General Purpose Compositional Model
Gabor Acs, * Hungarian Hydrocarbon Inst.
Sandor Doleschall, Hungarian Hydrocarbon Inst.
Eva Farkas, Hungarian Hydrocarbon Inst.
Abstract
A direct sequential method has been developed to simulate
isothermal compositional systems. The solution technique
is the same as that of the implicit pressure, explicit satura-
tion (IMPES) method: one pressure is treated implicitly
and (instead of the phase saturation) the component
masses/moles are treated explicitly. A "volume balance"
equation is used to obtain the pressure equation. A
weighted sum of the conservation equations is used to
eliminate the nonlinear saturation/concentration terms
from the accumulation term of the pressure equation. The
partial mass/mole volumes are used as "constants" to
weight the mass/mole conservation equations.
The method handles uniformly a range of cases from
the simplified compositional (i.e., black-oil) models to the
most complicated multiphase compositional models of in-
compressible and compressible fluid systems. The
numerical solution is based on the integrated finite-
difference method that allows one- (1D), two- (2D), and
three-dimensional (3D) grids of regular or irregular
volume elements to be handled with the same ease. The
mathematical model makes it possible to develop modular
versatile computer realizations; thus the model is highly
suitable as a basis for general-purpose models.
Introduction
During the last three decades reservoir simulators have
been well developed. The enormous progress incomputer
techniques has strongly contributed to the development
of increasingly effective and sophisticated computer
models. The key numerical techniques of modeling con-
ventional displacement methods had been elaborated upon
by the beginning of the 1970's, 1,2 and it was possible to
develop a single simulation model capable of addressing
most reservoir problems encountered. Since the 1970's,
however, because of the sharp rise in oil prices, the need
for new enhanced recovery processes has forced reservoir-
simulation experts to develop newer computer models that
account for completely unknown effects of the new dis-
placement mechanisms. The proliferation of recovery
methods since the 1970's has resulted in a departure from
the single-model concept because individual models tend
to be developed to simulate each of the new recovery
schemes. This proliferation of models, however, seems
to be a less than ideal situation because of the expense
involved in the development, maintenance, and applica-
tions training for the multiple new models. In addition,
when different models are applied to simulate various en-
'Deceased.
Copyright 1985 Society of Petroleum Engineers
AUGUST 1985
hanced recovery methods, no common basis exists to help
survey, compare, and thus understand the different
recovery mechanisms.
The importance of a single, general simulator capable
of modeling all or most recovery processes of interest was
emphasized by Coats,3 who worked out a model 4 as a
step in this direction. Economic restrictions have also
forced various companies to develop mUltiple-application
reservoir models. The multiple-application reservoir
simulator (MARS) program presented by Kendall et al. 5
is one realization of the goal: a single program for multi-
ple application.
From a mathematical point of view, reservoir simulators
consist of a set of partial differential equations and a set
of algebraic equations, both with the appropriate initial
and boundary conditions. In isothermal cases the partial
differential equations, taking into account Darcy's law,
describe the mass/mole/normal-volume conservation for
each component of the reservoir fluid system. Phase
and/or component transport caused by capillarity, gravity,
and/or diffusion also can be taken into account. The
algebraic equations describe the thermodynamic proper-
ties of the reservoir fluid/rock system. The existence of
local and instant thermodynamic equilibria is a generally
accepted assumption of reservoir simulation. This means
that the number of mass/mole/normal-volume conserva-
tion equations is equal to the number of components used
to describe the reservoir fluid/rock system. During the
simulation the reservoir examined is divided into volume
elements by aiD, 2D, or 3D grid. Each of the volume
elements is characterized by the appropriate reservoir
properties and the displacement process is described by
a series of thermodynamic equilibria for each volume
element.
The difference between the simulators of conventional
and enhanced recovery methods essentially arises from
how many components are chosen as a means of appro-
priately describing the displacement process, and how the
thermodynamic equilibria (thermodynamic properties) of
the reservoir fluid/rock system are characterized. In cases
of conventional technologies a simplified (black-oil) ap-
proach of the hydrocarbon system by a pseudogas and a
pseudo-oil component generally is accepted, and the ther-
modynamic properties of the given system depend only
on the pressure. This approximation made it possible to
develop the direct sequential IMPES solution technique,
taking into account the advantage of black-oil models
wherein the number of components is equal to the number
of phases and thus the number of phases is equal to the
number of conservation equations. In such cases, by
assuming explicit transmissibility and capillarity ca1cula-
543
tions and by combining the conservation equations, the
saturation terms can be eliminated from the accumulation
terms and a direct equation can be derived for the
pressure. As is well known, this method has greatly con-
tributed to the development of general purpose black-oil
models that have been used successfully for many
problems.
EVen though compositional models of enhanced
recovery technologies generally use the IMPES
technique 6- 11 (i.e., one pressure is calculated implicitly,
and saturations and concentrations are calculated explicit-
ly), no direct method has been derived for compressible
systems. IMPES-type compositional methods apply the
ideas of deriving an equation for the pressure from the
conservation equations, but this pressure equation can-
not be solved directly, which means that pressure equa-
tions must be formed and solved over a timestep for every
iteration step required for convergence. In addition, con-
vergence problems also may occur for some cases. 12 The
necessity of the iteration technique is caused by the
nonlinear accumulation terms. The various methods apply
some linearization technique, but they are not able to
eliminate correctly the saturation/concentration terms
from the accumulation term of the pressure equation.
We present here a direct method for 1M PES-type com-
positional modeling. The pressure equation is derived
from an in-situ volume conservation equation. A weighted
sum of the material balance equations is applied to
eliminate the nonlinear saturation/concentration terms
from the accumulation term of the pressure equation. the
partial volumes are used as the "constants" to weight the
material balance equations. The pressure equation can be
