You are on page 1of 9

High-Accuracy Implicit Finite-Difference

Simulations of Homogeneous and


Heterogeneous Miscible-Porous-Medium
Flows
Eckart Meiburg and C.-Y. Chen,* U. of Southern California

Summary dynamical interactions remains a formidable challenge. The nu-


A novel computational algorithm for the direct numerical simula- merical approach to be discussed in the present paper aims at this
tion of miscible-porous-media flows is developed and applied that area of direct numerical simulation.
offers multiple benefits. It is based on the streamfunction-vorticity Efforts in the field of porous media flow simulation date back
formulation of Darcy’s law and resolves all physically relevant at least to the early finite difference work by Peaceman and
length scales including diffusion. By employing an implicit dis- Rachford,1 as well as the mixed Lagrangian-Eulerian approach by
cretization based on compact finite differences, it combines ease Garder et al.2 Subsequently, a host of novel numerical approaches
of implementation and the ability to handle nontrivial geometries for the simulation of miscible displacements were developed. The
with the superior computational accuracy usually reserved for research groups around Wheeler and Ewing played a very active
spectral methods. The formal accuracy is O(⌬x 4 ,⌬t 2 ), although role in developing new computational techniques for porous-
diffusion terms are discretized with O(⌬x 6 ). Test calculations are media flows, cf. the early review by Russell and Wheeler.3 Dou-
performed that provide a quantitative comparison with linear sta- glas et al.4 employ a self-adaptive finite-element method, cf. also
bility results for strong mobility contrasts in the practically the analysis by Bell et al.5 Darlow et al.6 describe so-called
relevant-quarter five-spot configuration. The excellent agreement, mixed finite-element methods, which realize certain computa-
both with regard to the algebraic growth rate as well as the pre- tional advantages, in particular for strongly heterogeneous media,
ferred wave number, demonstrates the very low levels of numeri- by solving for pressure and velocity simultaneously. Ewing et al.7
cal diffusion, thereby eliminating grid orientation effects. The ca- analyze a modified method of characteristics for handling the con-
pabilities of the numerical method are furthermore demonstrated centration equation in miscible porous-media flows. Ewing et al.8
by means of representative calculations for both homogeneous perform more-detailed finite-element simulations for anisotropic,
and heterogeneous displacements. Among other findings, these heterogeneous, miscible flows in the quarter five-spot geometry,
simulations exhibit a minimal recovery efficiency for intermediate using finer grids up to 100⫻100. While these detailed simulations
values of the correlation length of the permeability heterogene- allow the authors to draw a variety of conclusions about the un-
ities. derlying physics, they nevertheless state a number of issues left
unresolved, thereby emphasizing the need for further, higher-
accuracy simulations. A comprehensive review of the above
methods, as well as related work on Eulerian-Langrangian
approaches9,10 is provided in Refs. 11 and 12, along with a dis-
Introduction cussion of these methods in light of recent developments in com-
The continuous enhancement of simulation techniques for flows in puter architectures, cf. also Ref. 13.
porous media poses a longstanding and ongoing challenge. Accu- More conventional finite-difference discretizations have been
rate predictive capabilities are crucial for, among other areas, used as well to obtain improved physical understanding,14-20 al-
problems in hydrology and groundwater contamination, as well as though alternate approaches such as particle tracking have also
enhanced oil recovery. Here, improved reservoir simulations will been pursued.21 These methods are typically of second-order
yield qualitative and quantitative information essential both for overall accuracy, but often still suffer from significant levels of
correctly anticipating oil reservoir performance, as well as for numerical dispersion. In terms of formal accuracy, an important
assessing the effectiveness of strategies for enhancing recovery. milestone is represented by the work of Leventhal,22 who applied
All of the above areas require the quantitatively accurate simula- the fourth-order operator compact implicit family of finite-
tion of complex physical processes in complicated geometries. difference schemes23,24 to one-dimensional, two-phase immiscible
The main difficulty lies in the correct reproduction of the wide waterflood problems, and demonstrated a significant reduction of
range of relevant length and time scales that typically characterize the adverse effects of numerical diffusion.
these problems. At one end of the spectrum, the size of the small- A different line of research has aimed at modifying numerical
est scales is determined by the action of physical diffusion or approaches originally developed for compressible flows with
dispersion mechanisms, while at the other end, reservoir size or shock discontinuities, e.g., Godunov-based approaches25,26 and
well distances determine the large scales. An improved under- true verticle depth 共TVD兲 techniques.27,28 These methods typically
standing at the fundamental level of how the governing physical are of second-order accuracy, although Liu et al.29 have recently
mechanisms and interactions between them are affected, for ex- developed a TVD scheme that discretizes first-order spatial de-
ample, by the mobility contrast or the level of heterogeneity, can rivatives with third-order accuracy. More recently, Batycky
potentially be obtained by direct numerical simulations that re- et al.30 utilize mapping of numerical solutions along streamlines
solve the entire spectrum of length scales, without introducing in order to simulate first-contact miscible displacement. Their
significant levels of numerical diffusion. Consequently, especially simulations, which take into account gravitational forces as well,
for low levels of physical diffusion and dispersion, the accurate exhibit vigorous viscous fingering. The method furthermore al-
and detailed representation of the dominant mechanisms and their
lows big improvements in efficiency, although in the absence of
physical diffusion or dispersion, numerical diffusion sets the
*Now at Da-Yeh U. short-wave cutoff length scale and, hence, has a substantial effect
Copyright © 2000 Society of Petroleum Engineers on the results.
Original SPE manuscript received for review 4 May 1998. Revised manuscript received 31
In order to overcome the accuracy limitations of the above
January 2000. Paper (SPE 50698) peer approved 16 February 2000. grid-based methods, several researchers have recently based their

SPE Journal 5 共2兲, June 2000 1086-055X/2000/5共2兲/129/9/$5.00⫹0.50 129


computational investigations of unstable rectilinear displacements handle arbitrary geometries. By realizing that the overall geom-
on significantly more accurate spectral methods.31-33 This class of etry is composed of many identical building blocks, i.e., many
numerical methods displays superior accuracy for smooth flows, quarter five-spot elements, and by assuming that the flow is iden-
i.e., for flows without discontinuities and singularities,34 and tical in each one of these, we can take one such building block as
avoids problems related to numerical diffusion. In addition, the our computational domain, provided that the proper symmetry
spectral simulations mentioned above were performed for the gov- boundary conditions are specified along its borders. We consider
erning equations formulated in terms of the vorticity and stream- the physical problem of an unstable, time-dependent, incompress-
function variables. By no longer needing to solve for the pressure ible flow involving two fluids miscible with each other in all pro-
variable, this form of the governing equations often leads to portions. Under Darcy’s law, such a displacement in a heteroge-
higher computational efficiency. Furthermore, it offers the advan- neous porous medium of permeability distribution k(x,y), with a
tage of satisfying the conservation of mass identically. In this representative value K, is governed by the dimensionless equa-
way, it has allowed for the detailed investigation of mobility and tions
gravity-driven fingering processes in rectilinear displacements at
relatively low levels of physical dispersion. Several parameter re- “•u⫽0, 共1兲
gimes of different dynamical properties could be identified, and

