Fluid Phase Equilibria 433 (2017) 56e66

Contents lists available at ScienceDirect

Fluid Phase Equilibria

j o u r n a l h o m e p a g e : w w w . e l s e v i e r . c o m / l o c a t e / fl u i d

Stability analysis for multicomponent mixtures including capillary

pressure
Marjan Sherafati, Kristian Jessen*
Mork Family Department of Chemical Engineering and Materials Science, University of Southern California, Los Angeles, CA, 90089-1211, USA

a r t i c l e i n f o a b s t r a c t

Article history: Despite the large potential of unconventional resources, many unknowns still exist regarding the physics
Received 18 August 2016 controlling the extraction processes in these settings. These include accurate representation of phase
Received in revised form equilibrium in tight formations and effective implementation of relevant models in simulation tools.
14 October 2016
When a fluid is confined in pore spaces of nanometer size, significant interfacial curvatures may occur
Accepted 11 November 2016
that can result in large capillary pressures between the liquid and vapor phases: The pressure difference
Available online 14 November 2016
between the two phases will likely affect the vapor-liquid equilibrium state. Previous efforts have shown
that this effect is negligible for conventional reservoirs (with pores in the micron range) and current
Keywords:
Stability analysis
commercial reservoir simulators commonly ignore the effect of capillary pressure in the VLE calculations.
Capillary pressure However, experimental and modeling efforts have shown that ignoring capillary pressure in the VLE
Algorithms calculations will not be a valid approximation for unconventional (tight) reservoirs.
Convergence behavior In this work, we analyze the numerical aspects of including capillarity phenomena in VLE calculations
in an effort to arrive at robust and efficient algorithms for stability analysis that can be used in
compositional modeling/simulation of unconventional reservoirs. While the equality of chemical po-
tentials is a necessary condition for equilibrium, it is not a sufficient one. A sufficient condition for
equilibrium is the minimization of Gibbs energy, and the latter can be tested using the tangent plane
distance (TPD) criteria. We show that stability analysis based on the TPD criteria remains valid for sys-
tems with large capillary pressures and propose effective/robust algorithms for stability testing. The
proposed algorithms are tested for multicomponent reservoir fluid systems over a range of relevant
temperatures, pressures and pore radii.
© 2016 Elsevier B.V. All rights reserved.

1. Introduction confined spaces, e.g. in pores at the nanometer scale, and to

Recent advances in drilling/completion technology have
enabled hydrocarbon extraction from tight unconventional re-
sources such as shale gas, liquid-rich shales and tight oil systems.
These resources currently contribute significantly to the total oil
and gas production in the US. Hydrocarbon in organic-rich shales is
one of the most significant unconventional resources [1]. Despite
the large potential and the advancement in production technology
for these resources, gaps still exist between the physical models
used in currently available simulation tools and the well-
documented additional complexity of fluid flow and mass trans-
fer in micro- and meso-porous materials. A central challenge is
related to the understanding and modeling of phase equilibrium in
* Corresponding author.
E-mail address: jessen@usc.edu (K. Jessen).
develop appropriate and accurate tools/algorithms to be used for
estimation of reserves and for forecasting of production [2]. Con-
ventional experimental PVT analysis is usually performed based on
the assumption that the porous medium does not influence the
phase behavior [3]. However, fluid properties and VLE behavior in
confined spaces depart substantially from the corresponding bulk
measurements (PVT cell) where vapor-liquid interface curvature
can be neglected [4].
When the fluid is confined in small pores, significant interfacial
curvatures can arise that results in large a capillary pressure be-
tween the equilibrium liquid and vapor phases: A pressure differ-
ence that is known to impact the VLE. It has, however, been
demonstrated previously that this effect is negligible for conven-
tional reservoirs with pore spaces in the micron range [5,6].
Current commercial reservoir simulators commonly ignore the
effect of capillary pressure in VLE calculations. However, experi-
mental and modeling work presented over the past few years

http://dx.doi.org/10.1016/j.fluid.2016.11.013
0378-3812/© 2016 Elsevier B.V. All rights reserved.
 M. Sherafati, K. Jessen / Fluid Phase Equilibria 433 (2017) 56e66 57

[3,4,7e10] have shown that neglecting capillary pressure in VLE cal- simplicity. We also assume, by default, that the rock is strongly
culations is not a good assumption for unconventional reservoirs. liquid wet ðq ¼ 0Þ. We use the correlation of Macleod [13] and
Furthermore, a standard and reliable measurement of fluid properties Sugden [14] to calculate interfacial tension (IFT),
and VLE for confined fluids in ultra-tight rocks is still not available [3].
" #E
Numerous researchers have worked on analyzing phase enve- X
nc  
lopes of confined fluids. Nojabaei et al. [8] coupled capillary pres- s¼ ci xi rL  yi rV ; (4)
sure and phase equilibrium calculations for binary and i¼1

