
0.05740
0.082317
0.001380
0.878809
0.037836
Two-phase flash
Stability
Three-phase flash
x1
x2
0.979116
0.000001
0.007984
0.012899
0.118005
0.047214
0.741901
0.092880
0.286047
0.349603
0.293955
0.070395
0.977599
0.000001
0.007900
0.014499
0.308646
0.227252
0.363484
0.100618
0.068425
0.007764
0.833419
0.090391
minimum found by Sun and Seider (1995), with the

phase distribution: V /0.4482, L1 /0.1002, L2 /
0.4516.
At T /200 K and p/42.5 bar the system splits into
two phases, with L /0.6741. Stability analysis of the
liquid phase finds TPD /0. Same results were obtained
by Sun and Seider (1995). At slightly different conditions (T/199 K and the same pressure), the equilibrium
state is three-phase VLL. These are particularly difficult
conditions for this system, in the vicinity of a two-phase
liquid/liquid critical point located on the three-phase
boundary. The results are presented in Table 14. The
extent of the three-phase region for p /42.5 bar is about
5 K.
al. (1999) is used. BICs used with the PR EOS are given
in Table 15.
The p /T phase diagram of this mixture is given in
Fig. 8, showing the vapor /liquid and vapor /solid phase
boundaries as well as the vapor/liquid /solid/vapor/
solid and vapor /liquid/solid/vapor/liquid phase
boundaries. Results are identical with the results obtained with the SA method (Nichita et al., 2001).
Construction of the phase envelope does not require
carrying out three-phase VLS calculations.
For a multisolid model, the condition of solid stability
as a pure phase is obtained from Eq. (12) (Firoozabadi,
1998):
S
fi (p; T; z)fpure;
i ]0;
i 1; ns
(49)
4.4. Three-phase vapor /liquid /solid equilibria

Wax precipitation from hydrocarbon mixtures is an
important problem for the oil industry, which implies
vapor/liquid /solid equilibria calculations. Two examples, including vapor/liquid /solid equilibria, vapor/
solid, and liquid /solid equilibria for wax precipitation
are presented, one for a synthetic gas-condensate, the
other for a North Sea crude oil (at Stock-Tank and
reservoir conditions).
4.4.1. The SHF4 mixture
The first example is a synthetic gas condensate fluid of
Ungerer et al. (1995), designated by the acronym SHF4
(Synthetic Hyperbaric Fluid), containing seven components. The thermodynamic multisolid model for wax
precipitation from hydrocarbon mixtures of Nichita et
where ns is the number of components that may

precipitate.
This relationship is used to locate the V/VS phase
boundary (cloud point locus); it is also used to locate the
VL/VLS phase boundary, for z/x, where, x is the
liquid mol fraction resulted from the two-phase VL
flash.
For several conditions, local minima are observed
near the VL/VLS and V/VS phase boundaries. Table 16
gives the results obtained with the Tunneling method for
some of the points considered for the construction of the
phase diagram by Nichita, Goual and Firoozabadi,
2001. The total number of FE is reported, including
those required by the tunnelization phases. The number
of FE for the minimization step(s) is given in parenthesis. The computational effort required by the SA
algorithm is also given for comparison. For all points,
Table 14
Results for the quaternary system at p 42.5 bar and T 199 K
Component
C1
n C6
H2S
CO2
Feed
Stability
Two-phase flash
Stability
Three-phase flash
x1
x2
0.50000
0.02627
0.41633
0.05740
0.961720
0.000023
0.024470
0.013787
0.956278
0.000021
0.025186
0.018515
0.289556
0.038376
0.596733
0.075334
0.244579
0.029114
0.653089
0.073218
0.956229
0.000021
0.025179
0.018571
0.331272
0.046211
0.545958
0.076560
0.250611
0.030348
0.645240
0.073801
1720
Table 15
BIC for the SHF4 mixture
Component
C2
C3
C7H8
n C8
n C16
n C36
kC j
1
kC j
2
kC j
3
kC H8j
7
Kn C j
8
Kn C j
0.0000
/
/
/
/
/
0.0384
0.0417
/
/
/
/
0.0439
0.0000
0.0000
/
/
/
0.0475
0.0358
0.0221
0.0000
/
/
0.0658
0.0242
0.0104
0.0000
0.0002
/
0.0960
0.0050
0.0188
0.0000
0.0294
0.0172
16
Fig. 9. Phase compositions of SHF4 fluid at p/400 bar.