scaling laws for some of the global displacement features were “p⫽⫺ u, 共2兲
derived. k
The above discussion demonstrates the need to develop reser-
⳵c 1
voir simulation techniques that combine the superior accuracy of ⫹u•“c⫽ “ 2 c, 共3兲
spectral methods with the flexibility and ease of implementation ⳵t N Pe
of conventional grid-based methods, and with their ability to where the Peclet number N Pe has the form
handle nontrivial geometries. In this regard, the compact finite-
difference methods with spectral-like accuracy recently described Q
by Lele35 offer new opportunities. They are straightforward to N Pe ⫽ . 共4兲
D
implement, as they merely require the standard solution of typi-
cally tri- or pentadiagonal systems of equations. They can be com- Here, the lateral extent L of one unit of the flow field serves as the
bined with implicit or explicit time-stepping schemes, as will be characteristic length scale. In conjunction with the source strength
discussed below, which is important with regard to the overall per depth 2 ␲ Q, the time scale is L 2 /Q, and the velocity scale has
stability properties of convection-dominated simulations. Further- the value of Q/L. Viscosity and pressure are scaled with ␮ 1 and
more, their excellent properties regarding numerical dispersion Q ␮ 1 /K, respectively, where the subscript ‘‘1’’ indicates the dis-
and diffusion errors allow the accurate tracking of steep fronts in placing fluid. The above equations express the conservation of
miscible displacements at low levels of physical diffusion. It mass, momentum, and species, with c denoting the concentration
should be pointed out that the higher-order compact finite- of the displacing fluid. Two-dimensional flow in a horizontal
difference schemes described by Lele are qualitatively different in plane is assumed, so that gravitational effects do not enter into the
nature from compact schemes as defined by Spotz and Carey.36 problem. It is straightforward, however, to extend the present
These authors consider any finite-difference stencil ‘‘compact’’ computational approach to three-dimensional inhomogeneous po-
that is restricted to the cells immediately surrounding a given rous media flows including gravity override, cf. the formulation of
node, such as, for example, the standard second-order central dif- the governing equations presented by Zimmerman and Homsy.32
ference scheme. Higher accuracy is then obtained by using the Physical diffusion is assumed to be governed by a scalar diffusion
governing differential equation to approximate the leading order coefficient D. Again, the extension to a more realistic dispersion
truncation error terms. tensor is straightforward. To close the problem, we assume that
In the following, we develop and apply a computational algo- the viscosity ␮ depends on the concentration like
rithm for the simulation of homo- and heterogeneous miscible
porous media flows based on the compact finite-difference 1 d␮
␮ 共 c 兲 ⫽e R(1⫺c) , R⫽⫺ , 共5兲
schemes described by Lele. It should be pointed out that our goal ␮ dc
is, within the framework of Darcy flows, the direct numerical
simulation of all physically relevant processes within the frame- where R is a constant determined by the given combination of
work of Darcy flow, including physical diffusion or dispersion. fluids. For the purpose of solving these equations numerically, we
The numerical approach to be developed below is not intended for follow earlier authors31,33 and recast them into the well-known
the simulation of problems containing discontinuities or sharp streamfunction and vorticity ( ␺ , ␻ ) formulation
fronts, such as immiscible flows, flows with jumps in the perme- ⳵␺ ⳵␺ ⳵v ⳵u
ability distribution, or miscible flows in the limit of zero diffusion. u⫽ , v ⫽⫺ , ␻⫽ ⫺ . 共6兲
⳵y ⳵x ⳵x ⳵y
In other words, any variations, for example, in the concentration
or permeability fields, are to be resolved by a sufficient number of In this way, the continuity equation is satisfied identically, and the
gridpoints, so that gradients can be calculated accurately. For the governing set of equations takes the form
reasons mentioned above, the computational technique will be
based on the governing equations in the vorticity and streamfunc- 1
c t⫹ ␺ yc x⫺ ␺ xc y⫽ “ 2 c, 共7兲
tion formulation. In order to evaluate its capabilities and perfor- N Pe
mance, and its ability to handle more demanding geometries than
the rectilinear one, we will subsequently apply it to the practically ⳵ ln k ⳵ ln k
␻ ⫽⫺R 共 ␺ x c x ⫹ ␺ y c y 兲 ⫺ ␺ y ⫺␺x , 共8兲
relevant problem of the quarter five-spot pattern. This configura- ⳵y ⳵x
tion is computationally challenging, due to the nonperiodic
boundaries, the singularities near the injection and production ⵜ 2 ␺ ⫽⫺ ␻ . 共9兲
wells, and the large variations in velocities encountered in the One computational difficulty arises from the fact that the flow has
flow. Furthermore, many earlier authors have applied their meth- singularities at both well locations, making accurate computations
ods to this problem, so that it is well suited for comparing the difficult due to problems related to numerical differentiation near
performance of different approaches. these locations. We will employ an approach that elegantly over-
comes this obstacle. It is based on a splitting of the solution into a
Physical Problem and Governing Equations potential component that absorbs the singularities, plus a rota-
The algorithm will be developed for the practically relevant prob- tional part which is smooth and can be obtained with high accu-
lem of the quarter five-spot geometry. It should be kept in mind, racy via the above-mentioned compact finite-difference approach.
however, that the computational approach has the flexibility to The potential part of the flow consists of a two-dimensional stag-

130 E. Meiburg and C.-Y. Chen: High-Accuracy Implicit Finite-Difference Simulations SPE Journal, Vol. 5, No. 2, June 2000
gered array of point sources and sinks arranged in the quarter should be mentioned that the crucial step in demonstrating the
five-spot configuration. The corresponding velocity field can be feasibility of high-accuracy finite-difference algorithms for reser-
evaluated on the basis of the results provided by voir modeling lies in the implicit solution of the concentration
Morel-Seytoux,37,38 and also Muskat,39 who derives the following equation, as opposed to the Poisson equation.
expression for the velocity potential: Potential Velocity Field. For reasons that will become clear dur-

