Вы находитесь на странице: 1из 7

IMPES Stability: Selection of

Stable Timesteps
K.H. Coats, SPE, Coats Engineering, Inc.

Summary expense associated with the final results. Therefore, a Summary of


An IMPES stability criterion is derived for multidimensional Final Results section gives the final results, followed by sections
three-phase flow for black-oil and compositional models. The grid describing the derivations.
may be structured or unstructured. Tensor considerations are ne-
glected. The criterion can be used to set the time steps in an IMPES Summary of Final Results
formulation or as a switching criterion in an adaptive implicit model. For the unstructured grid case, the subscript i denotes a gridblock,
The criterion extends previous work by accounting for three- and the subscript j denotes one of its neighbors. Derivations using
phase flow, including capillary, gravity, and viscous forces, with a Cartesian grid use subscripts i, j, k as the gridblock indices in the
all the possible cocurrent and countercurrent flow configurations x, y, and z directions, respectively. In all equations throughout the
in a general grid. The criterion derivation uses stability theory to the paper, each phase mobility and its derivatives are evaluated at the
limits of its applicability, augmented by numerical experimentation, upstream block for the phase.
including extensive 1D tests and numerous field study datasets.
Maximum Stable Step Owing to Explicit kr and Pc. For three-
Introduction phase multidimensional flow, for each gridblock i, the maximum
A reservoir simulation model consists of N*N nonlinear difference stable timestep is limited by Condition 2 with

Fi = 1 Ⲑ 2 | f 11i + f 22i + 公共f 11i + f 22i兲2 − 4det共Fi兲 |, . . . . . . (3)


equations which express conservation of mass of Nc components in
each of N gridblocks comprising the reservoir. The form of each of
these equations, for a given block, is where each f is a sum of Ji terms, one for each of the Ji neighbors
of block i,j⳱1,2, . . .,Ji. For example,
MI,n+1 − MI,n = ⌬t 关sum (interblock flow rates)
Ji
− sum (well rates)], . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (1)
where MI,n⳱the mass of component I in the block at time level n.
f 11i = 兺 f 11 .
j=1
ij

The first sum is over all block neighbors; the second sum is over f 11ij = Tij 关共␭0 + ␭g兲 ␭ww
⬘ | ⌬⌽w | − ␭w␭ow
⬘ | ⌬⌽o|
all wells completed in the block. Well terms are assumed to be
treated fully implicitly and are dropped from consideration. − ␭w共␭p + ␭g兲 共Pcwoi ⬘ 兲兴 Ⲑ ␭t, . . . . . . . . . . . . . . . . . . . (4a)
⬘ + Pcwoj
The IMPES formulation1,2 treats the interblock flow rates im- f 12ij = −Tij 关␭w␭⬘og | ⌬⌽o | + ␭w␭⬘g | ⌬⌽g |
plicitly in pressure, but explicitly in saturations and compositions.
This explicit treatment gives rise to a conditional stability for − 共␭0 + ␭g兲 ␭⬘wg | ⌬⌽w |
IMPES, + ␭w␭g 共Pcgoi ⬘ 兲兴 Ⲑ ␭t, . . . . . . . . . . . . . . . . . . . . . . . . . (4b)
⬘ + Pcgoj
Fi⌬t f 21ij = −Tij 关␭g␭ww
⬘ | ⌬⌽w | ␭g␭ow
⬘ | ⌬⌽o|
⬍ 1, . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (2)
Vpi − ␭g␭w 共Pcgoi ⬘ 兲兴 Ⲑ ␭t, . . . . . . . . . . . . . . . . . . . . . . . . . (4c)
⬘ + Pcgoj
where ⌬t⳱maximum stable timestep and Fi⳱some function of f 22ij = Tij 关−␭g␭og
⬘ | ⌬⌽o | + 共␭w + ␭o兲 ␭⬘g | ⌬⌽g |
rates and reservoir and fluid properties. This paper derives the
function Fi for compositional and black oil models, accounting for − ␭g␭wg
⬘ | ⌬⌽w |
viscous, gravity, and capillary pressure forces in cocurrent and + ␭g 共␭o + ␭w兲 共Pcgoi ⬘ 兲兴 Ⲑ ␭t, . . . . . . . . . . . . . . . . . . . (4d)
⬘ + Pcgoj
countercurrent three-phase flow. The grid may be unstructured, or
structured (e.g., Cartesian) with or without nonneighbor connec- det共Fi兲 = f 11i f 22i − f 12i f 21i. . . . . . . . . . . . . . . . . . . . . . . . . . . (4e)
tions. The flow may be 1D, 2D, or 3D. Tensor considerations are The flow potential definitions are
neglected. Condition 2 gives a different stable step value for each
block. In an IMPES model, the time step used is the minimum of ⌬⌽w = ⌬p − ␥w⌬Z − ⌬Pcwo . . . . . . . . . . . . . . . . . . . . . . . . . . . . (5a)
all blocks’ stable step values. In an AIM (adaptive implicit) model,3
each block’s stable step size can be used to determine if the block ⌬⌽o = ⌬p − ␥o⌬Z . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (5b)
needs implicit treatment. The effects on stable step size of (a) inter- ⌬⌽g = ⌬p − ␥g⌬Z + ⌬Pcgo, . . . . . . . . . . . . . . . . . . . . . . . . . . . . (5c)
phase mass transfer and (b) the pressure and composition dependence
of fluid properties are assumed small and neglected. where ⌬X is Xj−Xi, X being any of p, Z, Pcwo, and Pcgo. For each
Two functions Fi are derived for use in Condition 2. The first neighbor j, the four frsij (rs⳱11,12,21,22) values are stored in the
relates to effects of explicit treatment of the saturation-dependent four frs arrays at the gas-phase upstream block position, if ␭g is not
terms (relative permeability and capillary pressure) in the inter- zero. If ␭g is 0, they are stored at the water phase upstream block
block flow rates. The second relates to the explicit treatment of position. The frs satisfy f11i>0, f22i>0, and det(Fi)>0, which pro-
compositions in the interblock flow rates. vide a check on coding errors.
Derivations of the function Fi are lengthy and, at various The Fi given by Eq. 3 accounts for viscous, gravity, and cap-
points, tedious. This tends to obscure the simplicity and low cpu illary forces, and for all possible cocurrent and countercurrent
flows of three phases between two blocks. The above equations
collapse to simple forms for each of the three two-phase cases. For
example, for 1D two-phase gas/oil flow, ␭w⳱␭wg ⬘ ⳱␭ow⬘ ⳱␭ww ⬘ ⳱0,
Copyright © 2003 Society of Petroleum Engineers which give f11⳱f12⳱f21⳱0 and
This paper (SPE 84924) was revised for publication from paper SPE 69225, first presented
at the 2001 SPE Reservoir Simulation Symposium, Houston, 11–14 February. Original F = f 22 = T 关␭0␭g⬘ | ⌬⌽g | − ␭g␭⬘og | ⌬⌽o|
manuscript received for review 23 March 2001. Revised manuscript received 20 February
2003. Manuscript peer approved 10 March 2003. + ␭g␭o 共Pcgoi ⬘ 兲兴 Ⲑ 共␭o␭g兲. . . . . . . . . . . . . . . . . . . . . . . . . . (6)
⬘ + Pcgoj