Fig. 8. Phase diagram of the SHF4 fluid.
TUNPEQ is at least one order of magnitude faster than

SA.
These minima are not a problem for the SA algorithm
if some code parameters are properly set (Pan &
Firoozabadi, 1998), but a local minimizer can be
attracted (this is the case of L-BFGS-B as shown in
Table 16). TUNPEQ is able to escape and finally finds
the global minimum. Results are identical with the

results obtained with SA.
As expected, only the heavy hydrocarbon component
(normal-hexatriacontane) precipitates. The variation of
the amount of precipitated solid and the molar fraction
of the liquid and vapor phases at p/400 bar are
illustrated in Fig. 9. A narrow three-phase region from
328 to 333 K exists at this pressure.
Table 16
VLS equilibria for SHF4 mixture
T (K), p (bar) conditions
Equilibrium state
Gibbs free energy function
FE SA
FE TUNPEQ
Tunneling phases
332, 600
335, 400
/
333, 400
332, 400
331, 400
/
330, 400
/
329.5, 400
329, 400
/
328.5, 400
328, 400
326, 400
325, 400
/
320, 400
VS
/
VL
VL
VLS
VLS
/
9.220000
8.683763
8.677826
8.659164
8.649628
8.645886
8.639555
8.635533
8.629042
8.623662
8.624830
8.618213
8.612708
8.607162
8.584619
8.580098
8.573146
8.513939
130 201
/
270 201
268 801
268 801
/
273 001
/
266 001
463 401
/
378 001
263 201
267 401
263 201
/
266 001
/
4869 (15)
/
20 181 (7916)
20 178 (62)
21 024 (67)
/
28 527 (7223)
/
26 333 (6613)
25 895 (92)
/
13 196 (130103)
14 596 (63)
12 613 (53)
13 821 (50)
/
12 265 (4358)
11 945 (58)
1
/
2
1
1
/
2
/
2
1
/
2
1
1
1
/
2
1
VLS
VLS
VLS
VS
VS
VS
VS
VS
1721
Fig. 10. Retrograde solid behavior and liquid dropout at T/330 K

for SHF4 fluid.
Table 17
Composition of crude oil 2
Component
Stock-Tank oil
Reservoir fluid
N2
CO2
C1
C2
C3
i C4
n C4
i C5
n C5
C6
C7
C8
C9
C10
/
/
/
0.00087
0.00536
0.00301
0.01240
0.01008
0.01730
0.03701
0.08213
0.10623
0.06672
0.65888
0.00225
0.01365
0.51283
0.07344
0.04430
0.00755
0.02037
0.00772
0.01097
0.01967
0.02524
0.03456
0.02101
0.20644
The retrograde solid behavior (reported by Nichita et

al., 2001) may imply the increase in the amount of the
precipitated solids with decreasing pressure at constant
temperature. Fig. 10 shows the calculated isothermal
expansion of fluid SHF4. At T /330 K the amount of
the precipitated solid decreases from about 13% (at the
three-phase dewpoint) to 3% (at p /185 bar). With
further pressure decrease the precipitation increases at
an increasing rate to over 15 wt.% at atmospheric
pressure.
4.4.2. Wax precipitation from a north sea crude oil

The second example deals with wax precipitation
from a crude oil. The composition of the Oil 2 (StockTank and reservoir fluid, see Table 17) as well as the
characterization of the plus fraction is given by Pan,
Firoozabadi and Fotland (1997). Again, the thermodynamic model, component properties, and BICs from
Nichita et al. (2001) are used.
Fig. 11a and b show the calculated amount of
precipitated wax versus temperature at p /1 and 120
Fig. 11. (a) Wax precipitation from Oil 2 at Stock-Tank conditions.