␾ 共 x,y 兲 ⫽
1
4␲
ln 冉
1⫺cn 2 xcn 2 y
cn 2 x⫹cn 2 y
. 冊 共10兲
ing the discussion of the solution procedure for the convection-
diffusion equation, it is useful to avoid the appearance of singu-
larities in the velocity components. For this purpose, we spread
Alternatively, approximate analytical expressions can be con- the effect of the point source and the point sink out over a small
structed to any desired accuracy, based on the relations provided but finite area, by assuming a Gaussian distribution of the diver-
by Lamb40 for a single, infinite, row of sources.41 By using these gence characterized by a core size ␴. Typically, our test simula-
analytically available expressions for the potential component of tions employed ␴ ⭐0.05. The radial velocity v near such a
the streamfunction ␺ pot and its derivatives, any inaccuracy related ‘‘Gaussian source’’ is
to the existence of the singularities can be avoided, yielding a
much more accurate representation of the flow near the injection 1 2 2
v⫽ 共 1⫺e ⫺ r / ␴ 兲 . 共17兲
and production wells. Furthermore, the closed-form solution of r
the potential flow component will aid in eliminating grid orienta-
The divergence due to the finite-source size is
tion effects.14,15,42 Since the Laplacian of the potential stream-
function component vanishes, we obtain 2 ⫺r 2 / ␴ 2
⑀ ⫽“•u⫽ e . 共18兲
ⵜ 2 ␺ rot ⫽⫺ ␻ , 共11兲 ␴2
␺ ⫽ ␺ pot ⫹ ␺ rot , 共12兲 It should be pointed out that this smearing out of the source will
have negligible effect on the flow as long as the size of the source
where ␺ rot is the rotational component of the streamfunction. The is smaller than the front radius. By adding to the above potential
symmetry boundary conditions at the sides are velocity field the effect of the local Gaussian spreading near 共0,0兲
x⫽0,1:c x ⫽0, ␻ ⫽0, ␺ rot ⫽0, k x ⫽0, 共13兲 and 共1,1兲, we obtain the final form of the potential velocity field
for the quarter five-spot geometry. Notice that the finite size of all
y⫽0,1:c y ⫽0, ␻ ⫽0, ␺ rot ⫽0, k y ⫽0. 共14兲 other sources and sinks can be neglected, since they are many core
sizes away.
Suitable initial conditions can be specified, for example, as
Spatial Discretization. The spatial discretization of the govern-
␺ 共 x,t⫽t i 兲 ⫽ ␺ pot 共 x兲 , 共15兲 ing equations is based on the compact finite-difference schemes
c 共 x,t⫽t i 兲 ⫽c 0 共 x兲 , 共16兲 presented by Lele.35 The evaluation of the first derivatives con-
tained in the convection terms is performed as follows: When they
where c 0 (x) is taken from the one-dimensional similarity solu- appear explicitly, we employ the sixth-order formulation
tions provided by Tan and Homsy43 for radial source flow. It
should be noted that, in contrast to rectilinear or pure radial flow, f i⫹2 ⫺ f i⫺2 f i⫹1 ⫺ f i⫺1
⬘ ⫹ f i⬘ ⫹ ␣ 1 f i⫹1
␣ 1 f i⫺1 ⬘ ⫽b 1 ⫹a 1 , 共19兲
the quarter five-spot pattern does not allow for a one-dimensional 4⌬ 2⌬
base state, from which a more complex flow would then evolve
through the development of instabilities. with
␣ 1 ⫽1/3, a 1 ⫽14/9, b 1 ⫽1/9. 共20兲
Computational Approach Here, ⌬ indicates the mesh size, which is identical in the x and y
While the spatial discretization will be based on compact finite directions. For reasons that will become clear in the discussion of
differences, the temporal discretization has to be chosen carefully the time-stepping scheme, we apply a fourth-order accurate dis-
as well. The traditional use of explicit methods, applied, for ex- cretization when the convection terms appear implicitly. In this
ample, by Tan and Homsy31 for rectilinear displacements, has case
considerable drawbacks for the quarter five-spot flow. The reason
lies in the singularities at the well locations, near which the fluid ␣ 1 ⫽2/11, a 1 ⫽16/11, b 1 ⫽⫺1/11. 共21兲
velocities are orders of magnitude larger than in the center of the In order to compute the second derivatives of the diffusion terms,
domain. For example, on a 1,0242 grid for a two-dimensional we apply the sixth-order formulation
simulation, the magnitude of the velocities will vary by O(103 )
between the gridpoints nearest to the wells and those at the center f i⫹2 ⫺2 f i ⫹ f i⫺2
⬙ ⫹ f i⬙ ⫹ ␣ f i⫹1
␣ f i⫺1 ⬙ ⫽b 2
of the computational domain. For explicit schemes, these small 4⌬ 2
regions near the wells consequently determine the maximum al-
lowable time step as a result of stability restrictions in terms of a f i⫹1 ⫺2 f i ⫹ f i⫺1
⫹a 2 , 共22兲
Courant-Fredrichs-Levy number. These problems can be over- ⌬2
come by employing an implicit computational approach for the
with
concentration equation, as will be described in detail below.
The elliptic Poisson equation for the determination of the ␣ 2 ⫽2/11, a 2 ⫽12/11, b 2 ⫽3/11. 共23兲
streamfunction from the vorticity distribution can be solved as
well using compact finite differences, in conjunction with a split- The discretization of the right-hand side of the vorticity equation
ting technique or a fast matrix inversion scheme. As a result, the is uniformly performed by means of the sixth-order compact
present numerical approach is applicable also to nonperiodic finite-difference expressions. At the boundaries, the relevant odd
boundaries, e.g., see Refs. 44 and 45. However, as a result of or even symmetry conditions dictated by the physical problem are
splitting the streamfunction ␺ into a rotational and an irrotational applied, respectively, so that the order of accuracy does not de-
component, the governing system of equations and boundary con- crease.
ditions for the quarter five-spot pattern can be solved more effi- Implicit Time Stepping. To overcome the limitations of explicit
ciently on the basis of fast Fourier transform methods. Conse- schemes, and in order to achieve second-order accuracy in time,
quently, for the present quarter five-spot geometry we opt for this we base our numerical solution of the concentration equation on
approach, although standard finite-difference algorithms for the an alternating-direction-implicit procedure.46 This approach splits
solution of the Poisson equation are available and can easily re- the full time step ⌬t into two half steps of size ⌬t/2. During the
place the present spectral module for more complex geometries. It first half step, the convection and diffusion terms in the x direc-

E. Meiburg and C.-Y. Chen: High-Accuracy Implicit Finite-Difference Simulations SPE Journal, Vol. 5, No. 2, June 2000 131
tion are treated implicitly, while those in the y direction are dealt
with explicitly. During the second half step, the roles are reversed,
so that now the y direction is implicit while the x direction is
explicit. The source term is treated implicitly during both half
steps.
With the spatial discretization in the form of compact finite
differences as described above, we obtain the following system of
semidiscrete equations for the first half step:
c i,*j ⫺c i,n j
⫽⫺ 共 uc 兲 x 兩 i,*j ⫺ 共 v c 兲 y 兩 i,n j ⫹ ⑀ i, j c i,*j
⌬t/2
1 1
⫹ c 兩* ⫹ c 兩n ,
N Pe xx i, j N Pe y y i, j
* ⫺c i⫹1,j
c i⫹1,j n
* ⫺ 共 v c 兲 y 兩 i⫹1,j
⫽⫺ 共 uc 兲 x 兩 i⫹1,j n
⫹ ⑀ i⫹1,j c i⫹1,j
*
⌬t/2
1 1
⫹ c 兩* ⫹ c 兩n ,
N Pe xx i⫹1,j N Pe y y i⫹1,j
* ⫺c i⫺1,j
n Fig. 1–Maximum vorticity value ␻ max in the flow field as a func-
c i⫺1,j
⫽⫺ 共 uc 兲 x 兩 i⫺1,j
* ⫺ 共 v c 兲 y 兩 i⫺1,j
n
⫹ ⑀ i⫺1,j c i⫺1,j
* tion of time for R Ä5 and N Pe Ä400. Excellent agreement is ob-
⌬t/2 served between the growth rate of the simulation and the alge-
braic growth rate predicted by Tan and Homsy „see Ref. 43…,
1 1 shown as a dotted line.
⫹ c 兩* ⫹ c 兩n .
N Pe xx i⫺1,j N Pe y y i⫺1,j
Here, n and * denote evaluation at the old and intermediate time
placements in homogeneous porous media provides an ideal op-
levels, respectively. These equations can then be combined in or-
portunity for comparison, since in the neighborhood of the injec-
der to apply the compact finite-difference formulation. The result-
tion well the quarter five-spot flow is nearly identical to a radial
ing system of equations is
source flow. For R⫽5, i.e., an unfavorable mobility ratio of 148.4,
uc 兩 i⫺2,j
* 冉 b1

b2
4⌬x 4N Pe ⌬x 2 u i⫺2,j
⫹uc 兩 i⫺1,j
*
a1
2⌬x冊 冉 the authors provide both the algebraic growth rate ␴ and the
wave-number n max 共number of fingers around the azimuth兲 of
maximum growth as a function of N Pe . Specifically, when apply-

a2

2␣
N Pe ⌬x 2 u i⫺1,j ⌬tu i⫺1,j

␣⑀ i⫺1,j
u i⫺1,j 冊 ing linear theory for the asymptotic limit N Pe →⬁ to N Pe ⫽400,
the authors obtain ␴ ⬇27 and n max⬇60. After an initial transient
phase the present quarter five-spot simulation for R⫽5 and N Pe
⫹uc 兩 i,*j ⫺ 冉 b2