June 2003 SPE Journal 181


If Pcgo and gravity are neglected, this further reduces to F=qfg⬘, coefficients, mathematicians long ago suggested applying the
giving the well known result analysis using constant coefficients equal to local values and then
applying the resulting analysis criterion to each subregion sepa-
qfg⬘⌬t
⬍ 1, . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (7) rately.5 That procedure is followed here, with the local region
Vp being a single block-pair interface.
where fg⳱fractional flow ␭g/(␭o+␭g).
The coding requirements associated with Eq. 3 are low. All The Two-Phase Flow Case. For the case of 1D incompressible
terms in Eqs. 3 through 5, except the derivatives of phase mobili- gas/oil flow, the Darcy expressions for oil and gas flow rates from
ties and capillary pressures, exist in the interblock flow subroutine gridblock i to block i+1 are
of an IMPES model. The above results do not depend upon choice qoi = −Ti+1 Ⲑ 2␭o共⌬p − ␥o⌬Z兲 . . . . . . . . . . . . . . . . . . . . . . . . . . . (15a)
of IMPES variables or upon whether variable substitution is used.
qgi = −Ti+1 Ⲑ 2␭g共⌬p − ␥g⌬Z + ⌬Pcgo兲, . . . . . . . . . . . . . . . . . . . (15b)
Maximum Stable Step Owing to Explicit Compositions. The where ⌬X for any quantity X is Xi+1−Xi. Eliminating ⌬p from these
explicit treatment of compositions gives Condition 2 with two equations gives gas and oil phase flow rates as
Qo␳oxI + Qg␳gyI qgi = ␭g Ⲑ ␭t关q − Ti+1 Ⲑ 2␭o共g + ⌬Pcgo兲兴 . . . . . . . . . . . . . . . . . . . . (16a)
Fi = Max共I兲 , . . . . . . . . . . . . . . . . . . . . . . . . . . . . (8)
So␳oxI + Sg␳gyI
qoi = ␭p Ⲑ ␭t关q + Ti+1 Ⲑ 2␭g共g + ⌬Pcgo兲兴, . . . . . . . . . . . . . . . . . . . . (16b)
where I⳱1,2,. . .,Nc. The rate Qo is the sum of all oil outflow rates
where g is the gravitational term (␭o−␭g) ⌬Z, q is qoi+qgi, and ␭t
from block i to its neighbors; Qg is similarly defined. The maxi-
is ␭o+␭g. Each of qgi, qoi, q, g, and ⌬Pcgo may be positive or
mum stable step is the lesser of the two values given by Condition
negative.
2 when Fi is substituted from Eqs. 3 and 8. The Fi of Eq. 8 is
The equation expressing conservation of mass of gas in grid-
simple and inexpensive in cpu. The rates Qo and Qg are stored in
block i is
arrays in the interblock flow subroutine.
Vpi
A Comparison With Previous Work 共S − Sgi,n兲 = qgi−1共Sgi−1,n,Sgi,n兲
⌬t gi,n+1
Previous papers referenced elsewhere4 give an IMPES stability
condition for multidimensional flow in a Cartesian grid. For cocur- −qgi共Sgi,n,Sgi+1,n兲, . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (17)
rent, hyperbolic, 1D gas/oil flow, that condition is which emphasizes that qgi is a single-valued function of Sgi and
qfg⬘ ⌬t Ⲑ Vpi ⬍ 1. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (9) Sgi+1, depending upon the oil and gas flow rate directions. The
right side interblock flow terms are explicitly dated in saturation in
Previous work is unclear regarding use of this condition when the accordance with the IMPES treatment.
flow is countercurrent. For 1D countercurrent flow in a vertical Stability analysis requires a constant-coefficient, linear differ-
column with uniform grid spacing ⌬z, ence equation which is, in some sense, an approximation of Eq. 17.
qfg⬘ = −Tz共␥o − ␥g兲⌬z⌿⬘. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (10) That equation is obtained by writing Eq. 17 twice, once as shown
where the Sgi are the actual calculated values and again with the Sgi
Substituting this in Condition 9 gives values replaced by Sgi * which is the exact (error-free) solution.
−Tz共␥o − ␥g兲⌬Z⌿⬘⌬t Ⲑ Vpi ⬍ 1, . . . . . . . . . . . . . . . . . . . . . . . . . (11) Subtracting the resulting two equations gives
or, expanding the derivative ⌿⬘, Vpi
共␧ − ␧i,n兲 = ␦qgi−1 − ␦qgi, . . . . . . . . . . . . . . . . . . . . . . . . . (18)
⌬t ⌬t i,n+1
− T 共␥ − ␥g兲⌬⌮共␭2o␭g⬘ + ␭2g␭og
⬘ 兲 Ⲑ ␭t2 ⬍ 1. . . . . . . . . . . . . . . (12)
Vpi z o where ␧i,n⳱the error Sgi,n−S*
gi,n and