(b) Wax precipitation from Oil 2 at reservoir conditions.
bar, respectively. Results are given for both the multisolid (MS) and ideal solid solution (SS) models.
The number of FE for TUNPEQ (including the
Tunneling phase for checking global optimality) is one
order of magnitude less than for SA. The conditions for
this example are not difficult conditions, but what
makes it interesting is that the mixture is described
with a large components number (37). If only the
minimization step is used, only 15/47 FE were required
by L-BFGS-B to find the solution.
Wax precipitation is a major problem for the oil
industry, because wax deposition may plug the flow
lines. We believe such a code would be extremely
valuable for the industry. For wax precipitation from
crude oils, it is very important that the heavy fractions
distribution between species (paraffins, naphthenes, and
aromatics) is taken into account; this may lead to very
large number of components. The code has proved to be
able to handle such problems.
5. Some considerations on computational efficiency

Whenever data were available in the literature or
software was available for other methods, comparison is
given in terms of the number of FE. Some authors
1722
reports CPU time, but a comparison cannot be made

because one should have the same hardware support.
For the limited data we had, the Tunneling method has
shown to be faster than other global optimization
methods reported in the literature, for similar test
problems.
TUNPEQ is definitely more efficient than SA. For
phase stability analysis, it is more efficient than the
Enhanced-Newton-Interval method. It is expected to be
more efficient than branch and bound and homotopy
continuation methods. It would be very useful to test
TUNPEQ against MULPEQ, HOMPEQ, GLOPEQ,
etc. Further investigation of TUNPEQ efficiency is
needed.
A feature of our code is that the user may decide how
much computer time is to be spent on checking
globality. This depends on the particular problem to
be solved, mainly on how far we are from difficult
regions of the phase envelope. More initial points for the
Tunneling phase means more confidence, but also more
computer time. It is, however, essential to keep reliability by making meanwhile a particular run as efficient
as possible.
We have already embarked, or intend to embark on
the following directions:
. The global optimization Tunneling method used here

has shown to be a robust and efficient tool to solve
phase equilibrium problems even for extremely difficult cases, like calculations at critical points and in
the vicinity of the phase boundaries.
. The method checks global optimality, to be sure that
there is not a better solution and this checking is the
most expensive part of the algorithm. However, the
user can decide on the CPU effort spent for this
globality check.
. If only local optimization is performed, the results
show that the efficiency is comparable with classical
methods based on the equality of chemical potentials.
. The method has shown to be faster than other global
optimization methods reported in the literature, for
similar test problems.
. Our approach can be used as a powerful diagnostic
tool for reservoir compositional simulation and
process design models.
. The method is general and model independent
although an EOS formulation is used here for the
Gibbs free energy, any other thermodynamic model
can be used for various kinds of thermodynamic
equilibria.
i)
Acknowledgements
a systematic comparison of the efficiency of TUNPEQ with other global optimization methods used
for multiphase equilibria calculations;
ii) development of the code to be able to handle more
than two liquid phases;
iii) since the approach is model independent, the use of
different thermodynamic models will be investigated;
iv) testing the behavior of the method for a wider set of
phase equilibria problems.
The major feature of the method is its reliability.
There is no guarantee that a method would work for
every problem without failure, but so far we have not
experienced any failures with TUNPEQ. Therefore, our
approach is recommended for use as a powerful
diagnostic tool for reservoir compositional simulation
and process design models.
6. Conclusions
. A global optimization method has been developed for
multiphase equilibrium calculations (stability analysis
and phase splitting), and successfully tested for
various kinds of phase equilibria problems, both
hypothetical and real.
. In many of the addressed problems, several local
minima have been found, showing the need for global
optimization approaches.
Partial support for this work was provided by IIMAS,