multicomponent mixtures in confined pores and integrated their

where rL and rV are phase molar densities, xi and yi are phase
model into a compositional reservoir simulator. Based on example
compositions, ci is the component parachor and E is an (adjustable)
calculations, they conclude that better estimates of the producing
exponent. Researchers have suggested a variety of values for the
gas-oil ratio (GOR) are achieved when the capillary pressure is
exponent E. Weignaug and Katz [15] suggested that E ¼ 4, while
included in the equilibrium calculations. Du and Chu [3] proposed a
Hough-Stegemeier [16] modified the exponent to E ¼ 3.66 for low
thermodynamic model to calculate the confined PVT properties for
IFT systems. In 1984, Lee and Chien [17] suggested a value of
Bakken crude oil systems considering the effects of capillarity in the
E ¼ 3.91 using critical scaling theory of IFT to evaluate the de-
porous medium. Wang et al. [4] evaluated the effect of capillarity on
pendency on temperature. Danesh et al. [18] presented a density-
VLE calculations using a Leverett J-function approach to approxi-
dependent model for exponent E. Schechter and Guo [19] recom-
mate a saturation dependent capillary pressure. Alharthy et al. [11]
mended E ¼ 3.88 based on experimental data for hydrocarbon
proposed a correlation to shift of the phase envelope depending on
systems. In this work, we use E ¼ 4 (following Weignaug and Katz).
pore sizes, dividing the pore space into macro-, meso- and micro-
From Eq. (4), one can calculate the interfacial tension, and hence the
porosity, while Sandoval et al. [2] proposed a new algorithm to
capillary pressure, for a given set of phase compositions and den-
calculate phase envelopes including the effect of capillary pressure.
sities (pressures) of liquid and vapor phases given a pore radius and
Despite a substantial amount of research on the topic of phase
a contact angle. This allows for a simple algorithm to update the
behavior of confined fluids, the effect of capillarity in stability
phase pressures by direct substitution as discussed in more detail
analysis has not been extensively studied/documented to the best
later.
of our knowledge. In this work, we will start from Gibbs energy and
the requirement for phase stability and introduce modifications
needed to account for capillarity in stability analysis. We propose 2.1. Formulation of the stability analysis problem
four algorithms for stability testing: Two based on direct/acceler-
ated substitution and two based on minimization. The convergence The tangent plane distance (TPD) of Gibbs has been used
behavior of the proposed algorithms is illustrated for a range of repeatedly in developing algorithms for stability analysis of un-
relevant fluid systems and proximity to critical points. The stability confined fluids [20]. To apply this idea for confined fluids, we make
testing is demonstrated to be consistent with the phase boundaries the assumption that the volume (or mass) contribution to changes
predicted by phase envelope calculations including capillarity. in Gibbs free energy is larger than the interface contribution. In
Accordingly, the proposed algorithms provide for reliable tools to terms of nucleation theory, this is equivalent to assuming that the
study the effect of capillary pressure on phase stability over a wide volume (radius) of any trial phase exceeds the radius of a critical
range of compositions and capillary radii as needed in composi- nucleus. Based on this assumption, we can write the TPD at
tional simulation of unconventional reservoirs/formations. isothermal conditions in the following form

X
nc  
2. Methodology TPDðwÞ ¼ wi mw w z z
i ðw; p Þ  mi ðz; p Þ ; (5)
i¼1
In this section, we discuss the stability analysis of a confined
fluid and the related impact of capillary pressure. Starting from where z is the feed composition existing at pressure pz and w is a
Helmholtz energy, one can derive the equilibrium condition, in trial phase composition existing at pw. The assumption of neglect-
terms of the chemical potential, m, for a system that includes the ing the interfacial contribution to changes in Gibbs energy, as
contribution from surface energy [12]. outlined above, is consistent with the formulation of Sandoval et al.
[2] for tracing phase boundaries. This is seen from Eq. (5) where the
mli ðT; px ; xÞ ¼ mvi ðT; py ; yÞ; i ¼ 1; ::; nc ; (1) TPD will equal zero on the phase boundary where the incipient
phase (w) exists is in equilibrium with the bulk phase (z).
l v
In Eq. (1), the phase pressures (p and p ) are related through the At this stage, we can follow the approach of Michelsen [20], and
Young-Laplace equation (written for a spherical interface) introduce a modified TPD via the variable transformation

2s lnWi ¼ lnwi  k; (6)

pv  pl ¼ : (2)
R
where k is the reduced TPD at the stationary points of Eq. (5). The
The interfacial tension, s, is a given function of phase compo-
modified reduced TPD, tm, for a system including capillary pressure
sitions (x and y), temperature, T, phase pressures and the radius of
can then be written in terms of fugacity coefficients
curvature of the interface, R. In a capillary tube (pore) the capillary
pressure, Pc, is evaluated from
X
Nc  
tmðWÞ ¼ 1 þ bw
Wi ln Wi þ ln 4 w
i þ ln p  di  1 ; (7)
2s cos q
Pc ¼ pv  pl ¼ ; (3) i¼1
rc
with
where rc is pore radius and q is the contact angle as measured
through the denser phase. Although porous rocks are made up by b zi þ ln pz ;
di ¼ ln zi þ ln 4 i ¼ 1; ::; Nc (8)
complex networks of non-cylindrical pores of variable radii, we use
here a single value (average) representation of the pore radius for and
58 M. Sherafati, K. Jessen / Fluid Phase Equilibria 433 (2017) 56e66

Eigenvalue Method) of Orbach and Crowe [21] with extrapolation

pw  pz ¼ ±Pc :
If the feed composition (z) is assumed to be a liquid and the trial
phase (w) is vapor, the ± sign in Eq. (9) is positive and if the feed
composition is vapor and the trial phase is liquid, the sign becomes
negative (assuming that the liquid is the wetting phase). The focus
of stability analysis is to find the stationary points of Eq. (7). If the
value of the tm is greater than zero for all stationary point, the feed
composition is stable, and if not, we have demonstrated that the
feed is unstable and will form at least two phases. In order to find
the stationary points of Eq. (7), we need to solve
(9) attempted after each five direct substitutions.
Direct substitution and accelerated direct substitution are
rapidly convergent for most relevant settings. However, near crit-
ical points the convergence rates can be very slow and a second
order method may be preferred. To formulate a second order
convergent method, we use the variable transformation suggested
by Michelsen [20] and arrive at the following set of independent
variables
h pffiffiffiffiffiffiffiffi pffiffiffiffiffiffiffiffi pffiffiffiffiffiffiffiffiffi iT
v ¼ 2 W1 ; 2 W2 ; ::; 2 Wnc : (17)