2a 2

2

⑀ i, j
2N Pe ⌬x 2 u i, j N Pe ⌬x 2 u i, j ⌬tu i, j u i, j 冊 ⫽400 共5122 grid, ⌬t⫽10⫺5 兲 shows a distinct region of algebraic
growth, as can be seen from the data for the vorticity maximum
␻ max presented in Fig. 1. The slope corresponding to the growth
⫹uc 兩 i⫹1,j
* ⫺ 冉 a1

a2

2␣
2⌬x N Pe ⌬x 2 u i⫹1,j ⌬tu i⫹1,j

␣⑀ i⫹1,j
u i⫹1,j 冊 rate predicted by linear stability theory is indicated as well, and
visual inspection shows excellent agreement. By counting the

冉 冊
number of fingers emerging in the quarter five-spot flow, we ob-
b1 b2 tain n max⬇48– 56, which again indicates very good agreement.
⫹uc 兩 i⫹2,j
* ⫺ ⫹ This comparison with linear stability theory indicates that the ef-
4⌬x 4N Pe ⌬x 2 u i⫹2,j
fects of diffusion are properly simulated, so that the correct diffu-
2␣ n 2 n 2␣ n sive cutoff scale is obtained. The requirement to accurately repro-
⫽⫺ c ⫺ c ⫺ c duce the instability growth rate and the wave number of maximum
⌬t i⫺1,j ⌬t i, j ⌬t i⫹1,j
growth also serves as the main criterion to establish that the flow
␣ field is represented by a sufficiently fine grid. If the grid spacing
⫹ ␣ 共 v c 兲 y 兩 i⫺1,j
n
⫹ 共 v c 兲 y 兩 i,n j ⫹ ␣ 共 v c 兲 y 兩 i⫹1,j
n
⫺ c 兩n were too coarse, excessive numerical diffusion would be gener-
N Pe y y i⫺1,j
ated, which would affect the growth rate as well as the wave
1 ␣ number of the instability.
⫺ c y y 兩 i,n j ⫺ c 兩n . A major concern in the numerical simulation of unstable
N Pe N Pe y y i⫹1,j
porous-media displacements is the issue of grid orientation ef-
For the second half step, we obtain a corresponding set of equa- fects. For a careful recent analysis of this topic and a review of the
tions. This scheme has a formal accuracy of O(⌬t 2 ) in time, pertinent literature, see Ref. 42. The authors identify as the under-
combined with O(⌬x 6 ) for the diffusion terms. The convection lying cause for grid orientation effects the anisotropic nature of
terms are discretized with O(⌬x 6 ) when explicit, and with numerical diffusion introduced by discrete finite-difference opera-
O(⌬x 4 ) when implicit. tors. As a result, grid orientation effects are shown to persist as
long as the numerical diffusion is comparable to or larger than the
Validation physical diffusion, which is the case for most of the reservoir
Rigorous validation represents an important step in establishing simulations reported in the literature. Our above comparison with
the accuracy and convergence properties of a novel numerical linear stability results, on the other hand, demonstrates that due to
approach. A popular test in the field of porous-media flows con- the high accuracy of the present numerical technique, the amount
sists of a comparison with the analytical results obtained by of artificial diffusion introduced by the discretization is much
Morel-Seytoux37,38 for unit mobility ratio flows. However, such a smaller than the physical diffusion. If substantial numerical diffu-
comparison cannot show whether the complex coupling between sion were present, an effectively lower value of N Pe would result,
concentration and viscosity variations and the velocity field is which should affect both the algebraic growth rate and the number
represented accurately. Consequently, rigorous testing requires a of emerging fingers. The excellent agreement between simulation
more challenging test case. In this regard, the linear stability results and linear theory thus implies that grid orientation effects
analysis by Tan and Homsy43 of radially symmetric miscible dis- are not a concern for the present simulations. The comparison

132 E. Meiburg and C.-Y. Chen: High-Accuracy Implicit Finite-Difference Simulations SPE Journal, Vol. 5, No. 2, June 2000
Fig. 2–Homogeneous displacement with R Ä2.5 and N Pe
Ä2,000. Shown are concentration contours at times 0.07, 0.1,
0.16, and 0.24. After the formation of many fingers during the
initial stages, nonlinear interactions lead to a reduction in their
number. Shortly before breakthrough, the acceleration of the Fig. 3–Temporal evolution of the mixing length as a function of
dominant finger leads to a renewed generation of small scales. N Pe for R Ä2.5. - * -: N Pe Ä200, - x -: N Pe Ä400, -䊊-: N Pe Ä800,
-¿-: N Pe Ä1,200, and -"-: N Pe Ä2,000. Two solid lines represent
the slopes of 0.5 and 1.
with linear stability theory thus represents a very useful tool for
obtaining information about the presence of numerical diffusion,
and it should serve as a universal test case for direct numerical of the displacement process. For unstable rectilinear displace-
simulations of miscible quarter five-spot displacements. ments, this is typically done by analyzing the temporal evolution
Results for Representative Flow Fields. Homogeneous of the mixing length L mix , which tracks the distance between the
Displacement. In the following, a representative calculation for 0.1 and 0.9 contours of the spanwise-averaged concentration pro-
R⫽2.5 and N Pe ⫽2,000 in a homogeneous environment is de- file. It is well established that for small initial perturbations the
scribed. It should be emphasized that the simulation results given early stages are dominated by diffusion or dispersion, respec-
below merely serve to illustrate the capabilities of the numerical tively, so that L mix grows proportionally to t 1/2. Once finite-
approach. They are not intended to provide an in-depth analysis of amplitude fingers appear, the evolution of L mix becomes increas-
the physical mechanisms at work, as such an analysis is provided ingly dominated by convective effects, and its average growth is
elsewhere already.47-49 The simulation starts at t i ⫽0.005 with a approximately proportional to t. 31,50-55 It is not immediately ob-
nearly radially symmetric front. A very small asymmetry is im- vious how an equivalent mixing length can be defined for quarter
posed on the concentration front radius in order to break the arti- five-spot flows. Averaging over the azimuthal direction would be
ficial symmetry about the diagonal. Vigorous fingering with the appropriate for radially symmetric source flows, but not for the
correct growth rate and the number of fingers predicted by linear present, spatially varying base flow. Even taking the average
stability theory sets in even without this perturbation, however, along equipotential lines of the constant mobility irrotational flow
the fingering pattern would stay symmetric with respect to the does not give a meaningful quantity, due to the mobility ratio
diagonal of the quarter five-spot domain. Concentration contours, dependence of the base flow. In the following, we will attempt to
computed on a 1,0242 grid, are shown in Fig. 2 for times t provide a global characterization of quarter five-spot flows in
⫽0.07, 0.1, 0.16, and 0.24. During the early, nominally nearly terms of the mixing area, defined as the area between the 0.1 and
radially symmetric stages, the unstable flow gives rise to numer- 0.9 contours. By dividing this mixing area by the instantaneous
ous small fingers, whose algebraic growth rate and azimuthal perimeter of an equivalent circular area of injected fluid, we then
wave number are again in excellent quantitative agreement with obtain a mixing length whose temporal evolution can be tracked.
linear stability results. Following a period of strong growth, the However, it needs to be kept in mind that this mixing length is
fingers soon reach amplitudes large enough for nonlinear effects still not strictly comparable to the one commonly employed in
to become important. Complex interaction effects between them rectilinear displacements. For example, in the limit of vanishing
reduce their number to O(10), mostly due to shielding and merg- diffusion, the front thickness will tend to zero, and according to
ing events, and in spite of an occasional tip splitting. The remain- our definition, the mixing length will vanish as well, in spite of the
ing fingers continue to interact strongly with each other, in par- presence of vigorous fingering. The rectilinear mixing length defi-
ticular, through partial merging, shielding, and splitting events. In nition, on the other hand, will result in a finite value, due to the
the process, several small islands of resident fluid are formed and averaging process it is based on.
left behind. The fingers near the edges of the domain keep grow- Fig. 3 depicts the development of L mix with time for R⫽2.5,
ing until relatively late times, in spite of the eventual dominance with N Pe as a parameter. During the initial stages and for lower
by those fingers located near the diagonal. The calculation dem- N Pe values, when the nearly radially symmetric displacement is
onstrates the ability of the numerical approach to simulate com- dominated by diffusion, the self-similar solution of the base-flow
plex dynamical events involving strongly disparate length and concentration profile indicates that the front thickness grows pro-
time scales. portionally to t 1/2. Since the perimeter of the injected area grows
It is of interest to employ the detailed information provided by approximately like t 1/2 as well, the overall mixing area exhibits
the above simulations in order to characterize the global features growth proportional to t. The mixing length, calculated along the