solved directly. Using black-oil fluid characterization, the
method reduces to the conventional IMPES method.
Undersaturated systems are treated in the same way as
saturated ones.
The main purpose of our work is to provide a uniform
format of the wide variety of isothermal simulation equa-
tions, including the possibility of nonisothermal model-
ing. The fluid property calculations work as a completely
separate part of the model, which means that the model
is highly suitable for simulating various enhanced rpcovery
schemes. The numerical solution is based on the integrated
finite-difference form 13 that allows 10,20, and 30 grid~
of regular or irregular volume elements to be handled with
the same ease. The method enables a model of a high-
level modular structure to be developed; thus it is suitable
as a basis for general purpose simulators.
Thermodynamic Properties and Primary Variables
The local thermodynamic equilibrium is a generally ac-
cepted assumption in reservoir simulation. This allows
one to apply equilibrium thermodynamic concepts when
describing the nonequilibrium displacement processes.
The continuum models can be visualized as a large number
of differentially small subsystems. All the thermodynamic
properties are defined for each subsystem at each instant
of time, and each subsystem is treated as a system in ther-
modynamic equilibrium.
According to Gibbs' phase rule, the number of degrees
of freedom, F, of a system in thermodynamic equilibri-
um is given by F=N+2-N p, where N is the number of
components and N p the number of phases; thus,
S44
N +2 - N p independent intensive parameters determine all
the intensive properties of the fluid phases. In reservoir
simulation, the phase saturations also must be known, so
N p -1 additional parameters are needed. Consequently,
the number of independent parameters will be N + 1.
Because we are dealing with isothermal systems, this
number is reduced to N. The N independent intensive pa-
rameters give no information about the system as a whole.
To determine the total fluid volume (i.e., the extensive
properties), an additional parameter is needed and at least
one of the N + 1 independent parameters should be an ex-
tensive one. Summarizing, in isothermal reservoir simula-
tion the number of fluid properties that uniquely determine
all the others is N in the intensive case and N + 1 in the
extensive case.
In the following, the independent parameters that deter-
mine uniquely all the others are primary parameters. If
the number of thermodynamic parameters to be deter-
mined is NT, then NT=N[+N D , where NI is the number
of primary parameters. If we know the primary variables,
the N D additional parameters can be determined by ap-
plying N D algebraic equations describing the given
system. To determine the space- and time-dependence of
the primary parameters we need N[ conservation (i.e.,
differential equations). In compositional modeling NTis
generally much greater than N I. Besides the linearization
technique, the fundamental questions of numerical tech-
niques are (1) which parameters should be the primary
variables determined from the conservation equations and
(2) which parameter can best be determined from which
equation.
In IMPES black-oil models (where N=Np), one
pressure and N p - 1 saturations are accepted as the set
of primary variables of the governing differential equa-
tions. The various compositional models apply different
sets of primary variables, however, and one or more
saturations can very often be found among the primary
ones. It seems that the traditional importance of satura-
tion in reservoir engineering and the effective numerical
solutions of black-oil models using saturations as primary
variables have greatly influenced numerical solution tech-
niques when turning from black-oil modeling to composi-
tional modeling. Nevertheless, it is worth mentioning that
saturations that can be regarded as ratios of phase and total
volumes are not usual thermodynamic properties.
A common feature of the models presented previously
is that all use an iterative solution technique, and generally
handle the components asymmetrically. The latter con-
dition always raises the practical question of which com-
ponent(s) should be calculated differently from the others?
Another disadvantage of the asymmetric methods is that
they are not really convenient for a general-purpose
model. As for the iteration techniques, a frequent prob-
lem is how the numerical errors (such as overshooting)
can be eliminated during the iteration.
When selecting the primary variables of the proposed
model, based on the original thermodynamic concepts of
fluid transport equations, we endeavored to satisfy cer-
tain conditions.
1. The model should be applicable to black-oil and com-
positional simulation, thereby enabling it (a) to be used
more generally, (b) to be checked by comparison with
other black-oil models, and (c) to be used economically
to simulate the different periods of oil recovery (e.g.,
SOCIETY OF PETROLEUM ENGINEERS JOURNAL
primary and secondary displacement by a black-oil system
and enhanced recovery by a compositional system).
2. The form of differential equations should be sym-
metrical for all the components. This feature is advan-
tageous in every respect: numerical analysis, discret-
ization, and program code development.
As a consequence, we selected the pressure p of the
phase equilibrium and the component masses m j, i = 1 ...
N. These N + 1 parameters uniquely determine all the in-
tensive and extensive properties of arbitrary isothermal
systems by application of the appropriate set of algebraic
equations. (A method similar to the one presented here
can be developed if one chooses mole numbers instead
of the masses, but masses can be regarded as middle pa-
rameters between the normal volumes of black-oil and
mole fractions of compositional models.)
Some Characteristics of the Model Variables
The primary variables of the model are the pressure p and
the component masses m j, i = 1 ... N, or (using vector
notation) m. If the primary variables are known, then ap-
plying algebraic equations such as empirical fluid prop-
erty correlations or an equation of state, the equilibrium
and PVT properties of the system can be determined
uniquely. This second group of parameters,
Cji(p,m) = mass concentrations,
Pj(p,m) = phase densities,
V/p,m) = phase volumes,
Vf(p,m) = ~ Vj(p,m) = total fluid volume, and
j
Sj(p,m) = Vj (p,m)IVjP,m) = saturations,
(i = 1 ... N; j = 1 ... N p), are secondary parameters. The
partial mass volumes of the fluid system form a group
of thermodynamic properties that we range among the
secondary parameters. These variables have not been used
previously in reservoir simulation. The definition of par-
tial mass volumes for a multi phase system is
- - -+ oVf(p,m) .
Vi=V;(p,m)= , l=l. .. N .......... (1)
omi
Transport property functions of secondary parameters
are also regarded as functions of primary parameters
(p,m), such as
kj(p,m) = relative permeabilities,
11- j(p,m) = viscosities, and
Dji(p,m) = diffusion coefficients.
The capillary pressures are interpreted in a more general