vtm The gradients of Eq. (7) with respect to vi is written as

¼ 0; i ¼ 1; ::; nc ; (10)
vWi
vtm pffiffiffiffiffiffi 
gi ¼ bw
¼ Wi ln Wi þ ln 4 i  di þ ln p
w
; (18)
corresponding to the following system of nonlinear equations vvi

bw
ln Wi þ ln 4 w
i  di þ ln p ¼ 0; i ¼ 1; ::; nc : (11) while the Hessian matrix is calculated as

v2 tm qffiffiffiffiffiffiffiffiffiffiffiffi vlnb
4w 1 gi
i
Hij ¼ ¼ dij þ Wi Wj þ d i; j ¼ 1; ::; Nc :
vvi vvj vWj 2 vi ij
2.2. Solution methods (19)
The stationary points of Eq. (7) can be found via the restricted
To find the stationary points of the modified tangent plane
step algorithm suggested by Hebden [22].
distance, Eq. (11) must to be solved for the independent variables
(W, pw). Due to its simplicity, direct substitution is an obvious ðH þ bIÞDv þ g ¼ 0; (20)
choice for solving the above equations where subsequent iterates of
W are determined from to handle cases where the Hessian matrix is indefinite [20]. In Eq.
(20), b is found iteratively to arrive at jjDvjj equal to a specified
ln Witþ1 ¼ di  ln b w;t
4 i
w;t
 ln p ; i ¼ 1; ::; nc : (12) maximum step length (see Hebden [22] for additional discussion of
step adjustments). In this formulation, the trial-phase pressure is
The trial phase pressure (pw) is initialized as equal to the feed
not included as an independent variable and must be updated
pressure (pz). In subsequent iterations, the capillary pressure is
separately via direct substitution (as discussed above) or by a
calculated from Eq. (3) based on phase compositions and the phase
Newton iteration on the residual form of Eq. (9). Here, we use the
pressures. The trial phase pressure is then updated accordingly by
Newton update of the trial-phase pressure to promote convergence
direct substitution. It is important to note that the capillary pres-
of the overall (sequential) scheme. We note that the Newton update
sure is calculated at the end of the iteration, such that the capillary
only requires the pressure derivative of the molar volume of the
pressure updates are lagging one iteration behind the direct sub-
trial phase (available from the calculation of fugacity coefficients).
stitution of the trial-phase compositions.
Alternatively, the pressure of the trial phase can be treated as an
In order to locate the stationary points of the modified tangent
independent variable by forming a new variable vector
plane distance, it is necessary to test multiple initial estimates. For
two-phase VLE problems, we select two estimates: One vapor-like h pffiffiffiffiffiffiffiffi pffiffiffiffiffiffiffiffi pffiffiffiffiffiffiffiffiffi iT
initial estimate and one liquid-like. Wilson's equation provides a s ¼ 2 W1 ; 2 W2 ; ::; 2 Wnc ; pw : (21)
reasonable approximation for initial estimates
With this set of independent variables, we define a new set of
  
Pc;i Tc;i equations to be solved as follows
Ki ¼ z exp 5:42ð1 þ ui Þ 1  ; (13)
p T 
gi i ¼ 1; ::; Nc
FðsÞ ¼ (22)
where Pc;i ; Tc;i and ui are the critical properties and acentric factor Pc ±ðpw  pz Þ i ¼ Nc þ 1
of the component i and Ki is the equilibrium constant. The vapor-
The Jacobian of the above function can be defined as follows
like and liquid-like initial estimates (W) are then calculated from
8
>
> Hij i ¼ 1; ::; Nc ; j ¼ 1; ::; Nc
Wi ¼ Ki zi ; (14) >
>  
>
>
> pffiffiffiffiffiffi
>
> W
vlnb4w i
þ
1
i ¼ 1; ::; Nc ; j ¼ Nc þ 1
and >
> i w w
>
< vp p
Ji;j ¼ qffiffiffiffiffiffi vP c (23)
Wi ¼ zi=K : (15) >
> Wj i ¼ Nc þ 1; j ¼ 1; ::; Nc
i >
> vWj
>
>
>
>
To study the convergence behavior, we define an error for the >
> c
>
> vP
trial-phase composition as the norm of the vector e : ±1 i ¼ Nc þ 1; j ¼ Nc þ 1
vpw
bw
ei ¼ ln Wi þ ln 4 w
i  di þ ln p ; (16) Hebden's restricted-step algorithm can again be used to solve
the above set of equations
and consider the solution converged when the jjejj is less than 108 .
To enhance the convergence behavior of the direct substitution ðJ þ bIÞDs þ F ¼ 0: (24)
method, we utilize the acceleration approach (Dominant
 M. Sherafati, K. Jessen / Fluid Phase Equilibria 433 (2017) 56e66 59

In the formulation of Eq. (22) and Eq. (23), both the tangent Table 2
plane distance and the error in the capillary pressure are mini- Non-zero binary interaction coefficients.

mized simultaneously. However, the Jacobian matrix (Eq. (23)) is N2 CO2 CH4
less attractive to evaluate, due to the dependency of capillary N2
pressure (via IFT) on phase compositions and pressures. In addition, CO2 0.017
the factorization of the J is more CPU intensive than the factoriza- CH4 0.0311 0.12
tion of H (non-symmetric versus symmetric matrix). We discuss C2 0.0515 0.12
C3 0.0852 0.12
this further in the result section.
iC4 0.1033 0.12
In the following section, we compare the performance of the 4 nC4 0.08 0.12
different algorithms discussed above including: a) Direct substi- iC5 0.0922 0.12
tution, b) Accelerated direct substitution, c) Sequential minimiza- nC5 0.1 0.12
C6 0.08 0.12
tion of TPD with separate Newton update of trial-phase pressure
PS1 0.08 0.1 0.028349
and d) Minimization of the TPD with a full set of independent PS2 0.08 0.1 0.044813
variables. PS3 0.08 0.1 0.062256
PS4 0.08 0.1 0.077679
3. Results PS5 0.08 0.1 0.0952

First we introduce the fluid description used in our example

calculations and report the relevant phase envelopes. We then
report results from stability analysis calculations with emphasis on
convergence behavior near phase boundaries of oil and condensate
systems. Finally, we consider the effects of feed composition and
pore radius on phase boundaries and stability analysis calculations.
the critical point, the two curves converge, since the two phases
become identical (at the critical point) and the interfacial tension
approaches zero. In the dew point region, the saturation pressure is
increased by capillary effects and the cricondetherm is shifted to
the right.