E. Meiburg and C.-Y. Chen: High-Accuracy Implicit Finite-Difference Simulations SPE Journal, Vol. 5, No. 2, June 2000 133
Fig. 4–Isocontours of the permeability field for the correlation
length L Ä0.02. Lighter areas indicate higher-permeability val-
ues.

lines described above, then is characterized by growth propor-


tional to t 1/2. The smaller L mix values observed for larger N Pe
values and constant t i reflect the diminished importance of diffu-
sion. As the initially small perturbations grow to finite amplitudes, Fig. 5–Heterogeneous displacement with N Pe Ä800, R Ä2.5, L
the diffusion-dominated early stage gives way to a second phase Ä0.02, and s Ä0.5. Shown are the concentration contours at
characterized by significantly faster growth of L mix . Due to the times 0.05, 0.15, 0.2, and 0.2402, which represents the time of
larger instability growth rates, this transition occurs at earlier breakthrough. Vigorous fingering sets in considerably earlier
times for higher N Pe values. Fig. 3 shows that during this second than in the homogeneous case.
stage, the mixing length grows at rates up to t 2 . For the N Pe
values of 800 and 1,200, eventually a third phase is observed,
during which the mixing length grows approximately linearly with required approximately 5 hours of CPU time on a single CRAY
time. Surprisingly, the N Pe ⫽2,000 simulation shows a different C-90 processor.
behavior again, with growth proportional to t 1/2 during most of the Already at the early time of t⫽0.05, a vigorous fingering ac-
displacement. tivity is visible in a plot of the concentration contours. This is in
More detailed results, with a focus on the dynamical evolution marked contrast to the homogeneous case, which for the same
of homogeneous flow fields as a function of the mobility ratio and physical parameters at this stage still displays a nearly radially
the Peclet number, are reported elsewhere.47 For a comparison symmetric front. Hence, for the present parameter values the per-
with experimental investigations in the Hele-Shaw geometry, cf. meability inhomogeneity plays a crucial role in the evolution of
Ref. 56. the flow. The number of emerging fingers is comparable to the
Heterogeneous Displacement. In generating the desired statis- homogeneous case. Their nonlinear evolution is characterized by
tical distribution of the permeability field, we employ an algo- a sequence of tip-splitting, shielding, and merging events that
rithm provided by Shinozuka and Jen.57 This approach, success- qualitatively resemble some of the patterns observed earlier in
fully employed by Tan and Homsy58 in their numerical both rectilinear31 and quarter five-spot47 homogeneous flows at
simulations of rectilinear heterogeneous flows, yields a lognormal higher N Pe values. The heterogeneities are thus seen to encourage
permeabilty field k(x), characterized by a variance s and a spatial these mechanisms already at lower Peclet numbers. Furthermore,
correlation scale l. Special care has to be taken in order to satisfy in the heterogeneous environment the fingers display an increased
the symmetry boundary conditions for the permeability distribu- tendency to develop ‘‘side bumps’’ at locations where the displac-
tion at the edges of the quarter five-spot domain. This is accom- ing, less viscous fluid begins to enter a high-permeability region,
plished by adding a suitable term to the original permeability dis- only to be shielded soon thereafter by the growing main finger.
tribution, and by letting this additional term die out with Hence, these bumps are different in origin from the sidefingering
increasing distance from the boundary, cf. Ref. 41. The effect of observed by Rogerson and Meiburg,33 which were due to a sec-
this artificial modification of the permeability distribution dies out ondary instability. In contrast to its homogeneous counterpart, the
over a distance of approximately one half of the correlation heterogeneous case gives rise to the emergence of large-amplitude
length, so that its influence in the interior of the flow field is fingers also near the boundaries, and not just near the main diag-
negligible. In some sense, maintaining the symmetry of the het- onal. The earlier onset of fingering triggered by the permeability
erogeneity field is, of course, an artificial construct. However, we field also leads to the more rapid emergence of a few dominant
felt it would be most instructive to compare homogeneous and fingers near the main diagonal. The breakthrough time of 0.2402
heterogeneous true-quarter five-spot flows, i.e., flows in which is approximately 10% smaller than for the homogeneous case,
there is no flux across the boundaries of the quarter five-spot indicating a correspondingly reduced recovery.
element. For a permeability field characterized by L x ⫽L y ⫽L In order to investigate the effect of the correlation length on the
⫽0.02 and s⫽0.5 共cf. Fig. 4兲, a representative displacement for displacement process, a series of simulations for N Pe ⫽800, R
the parameter values N Pe ⫽800 and R⫽2.5 is shown in Fig. 5. ⫽2.5, and s⫽0.5 was conducted, with L taking the values 0.2,
This value of s results in a ratio of maximum-to-minimum per- 0.1, 0.05, and 0.02. The surprising observation was made that
meability of approximately 5. It should be mentioned that the breakthrough occurs earliest for intermediate values of L. This
effects of making the heterogeneity field symmetric at the bound- indicates the transition from a dispersion-dominated flow to one
aries is felt only within a very narrow strip of width 0.01 along the dominated by channeling, with an optimal coupling of viscous
boundaries. Since the flow develops mostly along the diagonal, fingering and heterogeneity effects at intermediate values of L. It
i.e., away from the boundaries, the modification to the heteroge- confirms earlier observations of a resonance phenomenon by Tan
neity field is not expected to have a significant influence on the and Homsy58 as well as De Wit and Homsy.59,60
flow. The calculation is started with the radially symmetric simi- Fig. 6 shows the development of the mixing length as a func-
larity solution for the concentration profile in a homogeneous en- tion of the governing parameters. For N Pe ⫽400, s⫽0.5, and L
vironment at time t i ⫽0.02. Carried out on a 512⫻512 grid, it ⫽0.1, we observe the growth to switch from an approximately t 1/2