way than usual using the formula
P c(p,m)=p/p,m)-p.
We now discuss some characteristics of the secondary
parameters. Intensive-type parameters are zero-order
homogeneous functions of the extensive ones. Thus, for
any c>O,
Cji(p,cm) = Cji(p,m),
Pj(p,cm) = Pj(p,m),
S/p,cm) = Sj(p,m).
AUGUST 1985
The partial mass volumes are also intensive properties
because they are the ratios of two extensive properties;
thus
Vi(p,cm) = Vi(p,m). . ...................... (2)
The total fluid volume, as an extensive parameter, is a
first-order homogeneous function of the extensive ones;
this means that
Vf(p,cm) =cVf(p,m). . .................... (3)
The extensive masses mi cannot be unknown functions
of differential equations. This is why the intensive overall
mass densities ri are introduced:
mi
ri= ............................. (4)
Vf(p,m)
or
ri= ~PjSjCji . ........................... (5)
j
The N+ 1 parameters p, rj .. . rN [or (p,7)] are,
however, not independent since the number of indepen-
dent intensive parameters is only N in an isothermal
system. Let Vp be an arbitrary volume and c= Vpl
Vf(p,m). Then, from Eq. 4, it follows that
and from Eq. 3
Vf(p,vp 7)=vp ' .......................... (6)
The N+ 1 independent parameters (p,m) can be used to
describe all the fluid properties. An equivalent intensive-
type fluid description can be realized by using the N + 1
parameters (p,7); however, these parameters are not in-
dependent and must satisfy Eq. 6.
Governing Differential Equations
Combining Darcy's law with the component mass con-
servation equations and taking into account diffusional
mass transport, a usual form of compositional models is
the system of differential equations
T
o(cf>r .) [ k .P .
=div K t ~ ~,.J Cji(grad Pj -Pjg grad h)
J t"'J
+ ~pjSjDji grad Cji] +qj, i=I .. .N . ... (7)
J
Let us introduce the notation -~ for the expression in
brackets on the right side of Eq. 7. This equation then
can be written in the concise form:
O(cf>ri) .-
----a;- = -dlV J i +qi' i= 1 . .. N. ............ (8)
545
Let us choose the pressure p(x,t) and overall densities
r(x,t) as unknown functions of the system.
The N mass conservation equations are, however, in-
sufficient for determining the N + 1 unknown variables
(p, r); therefore, an additional equation is needed. The
N+ 1 unknown variables (p, r) do not represent an in-
dependent thermodynamic parameter set. They must
satisfy Eq. 6 at every point of time for a given volume
Vp. Let Vr be a rock volume in which we assume local
equilibrium during the simulation, and
Vp=Vre/> . ................................ (9)
Then, in accordance with Eq. 6,
Vf(p,vre/>r)=Vp ......................... (10)
must be valid at every moment; thus dVtfdt=dVpldt:
de/> ap
=Vr dp a; ......................... (11)
By using the definition of the partial mass volumes (Eq.
1), we can write Eq. 11 in the form
aVf ap + Vr ~
L.J Vi. a(e/>r;) = Vr de/>ap . ....... (12)
ap at i at dp at
Combining Eqs. 9 and 10,
Vf
Vr=-;- ................................ (13)
can be written. Let Cf denote the total fluid compressibili-
ty and Cr the rock compressibility; then,
_ 1 iWf
cf= - - - - ........................... (14)
Vf ap
and
de/>
cr = - . ................................ (15)
dp
The derivatives a(e/>ri)lat can be eliminated from Eq. 12
by using Eq. 8. By applying Eq. 13 and definitions from
Eqs. 14 and 15, the final form of the N+ 1 differential
equation will be
Thus, our N+ 1 governing differential equations are Eqs.
8 and 16. The N + 1 unknown functions of the system are
546
the pressure p(x,t) and the overall densities r(x,t). All
the coefficients and variables occurring in the governing
differential equations are considered to be functions of
(p,r).
The model equations were originally developed in the
form of difference equations. The corresponding system
of differential equations given previously was worked out
later as a concise and more general mathematical represen-
tation of the model.
The Difference Equations of the Model
This section deals with the difference equations of the
model. In the Appendix we prove that the difference equa-
tions presented here can be regarded as one of the dis-
cretized forms of the governing differential equations.
The primary variables of the model are the pressure p
determining the phase equilibrium and the component
masses m i, i = 1 ... N. The time- and space-dependence
of these N + 1 independent thermodynamic variables are
determined from the N mass balance equations and a
volume balance equation.
The set of variables (p, r) also can be chosen as
unknown functions of the difference equations because
the discretized form of Eqs. 8 and 16 can uniquely deter-
mine this set of variables as well. Nevertheless, because
of the numerical errors of the solution technique, the
calculated values of (p, r) probably would be an incon-
sistent set of thermodynamic properties that would not
satisfy Eq. 6. To avoid this difficulty, it would be possi-
ble to determine only N - 1 values r i from N - 1 mass bal-
ance equations; however, in this case the symmetrical
nature of our model would be lost. It seems to be an ad-
vantage of the proposed primary variables that their ap-
plication enables the inconsistency of thermodynamic
properties that causes one of the most inconvenient prob-
lems of compositional models to be avoided.
When constructing the difference equations we started
from the integral instead of the differential form of the
mass conservation equations. The reservoir is divided into
.volume elements, but neither their shape nor the number
of their neighboring elements is given in advance. All the
volume elements are treated similarly; thus it suffices to
show how the difference equations for one volume ele-
ment are written. To simplify, the data of this element
are denoted by subscript 0; those of its neighboring
elements by k, k= 1 .. . M, where M denotes the number
of neighboring elements. The boundary conditions are not
discussed. Even so, if we wish to model a closed reser-
voir, the equations discussed here can be applied directly
because this condition arises only in the number of
neighboring elements.
The reservoir contained in the volume element is char-
acterized by (1) the PV VpO (which may be pressure
dependent), (2) the AkKk (k= 1 .. . M) products of bound-
ary surface areas and absolute permeabilities (defined in
the Appendix), and (3) the products Ake/>k (k=I ... M),
where e/> k is an average porosity referring to volumes O-th
and k-th. The fluid system is characterized by the primary
variables Po, m 0, and the secondary and transport
variables.
In accordance with the definition of the primary
variables, the entire system is uniquely defined if the func-
tions p(t) , m(t) are given for all volume elements. The
SOCIETY OF PETROLEUM ENGINEERS JOURNAL
equations to be described hereafter show how an in- the brackets by L iO, then !::.m iO can be written in the
dividual timestep can be calculated. That is, with the following concise form:
knowledge of all
M
pO and fit 0 !::.miO=!::.t ~ Lik(!::.Pk-!::.PO)
k=l
belonging to the old time-level to, we are able to see how
the changes +!::.tL iO , i=l. .. N. ................. (18)