3.1. Fluid description 3.3. Stability testing

In this study, we use a 15-component fluid description that is
representative of a sample from a realistic reservoir fluid. Table 1
lists the components, their critical properties, acentric factors,
molecular weights, volume shift parameters, parachors and
composition of oil and condensate systems. The last 5 components
of the fluid composition are pseudo-components created during
characterization of the C7þ fraction of the reservoir fluid. The
Soave-Redlich-Kwong (SRK) equation of state was used in all cal-
culations. Table 2 reports the non-zero binary interaction co-
efficients for the components.
3.2. The phase envelope
We first present the phase envelope for the 15 component fluid
using the oil feed composition. The phase envelope was initially
constructed for two scenarios: a) without confinement and b) with
confinement corresponding to an average pore size of 10 nm. The
construction of phase envelopes follows the work of Sandoval et al.
[2]. Fig. 1 shows the phase envelopes and demonstrates that
capillary pressure effects suppress the bubble point pressures for
the entire temperature range of the bubble point branch. Close to
The phase envelope constructed in Fig. 1 serves as the basis for
testing the stability analysis algorithms outlined above. The critical
temperature and pressure of the fluid sample shown in Fig. 1 are
721.38 K and 113.16 atm, respectively. Stability analysis was per-
formed along constant temperature lines (383.15 K, 700 K and
780 K) to identify the lowest pressure at which the feed composi-
tion is stable at a given temperature (last positive/zero value of tm):
This pressure must be identical to the saturation pressure dictated
by the phase envelope at that temperature. Fig. 1 reports the results
of stability analysis for three different regions of the phase enve-
lope. At each temperature, the lowest pressure where the given
feed is stable is marked with a black square. First, stability analysis
was performed on the bubble-point side of the PT phase envelope
(T ¼ 383.15 K). The highest pressure corresponding to a negative
tangent plane distance is in agreement with the suppressed liquid
phase saturation pressure at the given temperature. The stability
analysis indicates, as expected, an unstable vapor-like trial phase.
Next, we choose a point relatively close to the critical point
(T ¼ 700 K). In this region, the direct substitution method takes
considerably longer to converge compared to the minimization
approaches. The vapor-like trial phase converges to a negative

Table 1
Fluid properties and compositions.

Comp Tc [K] Pc [atm] Acentric factor [-] MW [g/mole] Volume shift [-] Parachor Oil composition Condensate composition