Intercambio Academico, and project PAPIIT No.
IN11299, UNAM. Partial support from SNP Petrom
SA and the permission to publish this paper are greatly
appreciated. We thanks Dr Bruce Kohse of CMG for
providing the same input data as used by Kohse and
Heidemann (1992) for the quaternary system in example
Section 4.3.3.
References
Baker, L. E., Pierce, A. C. & Luks, K. D. (1982). Gibbs energy analysis
of phase equilibria , pp. 731 /742). SPE Journal .
Barron, C., & Gomez, S. (1991). The exponential Tunneling method,
Technical Report , IIMAS, vol . 1, No. 3.
Bullard, L. G. & Biegler, L. T. (1993). Iterated linear programming
strategies for non-smooth simulation: a penalty based method for
vapor /liquid equilibrium applications. Computers and Chemical
Engineering 17 (1), 95 /109.
Bunz, A. P., Dohrn, R. & Prausnitz, J. M. (1991). Three-phase flash
calculations for multicomponent systems. Computers and Chemical
Engineering 15 (1), 47 /51.
Byrd, R. H., Lu, P., Nocedal, J. & Zhu, C. (1995). A limited memory
algorithm for bound constrained optimization. SIAM Journal of
Science and Computers 16 , 1190 /1208.
Castellanos, L., & Gomez, S. (1998). An implementation of the
Tunneling methods for bound constrained global optimization,
Tehnical Report , IIMAS.
Coats, K. H. (1985). Simulation of gas condensate reservoir performance , pp. 1870 /1885). Journal of Petroleum .
Corana, A., Marchesi, M., Martini, C. & Ridella, S. (1987). Minimizing multimodal functions of continuous variables with the Simulated Annealing Algorithm. ACM Transactions on Mathematical
Software 13 , 265 /274.
DeGance, A. E. (1993). Ab initio equation of state phase equilibria
computations via path continuation. Fluid Phase Equilibria 89 ,
303 /334.
Eubank, P. T., Elhassan, A. E., Barrufet, M. A. & Whiting, W. B.
(1992). Area method for prediction of fluid phase equilibria.
Industrial Engineering and Chemical Research 31 , 942 /949.
Firoozabadi, A. (1998). Thermodynamics of hydrocarbon reservoirs .
McGraw-Hill.
Fussel, L. T. (1979). A technique for calculating multiphase equilibria ,
pp. 203 /210). SPE Journal .
Fussell, D. D. & Yanosik, J. L. (1978). An iterative sequence for phase
equilibria calculations incorporating the Redlich /Kwong equation
of state , pp. 173 /182). SPE Jounal .
Gibbs, J.W. (1873). A method of geometrical representation of the
thermodynamic properties of substances by mean of surfaces.
Trans. Conn. Acad. 2.
Gomez, S. & Romero, D. (1994). Two global methods for molecular
geometry optimization. In Progress in mathematics , vol. 121, pp.
503 /509). Birkhauser.
Gomez, S., Levy, A. V. & Calderon, A. (1987). A global zero residual
least squares method. In Lecture notes in mathematics , vol. 1230,
pp. 1 /10). Springer.
Gomez, S., Keeping, B. & Pantelides, C. (1998). Tunneling methods for
the time optimal control of high dimensional systems, manuscript .
London: Imperial College.
Gomez, S., Gosselin, O. & Barker, J. W. (2001). Gradient-based
history */matching with a global optimization method, paper SPE
71307 , pp. 200 /208). SPE Journal .
Gomez, S., Solano, J., Castellanos, L. & Quintana, M. I. (2001).
Tunneling and genetic algorithms for global optimization, advances in convex analysis and global optimization . Kluwer
Academic Publishers.
Graboski, M.S., & Daubert, T.W. (1978). A modified Soave equation
of state for phase equilibrium calculations IEC PDD, vol . 17, No.
4, 1. Hydrocarbon Systems (pp. 443 /448); 2. Systems Containing
CO2 , H2 S, N2 , and CO (pp. 448 /454).
Green, K. A., Zhou, S. & Luks, K. D. (1993). The fractal response of
robust solution techniques to the stationary point problem. Fluid
Phase Equilibria 84 , 49 /78.
Han, G. & Rangaiah, G. P. (1997). A method for calculation of
vapor /liquid and liquid /liquid equilibria. Computers and Chemical Engineering 21 (8), 905 /913.
Heidemann, R. A. & Khalil, A. M. (1980). The calculation of critical
points. American Institute of Chemical Engineering Journal 26 ,
770 /778.
Hua, J. Z., Brennecke, J. F. & Stadtherr, M. A. (1996). Reliable
prediction of phase stability using an interval Newton method.
Fluid Phase Equilibria 116 , 52 /59.
Hua, J. Z., Brennecke, J. F. & Stadtherr, M. A. (1998). Enhanced
interval analysis for phase stability: cubic equation of state models.
Industrial Engineering and Chemical Research 37 , 1519 /1527.