We now have N difference equations relating to the N + 1

can be determined for the period !::.t=tn -to.
In the equations (e.g., when writing Darcy's law), the
x
distances of the grid points 0, X k of the neighboring
volume elements are needed. These distances can be writ-
ten as
x
where the depth belonging to grid point is given by h.
The mass change of the i-th component of the reser-
voir fluid existing in the volume element can be calculated
from the masses flowing across the boundary surfaces of
the volume element and from the flow rates of sources
or sinks. When calculating the mass change !::.m iO, the
pressure p determining the phase equilibrium is taken into
account implicitly. The effects of gravitational and
capillary forces, the diffusional mass transport, and the
composition of the phases flowing between neighboring
elements are taken into account explicitly.
Thus,
independent primary parameters. The N+ I-th equation
we use expresses that the fluid volume and the PV of the
volume element are equal during the process. To get this
last equation, used as the pressure equation, some assump-
tions must be made.
It is assumed that the volume change of the fluid system
can be regarded as directly proportional to the mass
change for small mass changes (i.e., for mass changes
of a timestep); thus
and similarly,
VtfpS,mS + !::.tCk (!::.Pk -!::.po)]- VIPS ,fit S)
!::.t(!::.Pk -Apo)
k= 1 .. . M, ........................... (20)
;-+
where the mass rate vectors L 0 = (L 1 0 ... L NO) and
C k =(L lk . . . LNd are defined by Eqs. 17 and 18. The
dimensions of Lo and Lk are, respectively, (mass/unit
time) and (mass/unit time/unit pressure difference).
In the same way, the effects of pressure change of a
timestep modifying the fluid volume

VAPo + !::.po ,fit g) - VAPo ,fit g)

Cf ......... (21)
!::.Po

and the PV

. ~ PjkOSjkODj;kO(Cfik - Cj;o) +QiO J, ........ (17)

1 ................ (22)

where PjkO, SjkO, and Dj;kO are average values referring

to the k-th and O-th volume elements (Djik is defined in are assumed to be linear functions. The proportionality
the Appendix): constants of Eqs. 21 and 22 can be regarded as numerical
approximations of fluid and rock compressibilities, de-
fined by Eqs. 14 and 15: cf= -cfVf ; c p =crVr .
kO 0 Using approximations of Eqs. 19 through 22, the equali-
jkPjk Cojik 1·f Pjk
0 (h h) 0
~PjO -PjkOg k - 0 > ,
0 0
- - 0-
ty of the pore and fluid volumes at the new time-level t n
/-tjk can be written
ejiku -- kQopQo
1 l co h .
[ - - 0- jO ot erWlse.
/-tjO

Here, the explicit upstream estimation of the transmissi- M

bilities is chosen. If in Eq. 17 the expressions in the first +!::.t ~ Ck(!::.Pk -!::.Po)+!::.tco . ............. (23)
parentheses are denoted by Vik> k= 1 .. . M, and those in k=l