N2 126.2 33.6 0.04 28.016 0.1927 43.85 0.0018 0.0106

CO2 304.2 72.9 0.228 44.01 0.0817 82.24 0.0082 0.0158
CH4 190.6 45.4 0.008 16.043 0.1595 71.52 0.2292 0.7702
C2 305.4 48.2 0.098 30.069 0.1134 113.71 0.0721 0.0939
C3 369.8 41.9 0.152 44.096 0.0863 151.14 0.0737 0.0522
iC4 408.1 36.0 0.176 58.123 0.0844 181.78 0.0158 0.0073
nC4 425.2 37.5 0.193 58.123 0.0675 191.03 0.0523 0.0202
iC5 460.4 33.4 0.227 72.15 0.0608 224.14 0.0225 0.0056
nC5 469.6 33.3 0.251 72.15 0.039 231.73 0.036 0.0078
C6 507.4 29.3 0.296 86.177 0.008 274.03 0.0484 0.0061
PS1 565.85 29.32 0.34612 120.312 0.01648 331.01 0.196107 0.0099
PS2 683.44 19.84 0.57838 207.159 0.04094 546.37 0.113893 0.0004
PS3 798.99 13.15 0.90175 354.274 0.0616 884.73 0.066598 3.041E-6
PS4 899.68 9.57 1.19183 574.812 0.05768 1391.97 0.041047 1.298E-8
PS5 1013.31 7.57 1.39383 1055.440 0.01401 2497.41 0.022355 1.987E-11
60 M. Sherafati, K. Jessen / Fluid Phase Equilibria 433 (2017) 56e66
140
normal envelope
PLiquid
120
PGas
100 unstable feed
Pres s ure (atm)
80
60
40
20
0
100 200 300
Fig. 1. Phase envelope of the 15-component oil with and without capillary pressure (rc ¼ 10 nm). Triangle ¼ critical point.
value of TPD, while the liquid-like trial phase arrives at the trivial
solution. Finally, we perform stability analysis on the dew-point
branch of the envelope (T ¼ 780 K), close to the cricondentherm.
The highest pressure resulting in a negative TPD is again in agree-
ment with the relevant phase envelope. Here, the liquid-like trial
phase results in a negative TPD. Fig. 2 reports the reduced TPD
(tpd ¼ TPD/RT) at the stationary point as a function of pressure
along constant temperatures of T ¼ 383.15 K, T ¼ 700 K and
T ¼ 780 K.
Let us first examine the tangent plane distance for T ¼ 383.15 K
on the bubble point side, with liquid as the reference phase and
vapor as the trial phase. Starting at a pressure higher than the
saturation pressure, tpd is positive at the stationary point before
reaching the saturation pressure (tpd ¼ 0), indicating a stable feed.
As we cross the phase boundary, tpd at the stationary points be-
comes negative and indicative of an unstable feed. By further
reducing the pressure we find even lower values of the tpd at the
stationary points inside the two-phase region of the phase
envelope.
Near the critical point (T ¼ 700 K), and at pressure above the
phase boundary, all solution approaches arrive, as expected [23], at
the trivial solution (z ¼ w). Right above the phase boundary, small
positive tpd values are observed, corresponding to a narrow
“shadow region” [23] in the vicinity of the critical point. When the
saturation pressure is reached, negative values of tpd are observed
by further reducing the pressure. On the dew point branch
(T ¼ 780 K), the reference phase (z) and the trial phase (w) are
assumed to be vapor and liquid, respectively. At pressures well
above the dew-point pressure, all algorithms arrive at the trivial
solution (with tpd ¼ 0) before reaching the so-called “shadow re-
gion” where positive values of tpd are found right above the satu-
ration pressure. Again, as the pressure is reduced further, tpd
transitions to negative values just below the saturation pressure.
The results presented above demonstrate that the proposed
algorithms for stability testing accurately identify the phase
boundaries over the full range of relevant temperatures. For most
calculations (with the exception of the vicinity of the critical point)
a modest number of iterations is required to converge to the sta-
tionary points. A comparison of the convergence behavior of the
different algorithms is provided below.
60
50
40
30
780 800 820 840
400 500 600 700 800 900
Temperature (K)
3.4. Convergence behavior
The calculation examples provided above illustrates the accu-
racy and robustness of the suggested algorithms for stability anal-
ysis including effects of capillary pressure. All calculations to locate
the stationary points of Eq. (7) were performed four times: Once
using direct substitution, once using accelerated substitution
(extrapolation based on the largest eigenvalue), once using the
minimization approach with separate Newton update of the trial-
phase pressure (Eqs. (9), (17) and (19)) and once using the mini-
mization approach including the trial-phase pressure as an inde-
pendent variable (Eqs. (21)e(23)). All algorithms identify the same
values of TPD at the stationary points. The main difference between
these approaches is the rate of convergence. The left tab of Fig. 3
provides a comparison of the convergence behavior of the four
algorithms at T ¼ 383.15 K and 83 atm (just below the bubble point
branch) and an average pore radius of 10 nm.
As can be seen, all algorithms perform well and converge after a
moderate number of iterations. The right tab of Fig. 3 provides a
comparison of the convergence rates at T ¼ 700 K (in the vicinity of
the critical point). Near the critical point, the minimization ap-
proaches require a considerably lower number of iterations to
converge than both the direct substitution approach and the
accelerated substitution approach. The accelerated substitution, as
tested here with a single eigenvalue, is not sufficient to promote the
convergence of the direct substitution significantly (since the
eigenvalue approaches unity near the critical point). At these con-
ditions, it is preferable to use a minimization approach (also
pointed out by Michelsen [20]). In practical applications, one may
start with direct substitution and transition to a minimization
approach if the convergence is progressing slowly.
As discussed above, the update of the capillary pressure (and
trial-phase pressure) is performed separately from the update of
trial-phase compositions in the first three algorithms. Near the
critical point, the algorithms proposed here will behave (become)
identical to the original work of Michelsen [20] and preserve the
favorable convergence behavior of the minimization approaches.
When the capillary pressure (and trial-phase pressure) is updated
outside the main iterative scheme, one would expect the conver-
gence of the trial-phase pressure (and capillary pressure) to be
 M. Sherafati, K. Jessen / Fluid Phase Equilibria 433 (2017) 56e66 61

T=383.15K
0.04

0.02

tpd
0

-0.02

-0.04
80 81 82 83 84 85 86 87 88 89 90
pressure(atm)
-4
x 10 T=700K

0
tpd

-2

-4

116 117 118 119 120 121 122 123 124 125 126
pressure(atm)
-3
x 10 T=780K
2

0
tpd

-2

-4

-6
80 81 82 83 84 85 86 87 88 89 90
pressure(atm)

Fig. 2. Reduced tangent plane distance vs. pressure at constant temperatures of T ¼ 383.15 K (top), T ¼ 700 K (middle) and T ¼ 780 K (bottom).

T=383.15 K T=700 K
5 5
Direct Substitution
Accelerated Direct Substitution
0 Minimization 1
0
Minimization 2
-5
-5

-10
log(||err||)

log(||err||)

-10
-15

-15
-20

-20
-25

-25 -30
0 5 10 15 0 10 20 30
iteration iteration

Fig. 3. Convergence behavior of direct substitution, accelerated substitution and minimization with lagged pw updates (Minimization 1) and minimization with the full Jacobian
(Minimization 2) at a) T ¼ 383.15 K, p ¼ 83 atm and b) T ¼ 700 K, p ¼ 119 atm for rc ¼ 10 nm.

slower than the convergence of the trial-phase compositions phase pressure). To investigate the convergence rate of the capil-
(particularly for the minimization approach with lagging trial- lary pressure in more detail, we define the error in the capillary
62 M. Sherafati, K. Jessen / Fluid Phase Equilibria 433 (2017) 56e66

T=383.15 K T=700 K
5 5
Direct Substitution
Accelerated Direct Substitution
0 0 Minimization 1
Minimization 2
-5 -5
log(p error)

log(p error)
-10 -10
c

c
-15 -15

-20 -20

-25 -25

-30 -30
0 5 10 15 0 10 20 30
iteration iteration

Fig. 4. Convergence of the capillary pressure for direct substitution, accelerated substitution, minimization with lagged pc updates (Minimization 1) and minimization with the full
Jacobian (Minimization 2) at T ¼ 383.15 K, p ¼ 83atm and T ¼ 700 K, p ¼ 119atm for rc ¼ 10 nm.