The derivative ⌿⬘ is positive for low to moderate Sg and negative ␦qgi = qgi共Sgi,n,Sgi−1,n兲 − qgi共Sgi,n
⬘ ,Sgi+1,n
⬘ 兲. . . . . . . . . . . . . . . . . . . (19)
for larger Sg. Condition 11 presents a dilemma because it gives
Using the first Taylor series terms gives
unconditional stability and conditional stability for positive and
negative ⌿⬘, respectively. For this counter-current flow, Eq. 3 ␦qgi = qgi,i
⬘ ␧i,n + qgi,i+1
⬘ ␧i+1,n, . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (20)
gives Fi=f22i and Condition 2 is
where q⬘gi,i⳱the partial derivative of qgi with respect to Sgi and
⌬t q⬘gi,i+1⳱the partial derivative of qgi with respect to Sgi+1. The term
T 共␥ − ␥g兲⌬Z共␥2o␥g⬘ − ␥2g␥og
⬘ 兲 Ⲑ ␥t2 ⬍ 1. . . . . . . . . . . . . . . . . (13) ␦qgi-l in Eq. 18 is expressed in similar fashion, with care to use
Vpi z o
locally constant derivatives (i.e., q⬘gi−l,i-l⳱q⬘gi,i and q⬘gi−l,i⳱q⬘gi,i+1),
Numerous 1D vertical countercurrent flow simulations were
performed, initializing the column with uniform saturation distri- ␦qgi−1 = qgi,i
⬘ ␧i−1,n + qgi,i+1
⬘ ␧i,n. . . . . . . . . . . . . . . . . . . . . . . . . . . . (21)
butions or with the upper half containing the heavier phase and the Substituting Eqs. 20 and 21 into Eq. 18 gives the error equation
lower half containing the lighter phase. In all cases, use of ⌬t from
Condition 13, using ⳱ in place of <, gave nonoscillatory stability. Vpi
In most cases, instability occurred for ⌬t values 10 to 20% larger. 共␧ − ␧i,n兲 = ai␧i+1,n − bi␧i,n + ci␧i−1,n, . . . . . . . . . . . . . . . . (22)
⌬t i,n+1
In all cases, results exhibited significant instability using ⌬t from
Condition 11 with the conservative choice of |⌿⬘| replacing −⌿⬘. where
Conditions 12 and 13 give the data-dependent ratio ai = −qgi,i+1
⬘ . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (23a)
⌬t共this work) +Max共k兲共|␭2o␭g⬘ ␭2g␭og
⬘ |兲 bi = qgi,i
⬘ − qgi,i+1
⬘ . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (23b)
= ⬍ 1, . . . . . . . . (14)
⌬t(previous work) Max共k兲共␭2o␭g⬘ − ␭2g␭⬘og兲 ci = qgi,i
⬘ . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (23c)
where Max(k) denotes maximum over all gridblocks at the current Eq. 23 shows that ai+ci⳱bi. Also, q⬘gi,iⱖ0 and q⬘gi+1ⱕ0 for all cases
timestep. Numerous similar simulations with three-phase flow gave of cocurrent and countercurrent flow, so that each of ai,bi,ciⱖ0.
stability using Eq. 3 and instability with ⌬t values 20% larger. The Appendix describes application of stability analysis to Eq.
22 with the variable coefficients Vpi, ai, bi, ci assumed to be con-
Derivation of F for Effects of Explicit Treatment stants independent of i and n. The resulting stability condition is
of Saturations
Methods of stability analysis generally apply to linear difference ⌬t
共a + bi + ci兲 ⱕ 2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (24)
equations with constant coefficients. For the case of variable Vpi i