134 E. Meiburg and C.-Y. Chen: High-Accuracy Implicit Finite-Difference Simulations SPE Journal, Vol. 5, No. 2, June 2000
Fig. 7–Breakthrough recovery data for N Pe Ä400 and R Ä2.5 as
a function of the heterogeneity index H Ä s 2 L .

numerical diffusion is to be avoided. We wish to state clearly that


in its present form the numerical method is not yet intended for
the simulation of flows with jumps or discontinuities, i.e., spatial
variations that are too steep to be adequately resolved by the grid,
such as occur in immiscible flows, miscible flows in the limit of
Fig. 6–Mixing length growth with time. As the flow transitions zero diffusion, or in flows with discontinuous jumps in the per-
from the diffusionÕdispersion regime to one in which fingering
meability distribution. However, one can envision extensions of
andÕor channeling are dominant, the growth rate increases
from approximately t 1Õ2 to faster than linear with time.
the present approach that would be capable of dealing with dis-
continuous permeability distributions, for example, by dividing
the overall domain into zones, with the zonal boundaries given by
the lines along which the permeability is discontinuous. The ac-
behavior for the dispersion-dominated displacement at R⫽0 to curacy of the present computational method was verified by com-
faster than t for larger R values. Similar trends are found as well parison with linear stability results derived by Tan and Homsy43
when we vary the other parameters: As the flow transitions from for radially symmetric source flow. Excellent agreement was ob-
the diffusion/dispersion regime to one in which fingering and/or tained, both with regard to the algebraic growth rate, as well as the
channeling are dominant, the growth rate increases to faster than preferred wave number. The correct number of fingers develops
linear with time. Keeping in mind the goal of predictive capabili- automatically, without any externally prescribed perturbations,
ties, it is also of interest to revisit the heterogeneity index:52,53,61 such as artificial permeability inhomogeneities. To our knowl-
H⫽s 2 L, 共24兲 edge, the above comparison with linear stability theory provides a
significantly more demanding check of the method’s accuracy
in light of the above findings. The hope is that this index will than the validation procedures reported in the literature. It implic-
account for the effects of both the variance and the correlation itly also checks for the presence of grid orientation effects. At the
length, so that recovery data, for example, will depend on H only, same time, it needs to be pointed out that the resolution require-
and not on s and l separately. Fig. 7 depicts the recovery data for ments for direct numerical simulations, in which all length scales
our simulations at N Pe ⫽400 and R⫽2.5, for a variety of s and L are to be resolved, are high.
values. While there is significant statistical scatter in our data, it The capabilities of the numerical method were demonstrated by
does not appear that H is well suited for reducing the number of means of a series of simulations for a variety of governing param-
independent parameters. If the recovery data depended on H only, eter values, for both homogeneous and heterogeneous quarter
the simulations for (s,L)⫽(0.5,0.5), 共0.8,0.2兲, and 共1,0.1兲 should five-spot flows. The large amount of detailed information allows
result in similar values, which obviously is not the case. This does us to perform in-depth investigations of the governing physical
not come as a surprise, considering that we found the recovery to mechanisms and their interactions over a large range of mobility
decline uniformly with increasing permeability variance,48 while ratios, Peclet numbers, and heterogeneity distributions. The tech-
at the same time it depends nonuniformly on the correlation nique has proved to be robust, provided that the grid size is suf-
length. A more-detailed discussion of heterogeneous quarter five- ficiently fine to resolve all relevant length scales. Empirical obser-
spot flows can be found in Chen and Meiburg,48 while Pankiewitz vations have furthermore shown that the allowable time step is
and Meiburg49 focus on the effects of nonmonotonic viscosity affected both by the mobility ratio and by the Peclet number. The
profiles. present method thus represents a tool suitable for highly accurate
investigations of miscible-porous-media flows in nontrivial geom-
Conclusions etries.
A highly accurate computational method for the direct numerical
simulation of heterogeneous miscible-porous-media flows has Nomenclature
been developed, validated, and applied to the practically relevant u ⫽ velocity vector
problem of flow in the quarter five-spot geometry. The numerical p ⫽ pressure
technique is intended for problems in which a short-wave cutoff ␮ ⫽ viscosity
mechanism exists as a result of physical diffusion or dispersion. k ⫽ permeability
For such problems, the goal is a direct numerical simulation of all c ⫽ concentration of solvent
of the physically relevant length scales, while the introduction of t ⫽ time