pressure (epc) relative to the converged solution (P*c) based on a
tolerance of 1.1e-15.
 
epc ¼ abs Pc  Pc* :
Fig. 4 depicts the convergence behavior as observed for direct
substitution, accelerated substitution and the two minimization
approaches.
Calculations at the two temperatures are representative of the
bubble point branch (T ¼ 383.15 K, p ¼ 83 atm) and the near-critical
region (T ¼ 700 K, p ¼ 119 atm) of the phase envelope, respectively.
A comparison of Figs. 3 and 4 demonstrates that the error in the
capillary pressure is reduced rapidly for the first three approaches
as well: For both calculation examples (383.15 K and 700 K) re-
ported in Fig. 4 the capillary pressure is stabilized (small error
compared to the converged solution) at rates that are similar to the
reduction of the errors in trial-phase compositions. As a result, a
separate update of the capillary pressure does not significantly
affect the convergence of the stability analysis calculations and the
proposed algorithms can be implemented via modest revision of
(25)
existing codes for stability testing.
A comparison of the convergence rates for the two minimization
approaches in Figs. 3 and 4 demonstrates the expected quadratic
convergence behavior for the algorithm using a full set of inde-
pendent variables, while a separate update (Newton) of the trial-
phase pressure results in a super linear convergence. Additional
computational work (a factor of two) is required for factorization of
the matrix J given in Eq. (23) (non-symmetric matrix) relative to the
factorization of the symmetric matrix H given in Eq. (19). In addi-
tion, working with a full set of independent variables requires the
evaluation of compositional and pressure derivatives of the capil-
lary pressure in Eq. (23). Accordingly, the algorithm with separate

140

120

100
Pressure (atm)

80
Pl , rc=10nm
Pg , rc=10nm
60
Pl , rc=5nm

40 Pg , rc=5nm
Pl , rc=3nm
20 Pg , rc=3nm

0
100 200 300 400 500 600 700 800 900
Temperature (K)

Fig. 5. Detection of phase boundaries by stability analysis at different capillary radii (rc  10 nm). Triangle ¼ critical point.
 M. Sherafati, K. Jessen / Fluid Phase Equilibria 433 (2017) 56e66 63

140
95
90 120
85
80 100
360 380 400

Pressure (atm)
80 Pl , rc =50nm
Pg , rc =50nm
60
Pl , rc =20nm
Pg , rc =20nm
40
Pl , rc =10nm

20 Pg , rc =10nm

0
100 200 300 400 500 600 700 800 900
Temperature (K)

Fig. 6. Detection of phase boundaries via stability analysis at different capillary radii (rc > 10 nm). Triangle ¼ critical point.

update of the trial-phase pressure (minimization 1) provides for than 10 nm and one for pore radii larger than 10 nm. Fig. 5 reports
the more efficient approach for stability testing when capillary the phase envelopes along with phase boundaries detected by
pressure is included. stability analysis at T ¼ 383.15 K for a capillary radius of 3, 5 and
10 nm.
As the pore radius decreases, the depression of saturation
3.5. Effect of pore radius
pressures becomes more significant. However, stability testing is
invariably able to detect the phase boundary without any conver-
In the calculation examples presented above, the capillary
gence problems. Although we report results for calculation with
pressure calculations, based on Eq. (3), were performed for an
pore radii down to 3 nm, continuum-scale considerations may not
average radius. In practice, however, a wide range of pore radii is
be suitable at this scale and more detailed analysis using molecular
observed in tight formations. The porosity of the Middle Bakken
dynamics simulation may be more appropriate [26e29]. Next, we
region is approximately 6% and the corresponding oil effective
repeat the calculations at T ¼ 383.15 K for capillary radii of 10, 20
permeability is in the range of 30e100 nanodarcies. From the
and 50 nm as reported in Fig. 6. As expected, when the average pore
Kozeny-Carman equation [24,25] Nojabaei et al. [8] concluded that
radius is increased, the phase envelope including capillary pressure
the corresponding pore-throat radius should be in the range of
approaches that of the bulk fluid. The differences between the two
10e40 nm. In the calculation examples provided above, we used a
curves are already minimal for a capillary radius of 50 nm.
constant pore radius of 10 nm. Here, we examine the effect of pore
Fig. 7 reports the calculated saturation pressures (detected via
radius on the calculated phase boundaries to further test the pro-
stability testing along a constant temperature line in p-T space) for
posed algorithms for stability testing by repeating the example
varying capillary radii at a constant temperature of T ¼ 383.15 K.
calculations for a range of pore radii of 3e50 nm. We present the
The saturation pressure of the bulk fluid at the given temperature is
calculation results in 2 separate figures: one for pore radii smaller

100

90

80
Psat (atm)

70

60

50

40
5 10 15 20 25 30 35 40 45 50
rc (nm)

Fig. 7. Calculated saturation pressure vs capillary radius at T ¼ 383.15 K.

64 M. Sherafati, K. Jessen / Fluid Phase Equilibria 433 (2017) 56e66

150

normal envelope
critical point
PLiquid
100
PGas
Pressure (atm)
unstable feed

50

0
100 150 200 250 300 350 400
Temperature (K)

Fig. 8. Detection of phase boundaries via stability analysis for a condensate system (rc ¼ 10 nm).