182 June 2003 SPE Journal


ai + ci ⱕ bi, . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (25) The Appendix describes application of the stability analysis to
these two simultaneous equations to obtain Condition 2 and Eq. 3.
and using Eqs. 23, the stability condition is
⌬t Derivation of F for Effects of
共q ⬘ − qgi,i+1
⬘ 兲 ⬍ 1. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (26) Explicit Compositions
Vpi gi,i
This section ignores the effects of explicit treatment of saturation-
The terms q⬘gi,i and q⬘gi,i+1 are obtained by differentiating Eq. dependent terms on stable step size. For the assumptions men-
16a with respect to Sgi and Sgi+l. It might appear that different tioned in the Introduction, the mass conservation equations in dif-
expressions for q⬘gi,i−q⬘gi,i+1 would result for the four possible ferential form for 1D flow are
cocurrent and countercurrent flow configurations. However, that is
not the case. For all four cases, ⭸ ⭸
共S ␳ x + Sg␳gyI兲 = − 共qo␳oxI + qg␳gyI兲,
⭸t o o I ⭸x
Ti+1 Ⲑ 2␭o␭g
⬘ − qgi,i+1
qgi,i ⬘ = 共P⬘cgoi + Pcgoi+1
⬘ 兲 I = 1, 2, . . , Nc . . . . . . . . . . . . . . . . . . . . . . . (34a)
␭t
␭o ␭g ⭸ ⭸
+ |q | ␭ ⬘ − |q | ␭⬘ . . . . . . . . . . . . . . . . . . . . . . (27) 共S ␳ 兲 = − 共qo␳o兲, . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (34b)
␭t␭g gi g ␭t␭o oi og ⭸t o o ⭸x

Note that the term q⬘gi,i and q⬘gi,i+1 is always positive. The last two ⭸ ⭸
共Sg␳g兲 = − 共qg␳g兲. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (34c)
(non-Pc) terms in Eq. 27 can be expressed equivalently as ⭸t ⭸x
Ti+1 Ⲑ 2 If qo, these equations give
关␭o␭g⬘|⌬⌽g|− ␭⬘g␭og
⬘ |⌬⌽o |兴. . . . . . . . . . . . . . . . . . . . . . . . . (28)
␭t ⭸y␫ qg␳gyI ⭸yI
+ = 0, . . . . . . . . . . . . . . . . . . . . . . . . . . (35a)
These results (Eqs. 26 through 28) for 1D gas/oil flow are equivalent ⭸t So␳oxI + Sg␳gyI ⭸x
to Condition 2 with Fi⳱f22i in Eq. 3. For multidimensional flow and if qg⳱0,
there are more contributions to Fi, each of the same form as f22i.
⭸xI q0␳oxI ⭸xI
The Three-Phase Flow Case. For the case of 1D three-phase + = 0. . . . . . . . . . . . . . . . . . . . . . . . . . . (35b)
⭸t So␳oxI + Sg␳gyI ⭸x
incompressible flow, the Darcy expressions for phase rates from
gridblock i to block i+1 are Eq. 35 is obtained from Eqs. 34 using yI⳱KIxI and neglecting the
dependence of the K-values on pressure and composition. Eq. 35
qwi = Ti+1 Ⲑ 2␭w共⌬p − ␥w⌬Z − ⌬Pcwo兲 . . . . . . . . . . . . . . . . . . . . (29a) is easily expressed in difference form, with explicit dating of the
qoi = Ti+1 Ⲑ 2␭o共⌬p − ␥o⌬Z兲 . . . . . . . . . . . . . . . . . . . . . . . . . . . . (29b) spatial derivatives and upstream weighting. Application of the von
Neumann stability analysis, as in the Appendix, gives stability
qgi = Ti+1 Ⲑ 2␭g共⌬p − ␥g⌬Z − ⌬Pcgo兲 . . . . . . . . . . . . . . . . . . . . . (29c) conditions
Eliminating ⌬p from these equations gives the phase rates as qg␳gyI ⌬t
⬍ 共qo = 0兲 . . . . . . . . . . . . . . . . . . . . . . . . . . (36)
qwi = ␭w关q + 共␭o + ␭g兲 awo + ␭gago兴 Ⲑ ␭t . . . . . . . . . . . . . . . . . . (30a) So␳oxI + Sg␳gyI Vpi

qgi = ␭g关q + 共␭o + ␭w兲 ago + ␭wawo兴 Ⲑ ␭t . . . . . . . . . . . . . . . . . . (30b) qo␳oxI ⌬t