E. Meiburg and C.-Y. Chen: High-Accuracy Implicit Finite-Difference Simulations SPE Journal, Vol. 5, No. 2, June 2000 135
N Pe ⫽ Peclet number 14. Christie, M.A. and Bond, D.J.: ‘‘Detailed Simulation of Unstable Pro-
Q ⫽ source strength cesses in Miscible Flooding,’’ SPERE 共November 1987兲 514; Trans.,
D ⫽ diffusion coefficient AIME, 283.
R ⫽ viscosity parameter 15. Christie, M.A.: ‘‘High-Resolution Simulation of Unstable Flows in
Porous Media,’’ SPERE 共August 1989兲 297; Trans., AIME, 287.
u, v ⫽ velocity components
16. Bratvedt, F. et al.: ‘‘A New Front-Tracking Method for Reservoir
x,y ⫽ spatial directions Simulation,’’ SPERE 共February 1992兲 107.
␺ ⫽ streamfunction 17. Fayers, F.J., Blunt, M.J., and Christie, M.A.: ‘‘Comparisons of Em-
␻ ⫽ vorticity pirical Viscous Fingering Models and Their Calibration for Hetero-
␴ ⫽ radius of Gaussian source, linear growth rate geneous Problems,’’ SPERE 共May 1992兲 195; Trans., AIME, 293.
r ⫽ radial coordinate 18. Christie, M.A., Muggeridge, A.H., and Barley, J.J.: ‘‘3D Simulation
⑀ ⫽ divergence of Viscous Fingering and WAG Schemes,’’ SPERE 共February 1993兲
␣ ,a,b ⫽ numerical coefficients 19; Trans., AIME, 295.
f ⫽ function value 19. Sorbie, K.S., Zhang, H.R., and Tsibulkis, N.B.: ‘‘Linear Viscous Fin-
⌬,⌬x,⌬y ⫽ grid spacing gering: New Experimental Results, Direct Simulation, and the Evalu-
⌬t ⫽ time step ation of Averaged Models,’’ Chem. Eng. Sci. 共1995兲 50, 601.
n ⫽ wave number 20. Zhang, H.R., Sorbie, K.S., and Tsibuklis, N.B.: ‘‘Viscous Fingering
in Five-Spot Experimental Porous Media: New Experimental Results
L mix ⫽ mixing length
and Numerical Simulation,’’ Chem. Eng. Sci. 共1997兲 52, 37.
s ⫽ correlation variance 21. Tchelepi, H.A. and Orr, F.M. Jr.: ‘‘Dispersion, Permeability Hetero-
L,L x ,L y ⫽ correlation lengths geneity, and Viscous Fingering: Acoustic Experimental Observations
H ⫽ heterogeneity index and Particle-Tracking Simulations,’’ Phys. Fluids A 共1993兲 5, 1558.
␾ ⫽ potential 22. Leventhal, S.H.: ‘‘The Operator Compact Implicit Method for Reser-
voir Simulation,’’ SPEJ 共June 1980兲 120; Trans., AIME, 269.
Acknowledgments 23. Berger, A.E. et al.: ‘‘Generalized OCI Schemes for Boundary-Layer
Partial support by a NATO Collaborative Research Grant, by the Problems,’’ Math. Comput. 共1980兲 37, 79.
National Science Foundation in the form of an equipment grant, 24. Morton, K.W.: Numerical Solution of Convection-Diffusion Prob-
and by the Chevron Petroleum Technology Company is gratefully lems, Chapman and Hall, London 共1996兲.
acknowledged. We, furthermore, thank the NSF-sponsored San 25. Bell, J.B., Dawson, C.N., and Shubin, G.R.: ‘‘An Unsplit, Higher-
Diego Supercomputer Center for providing computing time. Order Godunov Method for Scalar Conservation Laws in Multiple
Dimensions,’’ J. Comput. Phys. 共1988兲 74, 1.
26. Bell, J.B., Colella, P., and Trangenstein, J.A.: ‘‘Higher-Order Go-
References dunov Methods for General Systems of Hyperbolic Conservation
1. Peaceman, D.W. and Rachford, H.H. Jr.: ‘‘Numerical Calculation of
Laws,’’ J. Comput. Phys. 共1989兲 82, 362.
Multidimensional Miscible Displacement,’’ SPEJ 共December 1962兲
27. Blunt, M., and Rubin, B.: ‘‘Implicit Flux-Limiting Schemes for Pe-
327; Trans., AIME, 225.
troleum Reservoir Simulation,’’ J. Comput. Phys. 共1992兲 102, 194.
2. Garder, A.O., Peaceman, D.W., and Pozzi, A.L. Jr.: ‘‘Numerical Cal-
28. Rubin, B. and Edwards, M.: ‘‘Extension of the TVD Midpoint
culation of Multidimensional Miscible Displacement by the Method
Scheme to Higher-Order Accuracy,’’ paper SPE 25265 presented at
of Characteristics,’’ SPEJ 共March 1964兲 26; Trans., AIME, 231.
3. Russell, T.F. and Wheeler, M.F.: ‘‘Finite-Element and Finite- the 1993 SPE Reservoir Simulation Symposium, New Orleans, 28
Difference Methods for Continuous Flows in Porous Media,’’ The February–3 March.
Mathematics of Reservoir Simulation, R.E. Ewing 共ed.兲, SIAM, Phila- 29. Liu, J., Pope, G.A., and Sepehrnoori, K.: ‘‘A High-Resolution, Fully
delphia, Pennsylvania 共1983兲 35. Implicit Method for Enhanced Oil Recovery Simulation,’’ paper SPE
4. Douglas, J. Jr. et al.: ‘‘Self-Adaptive Finite Element Simulation of 29098 presented at the 1995 SPE Symposium on Reservoir Simula-
Miscible Displacement in Porous Media,’’ Comput. Methods Appl. tion, San Antonio, Texas, 12–15 February.
Mech. Eng. 共1984兲 47, 131. 30. Batycky, R.P., Blunt, M.J., and Thiele, M.R.: ‘‘A 3D Field Scale
5. Bell, J.B., Shubin, G.R., and Wheeler, M.F.: ‘‘Analysis of a New Streamline Simulator With Gravity and Changing Well Conditions,’’
Method for Computing the Flow of Miscible Fluids in a Porous Me- paper SPE 36726 presented at the 1996 SPE Annual Technical Con-
dium,’’ SIAM (Soc. Ind. Appl. Math.) J. Numer. Anal. 共1985兲 22, ference and Exhibition, Denver, Colorado, 6–9 October.
1041. 31. Tan, C.T. and Homsy, G.M.: ‘‘Simulation of Nonlinear Viscous Fin-
6. Darlow, B.L., Ewing, R.E., and Wheeler, M.F.: ‘‘Mixed Finite- gering in Miscible Displacement,’’ Phys. Fluids 共1998兲 31, 1330.
Element Method for Miscible Displacement Problems in Porous Me- 32. Zimmerman, W.B. and Homsy, G.M.: ‘‘Three-Dimensional Viscous
dia,’’ SPEJ 共August 1984兲 391; Trans., AIME, 277. Fingering: A Numerical Study,’’ Phys. Fluids A 共1992兲 4, 1901.
7. Ewing, R.E., Russell, T.F., and Wheeler, M.F.: ‘‘Convergence Analy- 33. Rogerson, A. and Meiburg, E.: ‘‘Numerical Simulation of Miscible
sis of an Approximation of Miscible Displacement in Porous Media Displacement Processes in Porous Media Flows Under Gravity,’’
by Mixed Finite Elements and a Modified Method of Characteris- Phys. Fluids A 共1993兲 5, 2644.
tics,’’ Comput. Methods Appl. Mech. Eng. 共1984兲 47, 73. 34. Gottlieb, D. and Orszag, S.A.: Numerical Analysis of Spectral Meth-
8. Ewing, R.E., Russel, T.F., and Young, L.C.: ‘‘An Anisotropic ods: Theory and Applications, SIAM, Philadelphia, Pennsylvania
Coarse-Grid Dispersion Model of Heterogeneity and Viscous Finger- 共1977兲.
ing in Five-Spot Miscible Displacement That Matches Experiments 35. Lele, S.K.: ‘‘Compact Finite-Difference Schemes With Spectral-Like
and Fine-Grid Simulations,’’ paper SPE 18441 presented at the 1989 Resolution,’’ J. Comput. Phys. 共1992兲 103, 16.
SPE Reservoir Simulation Symposium, Houston, 6–8 February. 36. Spotz, W.F. and Carey, G.F.: ‘‘High-Order Compact Scheme for the
9. Ewing, R.E. and Wang, H.: ‘‘Eulerian-Lagrangian Localized Adjoint Steady Stream Function Vorticity Equations,’’ Int. J. Numer. Methods
Methods for Variable-Coefficient Advective-Diffusive-Reactive Eng. 共1995兲 38, 3497.
Equations in Groundwater Contaminant Transport,’’ Advances in Op- 37. Morel-Seytoux, H.J.: ‘‘Analytical-Numerical Method in Waterflood-
timization and Numerical Analysis, S. Gomez and J.P. Hennart 共eds.兲, ing Predictions,’’ SPEJ 共September 1965兲 247; Trans., AIME, 234.
Kluwer Academic Publishers, Dordrecht, The Netherlands 共1994兲 38. Morel-Seytoux, H.J.: ‘‘Unit Mobility Ratio Displacement Calcula-
Vol. 275, 185. tions for Pattern Floods in Homogeneous Medium,’’ SPEJ 共Septem-
10. Wang, H., Ewing, R.E., and Russell, T.F.: ‘‘Eulerian-Lagrangian Lo- ber 1966兲 217; Trans., AIME, 237.
calized Adjoint Methods for Convection-diffusion Equations and 39. Muskat, M.: ‘‘The Flow of Homogeneous Fluids Through Porous
Their Convergence Analysis,’’ IMA J. Numer. Anal. 共1995兲 15, 405. Media,’’ McGraw-Hill Book Co., New York City 共1937兲.
11. Ewing, R.E.: ‘‘Recent Developments in Reservoir Simulation,’’ 40. Lamb, Sir H.: Hydrodynamics, Dover, New York City 共1932兲.
North Sea Oil and Gas Reservoirs, III, J.O. Aasen et al. 共eds.兲, Klu- 41. Chen, C.-Y.: ‘‘Topics in Miscible-Porous-Media Flows,’’ PhD disser-
wer Academic Publishers, Dordrecht, The Netherlands 共1994兲 233. tation, U. of Southern California, Los Angeles, California 共1998兲.
12. Ewing, R.E.: ‘‘Multiphase Flows in Porous Media,’’ Advanced Math- 42. Brand, C.W., Heinemann, J.E., and Aziz, K.: ‘‘The Grid Orientation
ematics: Computations and Applications, A.S. Alekseev and N.S. Ba- Effect in Reservoir Simulation,’’ paper SPE 21228 presented at the
khvalov 共eds.兲, NCC Publishers, Novosibirsk, Russia 共1995兲 49. 1991 SPE Symposium on Reservoir Simulation, Anaheim, California,
13. Arbogast, T. et al.: ‘‘Computational Methods for Multiphase Flow 17–20 February.
and Reactive Transport Problems Arising in Subsurface Contaminant 43. Tan, C.T. and Homsy, G.M.: ‘‘Stability of Miscible Displacements in
Remediation,’’ J. Comput. Appl. Math. 共1996兲 74, 19. Porous Media: Radial Source Flow,’’ Phys. Fluids 共1987兲 30, 1239.