AUGUST 1985 547

The PV Vp(pg) and fluid volume Vf(pg,mo) of the old
time-level to are theoretically equal; however, this equali-
ty is never satisfied in simulation practice. The difference
of these volumes (called volumetric error) is used to cor-
rect the pressure equation.
Eqs. 18 and 23 can be written in the same way for each
volume element. The pressure equation (Eq. 23) written
for all the volume elements gives a linear algebraic system
of equations referring to the unknown set of I1p. This
system of equations is similar to the usual ones; moreover,
in many special cases (applying the conventional volume
element system) it agrees with the systems of finite dif-
ference equations generally used. After the set of I1p has
been determined, the application of Eq. 18 allows 11m
to be calculated separately for each element. As pn and
mn uniquely determine all the fluid properties, the
calculation of the given timestep is completed by apply-
ing some equation of state and/or empirical fluid proper-
ty correlations.
Coefficients of the Pressure Equations
Eqs. 19 through 22 can be considered as first-order Taylor
approximations of functions Vf and V p:
Vf(pg,m 0+l1tCo) = Vt<pg,m 0)
+l1t ~
N (av)O
-.!.. Li,o + ... , ............... (24)
i=1 ami
+l1t(I1Pk-l1pO) ~N (av)o
-.!.. L ik + ... , ..... (25)
i=1 ami
+ (a:;:) °I1po + ... , ..................... (26)
If we compare Eqs. 19 and 24 we can see that
Co= ±
i=1
(avf)o'Li,o=
ami i=1
±
V?L iO ' ....... (28)
From Eqs. 20 and 25,
where the dimensions of volume rates Co and Ck are,
respectively, (volume/unit time) and (volume/unit
548
time/unit pressure difference). Comparison of Eqs. 21 and
26 yields
Cf= (a:;:) 0, ........................... (30)
and comparing Eq. 22 with Eq. 27 yields
Cp = (d~ )0 • ........................... (31)
Eqs. 28 and 29 show that the coefficients Co and Cb
k = 1 ... M can be written as scalar products of the partial
volume vector
(V}, V 2... V:z,), ......................... (32)
--
and mass rate vectors L 0, L k> k = 1. .. M.
The last decade has seen great progress in computer
methods of chemical engineering. Many new equations
of state have been published, and two of them 14,15 are
used extensively in reservoir simulation. The partial
volume vector as shown in Eq. 32 can be calculated direct-
ly from an equation of state or from a set of empirical
fluid property correlations in an analytical form, and then
Co, Cb k = 1 ... M can be derived by simple scalar prod-
ucts. The partial volumes also can be derived numerical-
ly instead of analytically. In that case, however, if
M+ 1 <N, then application of the directional derivatives
defined by Eqs. 19 and 20 is more economical. To derive
Co, Cb k=I ... M numerically, the formula
n=O, 1 .. . M, ........................ (33)
can be used, where a is a small dimensional value; the
dimension of ao is unit time, and that of ak is unit
time X unit pressure difference.
If there is no volume change at constant pressure mix-
ing, the simple equations
and
............................ (34)
are valid. Eq. 34 can be used to calculate coefficients Co
and Ck in cases of black-oil and micellar modeling. These
equations also can be used to improve program efficiency.
It is obvious that these equations are valid if the com-
positions of the inflowing and outflowing phases are the
same as those of the phases inside the volume element.
If diffusion is neglected, the outflowing compositions
always meet these conditions. On the other hand, signifi-
SOCIETY OF PETROLEUM ENGINEERS JOURNAL
cant differences in composition requiring Eqs. 19 and 20
or Eqs. 28 and 29 are to be found only in the vicinity of
fronts.
Volumetric Error as a Stability Term;
Timestep Control
There is another new feature of the pressure equation (Eq.
23). As mentioned earlier, the volumetric errors are cor-
rected by the terms
in each timestep. These terms correct the truncation errors
committed by the linear approximations of Eqs. 24
through 27 and allow a greater timestep, while keeping
the maximum volumetric error at the same level.
However, these terms are used only for compressible
systems.
Another practical advantage of the terms Ll Va is that
they can be applied to determine uniformly the timesteps
when modeling different compressible systems. The rela-
tionship
maximum [abs [ Ll VA
Vp(pO)
J] < 0.0005
has proved to be useful when determining a new timestep.
Another restriction sometimes results from the explicit
upstream transmissibility estimation that causes negative
masses to be calculated. Because this is an impossibility,
the timestep must be recalculated with a smaller Llt. To
avoid negative masses, an estimation is made taking into
account the explicitly calculated mass rates 1:0 , When
determining the new timestep, the relationship
0.75miO + LltL iO >0
must be satisfied (in addition to the volumetric error con-
dition) for each component in every volume element.
Simulation of Appearance or
Disappearance of Phases
The saturations are secondary variables of the proposed
model. This means that first the set of primary variables
(p,iit) is determined from the system of difference equa-
tions (Eqs. 18 and 23) and then the other thermodynamic
properties (such as saturation) are determined from a set
of algebraic equations with the knowledge of (p,iit). (The
saturations Sj' j = 1 ... N p are calculated as ratios of fluid
phase volumes Vj and the total fluid volume Vf .) Con-
sequently, calculation of the appearance or disappearance
of phases causes no difficulties. Nevertheless, when the
number of phases changes in one or more volume elements
during a timestep, the first-order Taylor approximations
in Eqs. 24 through 26 are less exact than otherwise.
Therefore, the volumetric errors of the appropriate ele-
ment(s) are greater than usual. By applying the stability
term Ll V and an automatic timestep reduction, the method
controls itself to overstep this changing.
Comparison of the Method With the
IMPES Black-Oil Method
The pressure equation of the IMPES black-oil model is
derived by a combination of the normal-volume conser-