⬍ 1 共qg = 0兲. . . . . . . . . . . . . . . . . . . . . . . . . (37)
So␳oxI + Sg␳gyI Vpi
qoi = ␭o关q + ␭w awo + ␭gago兴 Ⲑ ␭t, . . . . . . . . . . . . . . . . . . . . . . . . (30c)
If So⳱0 or Sg⳱0, these further reduce to the well known stability
where q⳱total rate qo+qw,qg, ␭t⳱total mobility ␭w+␭o+␭g, and conditions for multicomponent miscible flow,
awo = Ti+1 Ⲑ 2关共␥w − ␥o兲⌬Z + ⌬Pcwo兴 . . . . . . . . . . . . . . . . . . . . . (31a) qg⌬t
⬍ 1 共So = 0兲 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (38)
ago = Ti+1 Ⲑ 2关共␥o − ␥g兲⌬Z + ⌬Pcgo兴. . . . . . . . . . . . . . . . . . . . . . (31b) SgVpi
Because the fluids are assumed incompressible and saturations qo⌬t
add to 1.0, q is a constant independent of i and there are only two ⬍ 1 共Sg = 0兲. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (39)
SoVpi
independent conservation equations. Selecting the water and gas
equations, The four stability conditions just given for various choices of
rates and saturations are all included in the general stability con-
Vpi dition, written for the general case of multidimensional flow with
共S − Swi,n兲 = qwi−1,n − qwi,n . . . . . . . . . . . . . . . . . . . . . . . (32a)
⌬t wi,n+1 nonneighbors,
Vpi Qo␳oxI + Qg␳gyI ⌬t
共S − Sgi,n兲 = qgi−1,n − qgi,n . . . . . . . . . . . . . . . . . . . . . . . (32b) Max共I兲 ⬍ 1. . . . . . . . . . . . . . . . . . . . . . . . . (40)
⌬t gi,n+1 So␳oxI + Sg␳gyI Vpi
Following the same procedure as in the two phase case (Eqs. 17 Qo⳱the sum of all oil outflow rates from block i to its neighbors,
through 22), we obtain two equations in the two errors ␧1⳱Sw−S*
w
RB/D. The rate Qg is similarly defined. Condition 40 simply states
and ␧2⳱Sg−S* g,
that, for each component, the mols flowing out of a block in a time
step cannot exceed the mols in place in the block. All saturations,
Vpi densities, and compositions in Condition 40 are dated explicitly at
共␧1 − ␧1i,n兲 = −qwwi,i+1
⬘ ␧1i+1,n − 共qwwi,i
⬘ − qwwi,i+1
⬘ 兲 ␧1i,n
⌬t i,n+1 time n, while all rates Q are dated at time n+1.
The author knows of no theoretical basis for Condition 40 in
+ qwwi,i
⬘ ␧1i−1,n − qwgi,i+1
⬘ ␧2i+1,n − 共qwgi,i
⬘ − qwgi,i+1
⬘ 兲 ␧2i,n
the case where both Qo and Qg are nonzero. Young and Russell6
+ qwgi,i
⬘ ␧2i−1,n . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (33a) derived a different, approximate stability condition for those con-
ditions. Numerous 1D simulations were performed generating
Vpi steady-state (uniform saturation) distributions corresponding to in-
共␧2 − ␧2i,n兲 = −qgwi,i+1
⬘ ␧1i+1,n − 共qgwi,i
⬘ − qgwi,i+1
⬘ 兲 ␧1i,n
⌬t i,n+1 jection of different fixed-proportion gas/oil/water feed streams.
The proportions were such that the throughput Eq. 8 dominated
+ qgwi,i
⬘ ␧1i−1,n − qggi,i+1
⬘ ␧2i+1,n
(i.e., gave smaller ⌬t than) the displacement in Eq. 3. In all cases,
− 共qggi,i
⬘ − qggi,i+1
⬘ 兲 ␧2i,n + qggi,i⬘ ␧2i−1,n. . . . . . . . . . . . (33b) nonoscillatory stability occurred with ⌬t given by Condition 40