136 E. Meiburg and C.-Y. Chen: High-Accuracy Implicit Finite-Difference Simulations SPE Journal, Vol. 5, No. 2, June 2000
44. Ruith, M. and Meiburg, E.: ‘‘Miscible Rectilinear Displacements 55. Lenormand, R.: ‘‘Transport Equations for Fluid Displacements in
With Gravity Override. Part 1: Homogeneous Porous Medium,’’ J. Stratified Porous Media: The MHD Model,’’ paper SPE 30797 pre-
Fluid Mech. 共in press兲. sented at the 1995 SPE Annual Technical Conference and Exhibition,
45. Camhi, E., Meiburg, E., and Ruith, M.: ‘‘Miscible Rectilinear Dis- Dallas, 22–25 October.
placements With Gravity Override, Part 2: Heterogeneous Porous 56. Petitjeans, P. et al.: ‘‘Miscible Quarter Five-Spot Displacements in a
Media,’’ J. Fluid Mech. 共in press兲. Hele-Shaw Cell and the Role of Flow-Induced Dispersion,’’ Phys.
46. Peaceman, D.W. and Rachford, H.H.: ‘‘The Numerical Solution of Fluids 共1999兲 7, 1705.
Parabolic and Elliptic Differential Equations,’’ SIAM J. 共1955兲 3, 28. 57. Shinozuka, M. and Jen, C.-M.: ‘‘Digital Simulation of Random Pro-
47. Chen, C.-Y. and Meiburg, E.: ‘‘Miscible-Porous-Media Displace- cesses and its Applications,’’ J. Sound Vib. 共1972兲 25, 111.
ments in the Quarter Five-Spot Configuration. Part 1: The Homoge-
58. Tan, C.T. and Homsy, G.M.: ‘‘Viscous Fingering With Permeability
neous Case,’’ J. Fluid Mech. 共1998兲 371, 233.
Heterogeneity,’’ Phys. Fluids A 共1992兲 4, 1099.
48. Chen, C.-Y. and Meiburg, E.: ‘‘Miscible-Porous-Media Displace-
ments in the Quarter Five-Spot Configuration. Part 2: Effect of Het- 59. De Wit, A. and Homsy, G.M., ‘‘Viscous Fingering in Periodically
erogeneities,’’ J. Fluid Mech. 共1998兲 371, 269. Heterogeneous Porous Media. I: Formulation and Linear Instability,’’
49. Pankiewitz, C. and Meiburg, E.: ‘‘Miscible-Porous-Media Displace- J. Chem. Phys. 共1997兲 107, 9609.
ments in the Quarter Five-Spot Configuration. Part 3: Nonmonotonic 60. De Wit, A. and Homsy, G.M.: ‘‘Viscous Fingering in Periodically
Viscosity Profiles,’’ J. Fluid Mech. 共1999兲 388, 171. Heterogeneous Porous Media. II: Numerical Simulations,’’ J. Chem.
50. Koval, E.J.: ‘‘A Method for Predicting the Performance of Unstable Phys. 共1997兲 107, 9619.
Miscible Displacement in Heterogeneous Media,’’ SPEJ 共June 1963兲 61. Gelhar, L.W. and Axness, C.L.: ‘‘Three-Dimensional Stochastic
145; Trans., AIME, 228. Analysis of Macrodispersion in Aquifers.’’ Water Resour. Res. 共1983兲
51. Todd, M.R. and Longstaff, W.J.: ‘‘The Development, Testing, and 19, 161.
Application of a Numerical Simulator for Prediction of Miscible
Flood Performance,’’ SPEJ 共December 1972兲 874; Trans., AIME, Eckart Meiburg is a professor in the Dept. of Aerospace and
253.
Mechanical Engineering at the U. of Southern California in Los
52. Araktingi, U.G., and Orr, F.M. Jr.: ‘‘Viscous Fingering in Heteroge-
Angeles. e-mail: eckart@spock.usc.edu. His research interests
neous Porous Media,’’ SPE Advanced Technology Series 共April 1993兲
71. are in the area of computational fluid dynamics. Meiburg
53. Sorbie, K.S. et al.: ‘‘Flow Regimes in Miscible Displacements in holds Diploma and PhD degrees in mechanical engineering
Heterogeneous Correlated Random Fields,’’ paper SPE 24140 pre- from the U. of Karlsruhe, Germany. Ching-Yao Chen is an as-
sented at the 1992 SPE/DOE Symposium on Enhanced Oil Recovery, sistant professor in the Dept. of Mechanical Engineering at Da-
Tulsa, Oklahoma, 22–24 April. Yeh U. in Chang-Hwa, Taiwan. His research interests include
54. Waggoner, J.R., Castillo, J.L., and Lake, L.L.: ‘‘Simulation of EOR computational fluid dynamics and viscous flows. Chen holds a
Processes in Stochastically Generated Permeable Media,’’ SPEFE PhD degree in aerospace engineering from the U. of Southern
共June 1992兲 173; Trans., AIME, 293. California.

E. Meiburg and C.-Y. Chen: High-Accuracy Implicit Finite-Difference Simulations SPE Journal, Vol. 5, No. 2, June 2000 137

You might also like