AUGUST 1985
vation equations. The "constants" that are applied to
weight the conservation equations are practically partial
volumes, while the saturation constraint serves as the in-
situ volume conservation equation to complete the normal-
volume conservation equations. (The partial volumes and
total fluid compressibility of a black-oil system can easi-
ly be given. 16)
Main Steps of the Numerical Solution
The calculation of a timestep consists of the following.
1. Determination of the mass rate vectors 1:0 , Lb
k= 1 .. . M for each volume element: Eqs. 17 and 18.
2. Determination of the coefficients c of the pressure
equations: Eqs. 21 and 22 and Eqs. 19 and 20 or Eqs.
28 and 29.
3. Solution of the system of linear pressure equations:
Eq.23.
4. Calculation of the component mass changes: Eq. 18.
5. Calculation of all the fluid properties from the new
set of primary parameters pn ,iit n.
Possibility for Improving the Method
Explicit first-order transmissibilities and other explicit ap-
proximations are used (similar to the 1M PES method),
so the proposed numerical solution technique suffers from
stability limitations. Because the method gives the solu-
tion in the same fashion as the IMPES, second-order ex-
plicit transmissibilities or other techniques applied to
improve the 1M PES method can be used as well.
A sequential implicit compositional formulation was
presented by Watts, 16 who has combined and extended
the ideas of the proposed method with that of Spillette
et al. 17 Watts' procedure offers improved stability over
IMPES-type methods while not being as expensive com-
putationally as fully implicit methods.
A Possible Thermal Extension of the Model
In isothermal cases, the thermodynamic properties depend
only on the pressure and composition, but one can take
into account the effect of temperature as well. In this case,
the total fluid volume can be written as V/p,T,iit), and
the volume conservation equation equivalent to Eq. 23 is
+ ~ J'i(-div T;+qi)=O, ................. (35)
where
_ 1 iWf
CT=---
Vf aT
is the thermal expansivity. The N mass conservation equa-
tions (Eqs. 8 and 35) are inadequate for determining the
N+2 unknown functions p, T, rl .. . rN of the thermal
system. It is the energy conservation equation of the
fluid/rock system that serves as the N+2-th equation.
Let ef denote the energy of a unit volume of the total
fluid system and e r that of the reservoir rock:
ef= ~PjSjUj . .......................... (36)
j
549
Then the accumulation term of the energy conservation
equations can be written as
a(cpef) a[(I-cp)e r 1
--+ ..................... (37)
at at
The first part of the accumulation term in Eq. 37 is the
time derivative of the products as shown in Eq. 36. This
nonlinear term can be linearized in the following way.
The time derivative of cpef is
a(cpef) a(cpef) ap a(cpef) aT
--=---+---
at ap at aT at
+ ~ ~ a(cpr i ) , ..................... (38)
i a(cpri) at
from which the derivatives a(cpri)/at can be eliminated
by applying Eq. 8. Let eft denote the partial specific
energy; then
a(cpef) a(cpef) ap a(cpef) aT
--=---+---
at ap at aT at
+ ~eft(-div T;+qi), ................... (39)
is derived. If we apply some explicit calculations to give
the coefficients of the heat flux terms and substitute the
right side of Eq. 39 into the energy equation, then the
energy equation and Eq. 35 have only two unknown func-
tions, p and T. After determining p and T by a
simultaneous solution, we are able to calculate the overall
masses from the mass conservation equations.
Computer Program
The first program of the proposed method was realized
in FORTRAN in 1976. 18 It seemed, however, that because
of FORTRAN limitations, the versatile possibilities of dif-
ference equations could be realized only at the expense
of tedious and time-consuming programming, and even
in this case, the ready program could not easily be read,
debugged, or developed further. In view ofthis, a PLII
(IBM program language) program code was written. The
essence of the program is a program list of only two pages,
and it can be considered as a basis for a great number
of reservoir models; moreover, it can easily be developed
further. The best description of this basis program is the
program code itself. This code, together with a detailed
explanation of program identifiers and ~ome illustrations,
may be found in Ref. 19.
Comparison With Other IMPES
Compositional Models
The compositional models can be ranged into two
categories,11 Newton-Raphson (e.g., Refs. 4, 9 and 11)
and non-Newton-Raphson (e.g., Refs. 6, 8, and 10)
iterative methods. Each iterative step of these methods
essentially consists of three parts: (1) generation of the
550
linearized coefficients, (2) solution of a great system of
linear equations, and (3) calculation of the thermodynamic
properties of the fluid system.
In Newton-Raphson methods the number of linear equa-
tions is generally equal to the (number of grid points)
x(number of components); in non-Newton-Raphson
methods the number of linear equations is equal to the
number of grid points. The Newton-Raphson methods use
more computer time to calculate one iteration step than
the non-Newton-Raphson methods, but the latter methods
use more iteration steps for convergence. The number of
iteration steps in Newton-Raphson methods is 1 to 3, while
that in non-Newton-Raphson methods is 3 to 8. In the pro-
posed methods, Parts 1 through 3 are calculated only once
for a timestep. The proposed method gives the numerical
solution of the model in a direct way, which is the
method's fundamental advantage. As a consequence, it
is more effective than the known iterative methods and
guarantees a reliable technique for applying the most com-
plicated fluid systems.
Application
The model was first checked by comparing its results with
a 2D IMPES black-oil simulation, and with some ID com-
positional results obtained earlier by applying Newton-
Raphson and non-Newton-Raphson iterative composi-
tional techniques. The results agreed well; comparison