99 atm. We observe that the saturation pressure is reduced to less 3.6. Effect of composition
than half (46 atm) for a capillary radius of 3 nm. Changes in the
saturation pressure with respect to capillary radius is non-linear Next, we consider a condensate reservoir with the fluid
and increases exponentially with decreasing capillary radius. description and feed composition given in Table 1, assuming an

-4
x 10 T=250K

0
tpd

-2

-4

110 111 112 113 114 115 116 117 118 119 120
pressure(atm)
-3
x 10 T=300K
4

2
tpd

-2

-4
145 146 147 148 149 150 151 152 153 154 155
pressure(atm)
T=365K
0.02

0.01
tpd

-0.01

-0.02
75 76 77 78 79 80 81 82 83 84 85
pressure(atm)

Fig. 9. Reduced tangent plane distance vs. pressure at constant temperature for T ¼ 250 K (top), T ¼ 300 K (middle) and T ¼ 365 K (bottom) for condensate fluid.
 M. Sherafati, K. Jessen / Fluid Phase Equilibria 433 (2017) 56e66 65

average pore radius of 10 nm. Fig. 8 depicts the phase envelope and on saturation pressures and related production operations.
the phase boundaries as identified by stability testing at 4 different
temperatures. In comparison to the heavier reservoir fluid consid- Nomenclature
ered above (oil), the changes in saturation pressures along the
bubble point branch are smaller for the condensate fluid descrip- b Step size in Hebden's algorithm
tion. This is mainly because the interfacial tension is smaller due to q Contact angle for wetting phase
smaller density difference between the two equilibrium phases. As r Molar density, mole/m3
a result, smaller values of capillary pressure are observed that, in bi
4 Fugacity coefficient
turn, result in smaller changes in the saturation pressure relative to s Interfacial tension, dyne/cm
the bulk fluid. This is in agreement with previous results reported mi Chemical potential
for multicomponent systems based on the multicomponent Kelvin ci Parachor coefficient
equation [6]. The critical point for this fluid composition is at E General exponent for interfacial tension model
T ¼ 247.63 K and P ¼ 111.94 atm. The 4 different temperatures along F Alternative minimization formulation
the phase envelope represent the bubble point branch, the near- gi Gradient of tm function
critical region and the dew point branch including the criconden- Hij Hessian of tm function
therm. In all cases, the stability analysis detects the phase bound- Jij Jacobian of F, as described by Eq. (23)
aries accurately. Ki Equilibrium constant
Fig. 9 explores the reduced TPD at the stationary points as a n Number of moles
function of pressure at constant temperatures of T ¼ 250 K, nc Number of components
T ¼ 300 K and T ¼ 365 K for the condensate system. Similar to Fig. 2, Pc;i Critical pressure, atm
we observe, near the critical point, that the stability analysis finds Pc Capillary pressure, atm
the trivial solution until the pressure is close, but still greater than px Liquid phase pressure, atm
the saturation pressure. Very close to the phase boundary Py Vapor phase pressure, atm
(0.005 atm above psat), a positive value of tpd (1.25e-8) is found. Pw Trial phase pressure, atm
This is expected as the shadow region is very narrow near the Pz Reference pressure, atm
critical point. Finally, as we cross the phase boundary, we start rc Pore radius, nm
observing increasingly negative values of the tpd. For the temper- R Radius of curvature of interface, nm
atures on the dew point branch (away from the critical point), we s Transformed variable as defined by Eq. (21)
observe a change of sign in the reduced tangent plane distance T Temperature, K
more clearly as we cross the phase boundary. Tc;i Critical temperature, K
The proposed algorithms do not allow for negative values of tm Reduced modified tangent plane distance
vapor and liquid phase pressures. As a result, special care must be tpd Reduced tangent plane distance
taken in the lower portion of the dew point branch. Sandoval et al. TPD Tangent Plane Distance as defined by Eq. (5)
[2] suggested an algorithm for phase envelope calculations allow- v Transformed variable as defined by Eq. (17)
ing for negative phase pressures, which can be incorporated in the Wi Mole number of component i in the trial phase
algorithms for stability analysis proposed here. Negative phase Wt Total mole numbers of the trial phase
pressures are, however, unlikely to be encountered during wi Mole fraction of component i in the trial phase
compositional simulation at realistic reservoir conditions. xi Mole fraction of component i in the liquid phase
yi Mole fraction of component i in the gas phase
4. Summary zi Mole fraction of component i in reference phase