June 2003 SPE Journal 183


(using⳱in place of <). In cases of only one mobile hydrocarbon (GOR) are to be avoided. These results exhibit the stable flow that
phase, instability arose for ⌬t values 20% larger. With mobile oil can occur with uniform CFLi and CFL considerably above 1. In
and gas phases, instability occurred in most cases at ⌬t values 20% numerical tests of stability, perturbations should be introduced
larger, but in some cases did not occur until ⌬t was 60% larger. before concluding that a uniform CFLi, CFL > 1 flow regime is
“stable.”
Testing and Discussion Problem 4 includes capillary pressure in a 1D 20-block hori-
A large number of 1D test problems and numerous 3D field study zontal grid with initial Swi⳱0.2. The diffusivity or capillary term
data sets were run to test stability as a function of the CFL limit. is k(P cwo ) max /uL ␮ w ⳱13.5 with P cwo ⳱(P cwo ) max [(1−S w )/
The term CFL is Max(i)CFLi where CFLi is Fi⌬t/Vpi. The tests (1−Swc)]3. Water breakthrough occurred at about 0.24 pore vol-
included black oil and compositional, two-phase and three-phase umes water injected. Instability at CFL⳱1.1 gave fractional water-
cases, with cocurrent and countercurrent flow. Several of these cut oscillations with an amplitude of about 0.04 when the water cut
tests are described below. was 0.4.
With one exception, CFL⳱1 gave nonoscillatory stability in all Three-phase Problem 5 simulates gas injection in a 1D hori-
cases. The CFL limit for stability ranged from 1 to 2, depending zontal 20-block grid with initial Swi⳱0.4 and Soi⳱0.6. Gas vis-
upon the problem-dependent uniformity of the CFLi values. The cosity ␮ g ⳱0.05 cp, S o r g ⳱0.2, S g c ⳱0, and k r g ⳱[S g /
stable CFL limit was close to 1.0 for near-uniform CFLi and close (1−Sorg−Swc)]3. Gas, oil, and water are incompressible, and there
to 2.0 for the Buckley-Leverett, variable CFLi problem.4 As cap- is no dissolved gas or oil content of the gas phase. After gas
illary pressure becomes more dominant in the Fi of Condition 2, breakthrough at 0.177 pore volumes injected, ⌬t was controlled
the CFLi distribution becomes more uniform, and the CFL limit each timestep with one loop to exactly satisfy the specified CFL
tends toward 1.0, as shown by Problem 4 below. Tests of the value. Runs with CFL values up to 1.98 exhibited stability with no
heuristic Eq. 8, which perturbed uniform 1D, compositional gas/oil water-cut or GOR oscillations. Using CFL⳱2.2 gave large GOR
flowing mixtures, showed CFL limits ranging from 1.1 to 1.6. oscillations and significant water-cut oscillations, similar to the
The linear solver step of an IMPES Newton iteration gives a two-phase Problem 1 results.
new pressure distribution which may result in new upstream Problem 6 is the same as Problem 5 except a water/oil/gas
blocks for each phase. If more than one Newton iteration is taken, mixture containing equal parts by volume of each phase was in-
the model recalculates transmissibilities based on those new up- jected at constant total rate to establish uniform saturations
stream blocks. In addition, loops may be performed within a New- throughout the 20 blocks. Runs from that point were then made
ton when stable step logic is used. After the linear solver is called, using the 1-loop option so that the ⌬tof each timestep exactly
flow rates and stable step size are calculated using the upstream satisfied the specified CFL value. With no perturbations of the
blocks and interblock rates corresponding to the new pressure injection mixture, a CFL of 1.5 gave large water-cut and GOR
distribution. If that stable step is less than the timestep being used, oscillations that persisted indefinitely. A CFL of 1.2 gave stable
the model recalculates transmissibilities using the new pressure behavior—no change in water-cut or GOR. However, similar to
distribution, resets ⌬t to the stable step value, and returns to the the two-phase Problem 3, for CFL⳱1.2 perturbations in the in-
solver. For most problems, the average number of Newtons per jection mixture resulted in unstable travel of the perturbations and
timestep is close to 1.0, and the average number of loops per large GOR oscillations after perturbation breakthrough, followed
Newton is close to 0. by a return to stability.
An option used in the tests discussed below is the forcing of Problem 7 is Young and Russell’s 1/4 five-spot waterflood
one loop with ⌬t reset to stable step size regardless of whether the problem. The block-centered 13×13×1 grid had uniform spacing
stable step is smaller or larger than the timestep being used. This with half-edge and quarter-corner blocks. That is, the volume of
results in the CFL value of each timestep exactly equalling the each interior block was four times that of each corner well block
value cited in the discussion. Unless noted otherwise, the runs and twice that of each other edge block. Wells are located at grid
described here used no constraints other than the stable step con- points. The timestep was controlled by changes (⌬S max ,
dition. That is, no change constraints or maximum timestep con- ⌬xmax)⳱(.1, .1) and the specified CFL limit. The timestep was
straints were used. limited by the CFL limit alone after water breakthrough at 0.339
Problems 1 through 6 use the water/oil mobility data of Young pore volumes injected. The 1-loop option was activated at that
and Russell’s first two problems.6 Problem 1 is their 20-block 1D point, and runs were made with various CFL limits. As the CFL
vertical water/oil displacement with an implicit producer in Block limit was increased above 1.88, the water cut oscillated, with an
1. After water breakthrough, the CFL occurred in Block 2. For a amplitude increasing with increasing CFL. The run to 0.4 pore
CFL of 2.0, the after-breakthrough behavior was stable. For volumes injected, using CFL⳱1.88 (without the 1-loop option)
CFL⳱2.1 and larger, Sw oscillated in Block 2 and significant took 93 steps, 94 Newtons, and 0.44 seconds cpu on a 400 MHz
oscillations in oil production rate occurred. An explanation for this Gateway PC with the Compaq Visual Fortran Compiler version 6.0.
stability for CFL values up to 2.0 is given elsewhere.4 Identical The SPE Comparative Solution Project Problems 1, 3, and 10
results were obtained when the 20-block column was horizontal. are denoted here as Problems SPE1, SPE3, and SPE10. Problem
Problem 2 simulates countercurrent flow in a 20-block, 100-ft, SPE1 was run using the above changes and a CFL limit of 1.5. All
closed vertical column with k(␥o−␥g)⳱274 md-psi/ft. Runs (a) steps after gas breakthrough at about 1,230 days were controlled
were made for initial Swi⳱0.45, and runs (b) for initial Swi⳱1.0 in by the CFL limit alone, with Fi from the displacement Eq. 3. The
the upper half and 0.2 in the lower half of the column. The stable run took 145 steps, 154 Newtons, 1.42 seconds cpu and exhibited
CFL limit was 1.6 for runs (a) and 1.2 for runs (b). In runs (a), no GOR oscillations. A GOR reversal at 1,300 days is real. Un-
CFLi decreased significantly with distance from the CFL block. stable GOR behavior resulted for CFL>1.5. The stable step of
Runs (b) exhibited more uniform CFLi distributions. Condition 2 increased from 8 days at 1,250 days to 32 days at
Problem 3 illustrates stability behavior when the CFLi values 2,200 days and varied by less than ±10% from an average 31 days
are uniform throughout the grid. Water and oil were co-injected to value thereafter. This stable step vs. time profile differs consider-
establish uniform saturations through a 1D 20-block horizontal ably from that reported by Young and Russell for this SPE1 prob-
grid. The injection mix was then perturbed from 50% (by volume) lem. The controlling cell was generally a neighbor of the produc-
water to 45% water. After the new steady state, the injection mix ing block.
was again perturbed, from 45 to 55% water. For CFL⳱1.2, the The SPE3 9-component compositional problem was stable at a
results were stable with no perturbations. However, the perturba- CFL⳱2.0, requiring 105 steps, 106 Newtons, and 5.13 seconds
tions traveled unstably through the grid with the amplitude of Sw cpu. All timesteps were controlled by the CFL limit alone (Con-
oscillations at the perturbation front increasing with distance trav- dition 2 with <1 replaced by <2), and with Fi given by the through-
eled. Large water-cut oscillations occurred after perturbation put Eq. 8. The displacement Eq. 3 Fi limited none of the steps. The
breakthrough, followed by a return to stability. This behavior is not stable step of Condition 2 declined from 32 to 24.5 days in the first
tolerable if significant oscillations in water-cut and/or gas/oil ratio 10 years, and declined from 44 to 18.5 days in the last 5 years.