showed that the new method can be applied universally
for simulating various problems, and for compositional
cases it proved to be much more effective than the iterative
ones.
The model was first used to study CO 2 EOR. 20.21 A
three-phase lO-component fluid system was assumed. The
fluid properties were calculated on the basis of empirical
fluid correlations. The phase equilibrium calculations were
carried out by applying laboratory K values of a nine-
component gas/oil system and two-component
(C0 2/water) gas/water equilibrium data. The density of
the gas phase was calculated from the Redlich-Kwong
equation of state. When determining the densities of the
liquid phases, laboratory PVT measurement correlations
were used. On the basis of these measurements the FVF's
were calculated as functions of the dissolved gas/liquid
ratio for both iiquid phases. The solution gas contents Rso
and Rsw were determined by normal flash calculations
from the composition of the appropriate phase.
The model has been used successfully for simulating
ID laboratory micellar displacement experiments, and 2D
and 3D black-oil and compositional field performances.
Summary
I. A direct sequential IMPES-type technique has been
developed for black-oil and compositional cases. The
method treats the pressure determining the equilibrium
implicitly, the masses of components explicitly.
2. The pressure equation is derived from an in-situ
"volume balance" equation. The coefficients of the
pressure equation are determined from the derivatives of
the fluid volume function Vf(p,m) and from the
derivative of the PV Vp(p).
3. The primary variables and the governing difference
equations of the model (which are always symmetric for
the components) apply to any arbitrary fluid system. Ap-
SOCIETY OF PETROLEUM ENGINEERS JOURNAL
plication of the integrated finite difference method allows x = space vector
the use of an arbitrary grid system. a = constant
4. Systems with a different phase number are treated ak = unit time x unit pressure difference
in a uniform way: thus, modeling of the appearance or ao = unit time
disappearance of phases causes no difficulties. I.t= viscosity, miLt
5. By taking into account the volumetric error .1 va per p = density, miL 3
volume element, we have succeeded in avoiding the ac-
cP = porosity
cumulation of errors. The stability effect of this correc-
tion term permits the use of reasonable timesteps without
iteration. Subscripts
6. The computing work of a timestep of the method is
generally less than that of an iteration step of a Newton-
o == general volume element
Raphson composition model. The method is more effec- c = capillary
tive than the compositional IMPES techniques presented f = fluid
previously. In addition, it guarantees a reliable technique i = component, i = 1 ... N
for applying the most complicated fluid system. j = phase, j= 1. . . N P
k = neighboring volume element, k= 1 .. . M
Nomenclature P = pore
A = area, L2 r = rock
.it = surface vector
c = coefficient of the pressure difference Superscripts
equation n = new
c = compressibility, Lt 2 /m 0= old
c T = volume expansivity u = upstream
C = mass concentration
div = divergence
D = diffusion coefficient, L 2 It Acknowledgments
D t = diffusion tensor, L 2 It We thank the Hungarian Oil & Gas Trust for supporting
e = energy, mI/ It 2 this work, the Transdanubian Oil & Gas Production Co.,
g = gravity acceleration Zoltan Biro (who contributed extensively to the develop-
grad = gradient ment and first application of the model described here,
and Gyorgy Adler, Istvan Farago, and Csaba Hegedus
h = depth, L
r = mass rate vector for discussion during the elaboration of the mathematical
and PLI model.
k = relative permeability
K = absolute permeability, L 2
K t = absolute permeability tensor transmissibility
References
coefficient, L 2 1. Peaceman, D.W.: Fundamentals of Numerical Reservoir Simula-
tion. Elsevier Scientific Publishing Co., Amsterdam (1977).
L = component mass rate 2. Aziz, K. and Settari, A.: Petroleum Reservoir Simulation. Applied
Lo = masslunit time Science, London (1979).
Lk = masslunit timelunit pressure difference 3. Coats, K.H.: "Reservoir Simulation: State of the Art," J. Pet. Tech.
(Aug. 1982) 1633-42.
m = component mass, m 4. Coats, K.H.: "Reservoir Simulation: A General Model Formula-
M = number of neighboring elements tion and Associated Physical/Numerical Sources of Instability,"
N = number of components Boundary and Interior Layers-Computational and Asymptotic
Methods. J.J.H. Miller (ed.), Boole Press, Dublin (1980) 62-76.
N D = number of dependent thermodynamic 5. Kendall, R.P. et ai.: "Development of a Multiple Application Reser-
properties voir Simulator for Use on a Vector Computer," paper SPE 11483
N I = number of independent thermodynamic presented at the 1983 SPE Middle East Oil Technical Conference
and Exhibition, Manama, Bahrain, March 14-17.
properties 6. Roebuck, I.F. Jr. et al.: "The Compositional Reservoir Simulator:
N P = number of phases Case I-The Linear Model," Soc. Pet. Eng. J. (March 1969)
NT = number of thermodynamic properties 115-30; Trans., AIME, 246.
7. Nolen, J.S.: Numerical Simulation of Compositional Phenomena
p = pressure, m/Lt 2 in Petroleum Reservoirs. Reprint Series, SPE, Richardson, TX
Pc = capillary pressure, psi [kPa] (1973) 11, 269-84.
q = mass source, unit masslunit timelunit 8. Kazemi, H., Vestal, C.R., and Shank, G.D.: "An Efficient
Multicomponent Numerical Simulator," Soc. Pet. Eng. J. (Oct.
volume 1978) 355-68.
Q = mass source, masslunit time 9. Fussel, L.T. and FU8sel, D.O.: "An Iterative Technique for Com-
r = overall mass component density positional Reservoir Models," Soc. Pet. Eng. J. (Aug. 1979)
211-20.
Rs = solution gas content 10. Nghiem, L.X., Fong, O.K., and Aziz, K.: "Compositional Model-
S = phase saturation ing with an Equation of State," Soc. Pet. Eng. J. (Dec. 1981)
t = time 687-98.
11. Young, L.C. and Stephenson, R.E.: "A Generalized Compositional
V = volume, L 3 Approach for Reservoir Simulation," Soc. Pet. Eng. J. (Oct. 1983)
V = partial mass.volume, L3 727-42.