Hua, J. Z., Maier, R. W., Tessier, S. R., Brennecke, J. F. & Stadtherr,
M. A. (1999). Interval analysis for thermodynamic calculations in
process design: a novel and completely reliable approach. Fluid
Kohse, B. F. & Heidemann, R. A. (1992). Tricritical lines and
multiphase equilibria in quaternary mixtures. Fluid Phase Equilibria 75 , 11 /22.
Levy, A. V. & Gomez, S. (1985). The Tunneling method applied to
global optimization. In P. T. Boggs, R. H. Byrd & R. B. Schnabel
(Eds.), Numerical optimization , pp. 213 /244). SIAM.
Levy, A.V., & Montalvo, A. (1985). The Tunneling method for global
optimization. SIAM Science and Computers .
1723
Lucia, A., Guo, X., Richey, P. J. & Derebail, R. (1990). Simple process
equations, fixed-point methods, and chaos. American Institue of
Chemical Engineering Journal 36 , 641 /654.
McDonald, C. M. & Floudas, C. A. (1995). Global optimization and
analysis for the Gibbs free energy function using the UNIFAC,
Wilson, and ASOG equations. Industrial Engineering and Chemical
Research 34 , 1674 /1687.
McDonald, C. M. & Floudas, C. A. (1995). Global optimization for
the phase and chemical equilibrium problem: application to the
NRTL equation. Computers and Chemical Engineering 19 (11),
1111 /1139.
McDonald, C. M. & Floudas, C. A. (1995). Global optimization for
the phase stability problem. Thermodynamics 41 , 1798 /1814.
McDonald, C. M. & Floudas, C. A. (1997). GLOPEQ: a new
computational tool for the phase and chemical equilibrium
problem. Computers and Chemical Engineering 21 (1), 1 /23.
Mehra, R. K., Heidemann, R. A. & Aziz, K. (1982). Computation of
multiphase equilibrium for compositional simulation , pp. 61 /68).
SPE Journal .
Michelsen, M. L. (1982). The isothermal flash problem. Part I.
Stability. Fluid Phase Equilibria 9 , 1 /19.
Michelsen, M. L. (1982). The isothermal flash problem. Part II. Phase
splitting. Fluid Phase Equilibria 9 , 21 /40.
Michelsen, M. L. (1986). Simplified flash calculations for cubic
equations of state. Industrial Engineering and Chemical Process
Design and Development 25 , 184 /188.
Michelsen, M. L. (1994). Calculation of multiphase equilibrium.
Computers and Chemical Engineering 18 (7), 545 /550.
Nagarayan, N. R., Cullik, A. S. & Griewank, A. (1991). New strategy
for phase equilibrium and critical point calculations by thermodynamic energy analysis. Part I. Stability analysis and flash. Fluid
Nghiem, L. X. & Li, Y.-K. (1984). Computation of multiphase
equilibrium phenomena with an equation of state. Fluid Phase
Equilibria 17 , 77 /95.
Nichita, D.V. (1996). Ph.D. dissertation, University of Ploiesti.
Nichita, D. V., Minescu, F. & Cretu, I. (1996). A new algorithm for the
calculation of mixtures critical point. Romanian Petroleum Journal
(Revista Romana de Petrol in Romanian) 3 , 197 /210.
Nichita, D.V., Goual, L., & Firoozabadi, A. (2001). Wax precipitation
in gas condensate mixtures. SPE Production and Facilities (pp.
250 /259).
Orr, F.M., Jr, Yu, A.D., & Lien, C.L. (1980). Phase behavior of CO2
and crude oil in low temperature reservoir, paper SPE 8813
presented at the first SPE/DOE symposium on enhanced oil
recovery. Tulsa, OK.
Pan, H., & Firoozabadi, A. (1998). Complex multiphase equilibrium
calculations by direct minimization of Gibbs free energy by use of
simulated annealing. SPE Reservoir Evaluation and Engineering
(pp. 36 /42).
Pan, H., Firoozabadi, A., & Fotland, P. (1997). Pressure and
composition effect on wax precipitation: experimental data and
model results. SPE Production and Facilities (pp. 250 /258).
Peng, D. Y. & Robinson, D. B. (1976). A new two-constant equation
of state. I.&E.C. Fundamentals 15 (1), 59 /64.
Prausnitz, J. M., Lichtenthaler, R. N. & Gomez de Azevedo, E. (1986).
Molecular thermodynamics of fluid-phase equilibria . Englewood
Cliffs, NJ: Prentice-Hall.
Press, W. H., Flannery, B. P., Teukolski, S. A. & Vetterling, W. T.
(1989). Numerical recipes */the art of scientific computing . Cambridge University Press.
Risnes, R. & Dalen, V. (1984). Equilibrium calculations for coexisting
liquid phases , pp. 87 /96). SPE Journal .
Robinson, D.B., Kalra, H. & Rempis, H. (1978). The equilibrium phase
properties of a synthetic sour gas mixture and a simulated natural gas
mixture, GPA Research Report , No. RR-31, Tulsa.
1724
Soave, G. (1972). Equilibrium constants from a modified Redlich /