Accurate detection of phase stability is crucial in the modeling

and simulation of production from oil and gas reservoirs. Here we
propose a set of robust, efficient and accurate algorithms for sta-
bility testing, including capillary pressures, of relevance to
compositional simulation of tight oil/gas resources. The efficiency
and accuracy of the proposed algorithms were demonstrated for a
range of relevant temperature and pressure conditions in oil and
condensate systems. The formulation and proposed solution stra-
tegies for stability testing provide a simple path to incorporate ef-
fects of capillary pressure in existing codes for stability testing.
For large values of the capillary pressure, the saturation pres-
sures deviate significantly from the bulk fluid calculations for both
oil and condensate systems. Along the bubble point branch of the
phase envelope, saturation pressures are suppressed while, along
the dew point branch, the calculated saturation pressures are
increased compared to the bulk behavior (for both oil and
condensate systems). In the vicinity of the critical point, stability
analysis calculations including capillary pressure approach calcu-
lations for bulk-fluids, as the interfacial tension approaches zero.
The analysis presented here assumes a single average pore size
to represent unconventional resources. In practice, calculations will
be more complicated as a wide range of pore sizes are observed in
reservoir rocks. However, a single average value of the pore radius
will allow engineers to gauge the potential impact of confinement
References
[1] B. Yan, M. Alfi, Y. Wang, J.E. Killough, A new approach for the simulation of
fluid flow in unconventional reservoirs through multiple permeability
modeling, in: SPE Annual Technical Conference and Exhibition, 2013. No. SPE-
116173-MS.
[2] D.R. Sandoval Lemus, W. Yan, M.L. Michelsen, E.H. Stenby, The phase envelope
of multicomponent mixtures in the presence of a capillary pressure difference,
Ind. Eng. Chem. Res. 55 (2016) 6530e6538.
[3] L. Du, L. Chu, Understanding anomalous phase behavior in unconventional oil
reservoirs, in: SPE Canadian Unconventional Resources Conference, 2012. No.
SPE-161830-MS.
[4] Y. Wang, B. Yan, J. Killough, Compositional modeling of tight oil using dynamic
nanopore properties, in: SPE Annual Technical Conference and Exhibition,
2013. No. SPE-166267-MS.
[5] A.A. Shapiro, E.H. Stenby, Kelvin equation for a non-ideal multicomponent
mixture, Fluid Phase Equilibria 134 (1) (1997) 87e101.
[6] A.A. Shapiro, K. Potsch, J.G. Kristensen, E.H. Stenby, Effect of low permeable
porous media on behavior of gas condensates, in: SPE European Petroleum
Conference, 2000. No. SPE-65182-MS.
[7] T. Firincioglu, E. Ozkan, C. Ozgen, Thermodynamics of multiphase flow in
unconventional liquids-rich reservoirs, in: SPE Annual Technical Conference
and Exhibition, 2012. No. SPE-159869-MS.
[8] B. Nojabaei, R.T. Johns, L. Chu, Effect of capillary pressure on fluid density and
phase behavior in tight rocks and shales, SPE Reserv. Eval. Eng. 16 (03) (2013)
281e289.
[9] B. Nojabaei, N. Siripatrachai, R.T. Johns, T. Ertekin, Effect of saturation
dependent capillary pressure on production in tight rocks and shales: a
compositionally-extended black oil formulation, in: SPE Eastern Regional
Meeting, 2014. No. SPE-171028-MS.
66 M. Sherafati, K. Jessen / Fluid Phase Equilibria 433 (2017) 56e66
[10] L. Wang, E. Parsa, Y. Gao, J.T. Ok, K. Neeves, X. Yin, E. Ozkan, Experimental
study and modeling of the effect of nanoconfinement on hydrocarbon phase
behavior in unconventional reservoirs, in: SPE Western North American and
Rocky Mountain Joint Meeting, 2014. No. SPE-169581-MS.
[11] N.S. Alharthy, T. Nguyen, H. Kazemi, T. Teklu, R. Graves, Multiphase compo-
sitional modeling in small-scale pores of unconventional shale reservoirs, in:
SPE Annual Technical Conference and Exhibition, 2013. No. SPE-166306-MS.
[12] A. Firoozabadi, Thermodynamics of Hydrocarbon Reservoirs, first ed.,
McGraw-Hill, 1999.
[13] D.B. Macleod, On a relation between surface tension and density, Trans.
Faraday Soc. 19 (1923) 38e41.
[14] S. Sugden, The variation of surface tension with temperature and some related
functions, J. Chem. Soc. Trans. 125 (1924) 32e41.
[15] C.F. Weinaug, D.L. Katz, Surface tensions of methane-propane mixtures, Ind.
Eng. Chem. 35 (2) (1943) 239e246.
[16] E.W. Hough, G.L. Stegemeier, Correlation of surface and interfacial tension of
light hydrocarbons in the critical region, Soc. Pet. Eng. J. 1 (04) (1961)
259e263.
[17] S.T. Lee, M.C.H. Chien, A new multicomponent surface tension correlation
based on scaling theory, in: SPE Enhanced Oil Recovery Symposium, 1984. No.
SPE-12643-MS.
[18] A.S. Danesh, A.Y. Dandekar, A.C. Todd, R.A. Sarkar, Modified scaling law and
parachor method approach for improved prediction of interfacial tension of
gas-condensate systems, in: SPE Annual Technical Conference and Exhibition,
1991. No. SPE-22710-MS.
[19] D.S. Schechter, B. Guo, Parachors based on modern physics and their uses in
IFT prediction of reservoir fluids, SPE Reserv. Eval. Eng. 1 (03) (1998)
207e217.
[20] M.L. Michelsen, The isothermal flash problem. Part I. Stability, Fluid Phase
Equilibria 9 (1) (1982) 1e19.
[21] H. Orbach, C.M. Crowe, Convergence promotion in the simulation of chemical
processes with recycle e the dominant eigenvalue method, Can. J. Chem. Eng.
49 (1971) 509e513.
[22] M.D. Hebden, An Algorithm for Minimization Using Exact Second Derivatives,
UKAEA Theoretical Physics Division Harwell, 1973.
[23] C.P. Rasmussen, K. Krejbjerg, M.L. Michelsen, K.E. Bjurstrom, Increasing the
computational speed of flash calculations with applications for compositional,
transient simulations, SPE Reserv. Eval. Eng. 9 (2006) 2e38.
[24] J. Kozeny, Über kapillare Leitung des Wassers im Boden: (Aufstieg, Versick-
erung und Anwendung auf die Bewa €sserung), Ho€ lder-Pichler-Tempsky, 1927.
[25] P.C. Carman, Fluid flow through granular beds, Chem. Eng. Res. Des. 75 (1997)
S32eS48.
[26] R. Evans, U.M.B. Marconi, P. Tarazona, Fluids in narrow pores: adsorption,
capillary condensation, and critical points, J. Chem. Phys. 84 (4) (1986)
2376e2399.
[27] F. Restagno, L. Bocquet, T. Biben, Metastability and nucleation in capillary
condensation, Phys. Rev. Lett. 84 (11) (2000) 2433.
[28] N. Giovambattista, P.J. Rossky, P.G. Debenedetti, Effect of pressure on the
phase behavior and structure of water confined between nanoscale hydro-
phobic and hydrophilic plates, Phys. Rev. E 73 (4) (2006) 041604.
[29] Z. Jin, A. Firoozabadi, Thermodynamic modeling of phase behavior in shale
media, SPE J. 21 (1) (2016) 190e207.