184 June 2003 SPE Journal


Compared with Young and Russell, the last 5-year values are about q⬘wgi,i+1 ⳱ ⭸qwi/⭸Sgi+1
the same, but the first 10-year values here are approximately twice q⬘wwi,i ⳱ ⭸qwi/⭸Swi
as large. Instabilities occurred for CFL⳱2.2. q⬘wwi,i+1 ⳱ ⭸qwi/⭸Swi+1
Gas/oil Problem SPE10 simulates gas injection in a 100×1×20 S ⳱ saturation, fraction
grid with a geostatistical permeability distribution. The immiscible Sgc ⳱ critical gas saturation
gas and oil are incompressible, and there is no solution gas or
Sorg ⳱ residual oil saturation to gas
interphase mass transfer. Implicit is faster than IMPES for this
problem, running in 718 steps, 816 Newtons, and 214 cpu seconds. Swc ⳱ connate water saturation
The IMPES run using CFL⳱2.0 was stable and required 7,018 t ⳱ time
steps, 7,059 Newtons, and 558 cpu seconds. Both runs gave nearly T ⳱ transmissibility, kA/l, RB-cp/day-psi
identical time plots of rates and GOR, contradicting, in this case, Ti+1/2 ⳱ transmissibility for flow between gridblocks i and
the common belief that implicit results are less accurate than i+1
IMPES results because of larger numerical dispersion error. An Tij ⳱ transmissibility connecting block i and neighbor j
IMPES run using CFL⳱2.2 gave rate and GOR vs. time plots that u ⳱ superficial velocity
were smooth and identical to those for CFL⳱2.0. Nevertheless, Vp ⳱ gridblock pore volume
the run was unstable, reflected in the 8,494 steps, 8,535 Newtons, x, y, z ⳱ Cartesian coordinates
and 849 cpu seconds. The instabilities were saturation oscillations
xI ⳱ mol fraction of component I in the oil phase
at blocks not near the producer.
yI ⳱ mol fraction of component I in the gas phase
Conclusions Z ⳱ depth, measured vertically downward
An IMPES stability criterion is derived for the three-phase, cocur- ␤, ␤1, ␤2 ⳱ eigenvalues of von Neumann stability analysis,
rent and countercurrrent, multidimensional flow occurring in black Eqs. A-2, A-12
oil and compositional models. It accounts for viscous, gravity, and ⌬x, ⌬y, ⌬z ⳱ gridblock dimensions, length
capillary forces in structured or unstructured grids and gives a ⌬Smax ⳱ change over timestep of saturation, maximum
stable step size for each gridblock. These stable steps can be used over grid
as switching criteria in an adaptive implicit model or to set the ⌬t ⳱ timestep, tn+1−tn
timestep size in an IMPES formulation. ⌬xmax ⳱ change over timestep of mol fraction, maximum
The criterion is Fi⌬t/Vpi<1. We define the terms CFLi⳱Fi⌬t/
over grid
Vpi and CFL⳱Max(i)CFLi. A large number of 1D two- and three-
phase numerical tests and numerous field study data sets were run ␥ ⳱ phase density or gradient, psi/ft
using IMPES to check the criterion. Nonoscillatory stability re- ␧ ⳱ error in saturation
sulted for timesteps obeying CFL⳱1. Instability (oscillations) oc- ␭ ⳱ error amplification factor
currred at CFL⳱1.1 or 1.2 when the CFLi values were nearly ␭1 ⳱ error amplification factor for water saturation
uniform through the grid. Nonoscillatory stability occurred for error
CFL values between 1.2 and 2.0 as the (problem-dependent) CFLi ␭2 ⳱ error amplification factor for gas saturation error
distribution became more nonuniform. All runs with CFL>2.0 ex- ␭g ⳱ gas-phase mobility, krg/␮g, dependent upon Sg
hibited oscillations. ␭⬘g ⳱ d␭g/dSg
␭o ⳱ oil-phase mobility, kro/␮o, dependent upon Sw
Nomenclature
and Sg
fg ⳱ fractional flow of gas in a 1D gas/oil system,
␭⬘og ⳱ ⭸␭o/⭸Sg
␭g(1−gT ␭o/q)/␭t
␭⬘ow ⳱ ⭸␭o/⭸Sw
f ⬘g ⳱ dfg/dSg
␭t ⳱ total mobility, e.g. ␭w+␭o+␭g in a three-phase
F ⳱ stability function
system
g ⳱ (␭o−␭g)(Zi+1−Zi)
␭w ⳱ water-phase mobility, krw/␮w, dependent upon Sw
ı̂ ⳱ √−1
and Sg
k ⳱ permeability
␭⬘wg ⳱ ⭸␭w/⭸Sg
kr ⳱ relative permeability, fraction
␭⬘ww ⳱ ⭸␭w/⭸Sw
KI ⳱ component I K-value, yI/xI
␮ ⳱ viscosity
L ⳱ length of 1D column
␳g ⳱ molar density of gas phase, mols/rb
MI,n ⳱ mass of component I in gridblock at time level n
␳o ⳱ molar density of oil phase, mols/rb
N ⳱ number of gridblocks
␾ ⳱ phase flow potential
Nc ⳱ number of nonaqueous components
⌿ ⳱ ␭o␭g/(␭o+␭g), dependent upon Sg
p ⳱ oil-phase pressure
⌿⬘ ⳱ d⌿/dSg
Pcgo ⳱ gas/oil capillary pressure, pg−po, dependent upon Sg
P⬘cgo ⳱ dPcgo/dSg Subscripts
Pcwo ⳱ water/oil capillary pressure, po−pw, dependent g ⳱ gas phase
upon Sw i, j ⳱ gridblock indices
P⬘cwo ⳱ dPcwo/dSw i, j, k ⳱ gridblock indices in the x, y, and z directions,
q ⳱ interblock total flow rate, RB/D respectively
Qg ⳱ sum of gas outflow rates from a gridblock, RB/D I ⳱ nonaqueous component number, I⳱1, 2, …, Nc
Qo ⳱ sum of oil outflow rates from a gridblock, RB/D n ⳱ time level
qw, qo, qg ⳱ phase interblock flow rates o ⳱ oil phase
qgi ⳱ gas-phase flow rate from block i to block i+1 w ⳱ water phase
q⬘ggi,i ⳱ ⭸qgi/⭸Sgi x, y, z ⳱ Cartesian directions x, y, and z
q⬘ggi,i+1 ⳱ ⭸qgi/⭸Sgi+1
q⬘gwi,i ⳱ ⭸qgi/⭸Swi
q⬘gwi,i+1 ⳱ ⭸qgi/⭸Swi+1 Acknowledgments
qwi ⳱ water-phase flow rate from block i to block i+1 The author is indebted to Diane Korpics for transcribing the docu-
q⬘wgi,i ⳱ ⭸qwi/⭸Sgi ment.