Kwong equation of state. Chemical Engineering and Science 27 (6),
1197 /1203.
Sun, A. C. & Seider, W. D. (1995). Homotopy-continuation method
for stability analysis in the global minimization of the Gibbs free
energy. Fluid Phase Equilibria 103 , 213 /249.
Trangenstein, J.A. (1985). Minimization of Gibbs free energy in
compositional reservoir simulation, paper SPE 13520, presented
at the SPE 1985 Reservoir Simulation Symposium, Dallas, TX.
Ungerer, P., Faissat, B., Leibovici, C., Zhou, H., Behar, E., Moracchini, G. & Courcy, J. P. (1995). High pressure /high temperature reservoir fluids: investigation of synthetic condensate gases
containing a solid hydrocarbon. Fluid Phase Equilibria 111 , 287 /
311.
Vickery, D. J. & Taylor, R. (1986). Path-following approaches to the
solution of multicomponent, multistage separation process problems. American Institute of Chemical Engineering Journal 32 (4),
547 /556.
Wasylkiewicz, S. K., Sridhar, L. N., Doherty, M. F. & Malone, M. F.
(1996). Global stability analysis and calculation of liquid /liquid
View publication stats
equilibrium in multicomponent mixtures. Industrial Engineering

and Chemical Research 35 , 1395 /1408.
Wilson, G.A. (1969). Modified Redlich /Kwong equation of state,
application to general physical data calculations, paper no. 15C
presented at the American Institute of Chemical Engineering 65th
National Meeting, Cleveland, OH.
Yarborough, L. (1972). Vapor /liquid equilibrium data for multicomponent mixtures containing hydrocarbon and non-hydrocarbon components. Journal of Chemical Engineering Data 17 , 129 /
133.
Zhu, Y. (2000). High pressure phase equilibrium through the simulated
annealing algorithm: application to SRK and PR equations of
state, paper presented to the 2000 American Institute of Chemical
Engineering Annual Meeting.
Zhu, C., Byrd, R. H., Lu, P. & Nocedal, J. (1994). L-BFGS-BFORTRAN subroutines for large-scale bound constrained optimization . Department of Electrical Engineering and Computer Science,
Northwestern University.