AUGUST 1985 551

12. Aziz, K.: "Compositional Modelling of EOR Processes," Proc., where the mean value Jik (referring to surface A k ) is the
1981 European Symposium on Enhanced Oil Recovery, component of the vector ~ normal to A k.
Boumemouth, England, Sept. 21-23, 367-78.
13. Narasimhan, T.N. and Witherspoon, P.A.: "An Integrated Finite
The mean value of function J ik (t) in the time interval
Difference Method for Analyzing Fluid Flow in Porous Media," t n -to is approximated as follows:
Water Resources Research (Feb. 1976) 12, 57-64.
14. Soave, G.: "Equilibrium Constants from a Modified Redlich-Kwong
Equation of State," Chern. Eng. Sci. (1972) 27, No.6 1197-1203.
15. Peng, D.Y. and Robinson, D.B.: "A New Two-Constant Equa-
tion of State," Ind. Eng. Chern. Fundarn. (1976) 15, No.1, 59-64.
16. Watts, J. W .: "A Compositional Formulation of the Pressure and
Saturation Equations," paper SPE 12244 presented at the 1983 SPE
Reservoir Simulation Symposium, San Francisco, Nov. 15-18; to
be published in Soc. Pet. Eng. J. (April 1986).
17. Spillette, A.G., Hillestad, J.G., and Stone, H.L.: "A High-Stability
Sequential Solution Approach to Reservoir Simulation," paper SPE
4542 presented at the 1983 SPE Annual Meeting, Las Vegas, Sept. o 0

30-0ct. 3.
+K ~ fl!. Pcjk -PcjO
k L..J Jlk A_
18. Acs, G. and Farkas, E: "Compositional Modeling of Three-Phase j ~k
Systems, " paper presented at the 1977 Oil and Gas Conference of
Hungarian Mining and Metallurgical Soc., Balatonfiired, Hungary,
Sept. 26-29.
~ 0 So DO CJik -CJiO)
19. Acs, G., Doleschall, S., and Farkas, E.: "General Purpose Com- + 'I',/.. L..J PjkO jkO jikO ,
positional Model," paper SPE 10515 presented at the 1982 SPE j Illk
Reservoir Simulation Symposium, New Orleans, Jan. 31-Feb. 3.
20. Acs, G., Biro, Z., Doleschall, S., and Farkas, E.: "A Three-Phase,
Compositional Model and Its Application for Describing CO 2 where
Displacement of the Budafa-West Reservoir," K60laj es F6ldgaz,
(Jan. 1981) 1-5; (Feb. 1981) 51-62.
21. Doleschall, S. et al.: "Complex Study of CO 2 Flooding in
Hungary, " Proc., 1981 European Symposium on Enhanced Oil
Recovery, Boumemouth, England, Sept. 21-23, 299-311. '

APPENDIX Djik is defined by Dji' similar to K k ; ffik has already been

In the following we summarize how the difference equa- defined and PikO, SJkO are average values.
tions (Eqs. 18 and 23) can be obtained from the differen- The introduction of the relation
tial equations (Eqs. 8 and 16). The given domain is divided
into finite elements having (rock) volumes V r . The
general volume element is denoted again by subscript 0,
the neighboring ones by subscripts 1, 2 ... M. As we have
seen, the expression
gives the right side of Eq. 18 as the difference approxi-
div ~ mation of V rOdt( -div J i +qJ; thus

occurs in all differential equations where

For the discretization of Eq. 8 the quantity

, k·p· is introduced and, based on this, the approximations

+K t ~ ....!.......:LCji grad Pcj
j P-j

and
+cp ~pjSjDji grad Cji)'
J 1 dmiO
---z-----

We deal with the transformation of these first. From the at V rO dt

definition of divergence, the average value of div in r;
the volume element having the volume VrO and bounded can be written. Substitution of the difference approxima-
by surfaces that have areas AI' .. A M is tion of the derivative into Eq. 8 gives
M
dmiO
~ Lik(dPk -dpo)+L iO '
dt k=!

552 SOCIETY OF PETROLEUM ENGINEERS JOURNAL

That is, the difference equation, Eq. 18, is obtained. By changing the order of summation, the equation is
To approximate Eq. 16, the notation

is introduced and this gives

+ ~ (avJLiO) 0= (dVp)OApO .
i ami dp t:.t

and
That is, the difference equation,
d<t> dVp(p)
-,::::-- M
dp Vt{) dp cJt:.Po + t:.t ~ c k(t:.P k - t:.po) + t:.tco =c pt:.po,
k=!
Thus, Eq. 16 can be approximated as follows.
is obtained, which when supplemented by the volumetric
p ) ap error term VJ - V$, which results in Eq. 23.
-1- (avJ dV-
---- --
V t{) ap dp at SPEJ

Original manuscript received in the Society of Petroleum Engineers office Feb. 3, 1982.
Paper accepted for publication March 9, 1983. Revised manuscript received July 9,
1984. Paper (SPE 10515) first presented at the 1982 SPE Reservoir Simulation Sym-
posium held in New Orleans Jan. 31-Feb. 3.

AUGUST 1985 553