Multiphase Equilibria Calculation by Direct Minimization

Diunggah oleh

Informasi Dokumen

Hak Cipta

Format Tersedia

Bagikan dokumen Ini

Bagikan atau Tanam Dokumen

Opsi Berbagi

Apakah menurut Anda dokumen ini bermanfaat?

Apakah konten ini tidak pantas?

Hak Cipta:

Format Tersedia

Multiphase Equilibria Calculation by Direct Minimization

Diunggah oleh

Hak Cipta:

Format Tersedia

See

Multiphase Equilibria Calculation by Direct

French National Centre for Scientific Research

Universidad Nacional Autnoma de Mxico

61 PUBLICATIONS 640 CITATIONS

94 PUBLICATIONS 1,085 CITATIONS

Differentiation of noisy data with Total Variation Method View project

Computers and Chemical Engineering 26 (2002) 1703 /1724

Multiphase equilibria calculation by direct minimization of Gibbs free

PETROM-Institute for Research and Technology (ICPT), Campina, Romania

subspace of constraints in Rn defined by inequalities

Currently with University of Manchester.

fugacity of component i in phase j

Trangenstein (1985) formulated the two-phase

2. Statement of the phase equilibria problem

2.1. Multiphase equilibria calculation by direct

nji mji (nj )

where n/(n1, n2, . . ., nnp)T is a vector of nc /np

nji ln fji (xj )

where G is the reduced Gibbs free energy.

such that the component material balance constraints:

and bounds of independent variables:

where ni /Fzi are the feed component mol numbers for

In Eq. (2), one can eliminate the mol numbers of one

The gradient of the objective function is:

The expression of the reduced TPD in terms of

xi [ln fi (x)ln fi (z)]

If one of the mol fractions is eliminated:

According to the Gibbs /Duhem equation:

the optimization problem to be solved is:

for nc/1 mol fractions as independent variables

This is an unconstrained optimization problem. The

2.2. Phase stability analysis by minimization of the

xi [mi (x)mi (z)]

The TPD function is given by the difference between

Convergence to the correct solution of the problem

5) results from step 3 and 4 are used to initiate the

[xji (b)yi (b)];

Taking into account that:

2.4.1. Initialization of phase splitting calculations

The problem is to find the phase distribution. This

with the overall material balance:

The definition relationship for the equilibrium

we obtain a set of np/1 equations with the

[xji (b)x1i (b)];

and solved for the unknowns b2, . . ., bnp.

By combining Eqs. (22) /(24), the mol fraction of

2.5. Convergence criteria

and the fugacity coefficients are calculated using

The gradient vector is given by:

There are two stopping criteria for the minimization

where epsmch is the machine precision (automatically

The objective function is a convex function

We have set a high restrictive value for the tolerance

3. The Tunneling global optimization method

where estimates for equilibrium constants are obtained

A slightly modified initialization is used in this work.

The aim of the global optimization method is to

with B /{x jl 5/x 5/u }, f:Rn 0/R , x  /Rn , f  /C2, n being

Fig. 1. The basic idea of the Tunneling method.

3.1. The global optimization method

where xG is the global minimum of f(x ), ignoring all the

To perform the minimization phase any algorithm

3.2.2. Tunneling phase

where n/(n1, n2, . . ., nnp)T is a vector of nc /np

with B /{x jl 5/x 5/u }, f:Rn 0/R , x /Rn , f /C2, n being