June 2003 SPE Journal 185


References Substituting ␧ijkn⳱␭neı̂i␤x+ı̂k␤z leads to the amplificaction factor
1. Stone, H.L. and Garder, Jr., A.O.: “Analysis of Gas-Cap or Dissolved-
Gas Drive Reservoirs,” Trans., AIME (1961) 222, 92.
|␭|2 = 关1+ 共ax + cx兲cos␤x − bx + 共ay + cy兲cos␤y − by
2. Coats, K.H.: “A Note on IMPES and Some IMPES-Based Simulation + 共az + cz兲cos␤z − bz兴2
Models,” SPEJ (September 2000) 245.
3. Thomas, G.W. and Thurnau, D.H.: “Reservoir Simulation Using an + 共ax − cx兲2sin␤z2 + 共ay − cy兲2sin␤y2
Adaptive Implicit Method,” SPEJ (October 1983) 759.
+ 共az − cz兲2sin␤z2. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (A-8)
4. Coats, K.H.: “IMPES Stability: The CFL Limit,” paper SPE 66345
presented at the 2001 SPE Reservoir Simulation Symposium, Houston, The maximum value of |␭| occurs for cos␤x⳱cos␤y⳱cos␤z⳱−1
11–14 February. and the condition that |␭| be < 1 gives (provided a+c<b for each
5. Hildebrand, F.B.: Methods of Applied Mathematics, Prentice-Hall, Inc., direction)
Englewoods Cliffs, New Jersey (1952).
6. Young, L.C. and Russell, T.F.: “Implementation of an Adaptive Im-
ax + bx + cx + ay + by + cy + az + bz + cz ⬍ 2, . . . . . . . . . . . . (A-9)
plicit Method,” paper SPE 25245 presented at the 1993 SPE Sympo- which is simply the 1D result Condition A-5a with an additional
sium on Reservoir Simulation, New Orleans, 28 February–3 March. term of identical form for each additional direction.
7. Richtmyer, R.D.: Difference Methods for Initial-Value Problems, In-
terscience Publishers Inc., New York City (1957). The Three-Phase Case. The three-phase case involves the two
error Eqs. 33a and 33b which are of the form
Appendix—Stability Analysis
␧1i, n+1 − ␧1i, n = a11␧1i+1, n − b11␧1i, n + c11␧1i−1, n
The method of stability analysis used here was developed by J. von
Neumann and is described by many authors, including Richtmyer.7 + a12␧2i+1, n − b12␧2i, n + c12␧2i−1, n . . . . . . . . . . . (A-10a)

The Two-Phase Case. We consider the difference equation ␧2i+1, n − ␧2i, n = a21␧1i+1, n − b21␧1i, n + c21␧1i−1, n

␧i, n+1 − ␧i, n = ␣␧i+1, n − b␧i, n + c␧i−1, n , . . . . . . . . . . . . . . . . . . (A-1) + a22␧2i+1, n − b22␧2i, n + c22␧2i−1, n . . . . . . . . . . . (A-10b)
Comparing the coefficients in Eqs. 33 with those in Eq. A-10
similar to Eq. 22, where all coefficients a,b,c are ⱖ0. The von
shows that the latter satisfy
Neumann stability analysis method replaces ␧i,n by the Fourier
type term ars + crs = brs rs = 11, 12, 21, 22. . . . . . . . . . . . . . . . . . (A-11)
n ˆii␤
␧i, n = ␭ e . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (A-2) Substituting ␧1i,n⳱␭n1eı̂j␤1 and ␧2i,n⳱␭n2eı̂i␤2 in these equations
gives
The ratio ␧i,n+1/␧i,n is ␭ so that the stability condition is |␭|<1.
Substituting ␧i,n from Eq. A-2 into Eq. A-1 and using the identity ␧n+1 = E␧n, . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (A-12)
eı̂␤⳱cos␤+ı̂sin␤ gives where ␧n⳱the error vector (␧1i,n,␧2i,n)T and the elements ␧rs of
the amplification matrix E are
␭ = 1 − b + 共a + c兲cos␤ + î共a − c兲sin␤ . . . . . . . . . . . . . . . . . (A-3)
␧11 = 1 − d11 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (A-13a)
and
␧12 = −d12 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (A-13b)
|␭|2 = 关1 − b + 共a + c兲cos␤兴2 + 关共a − c兲sin␤兴2. . . . . . . . . . . . (A-4)
␧21 = −d21 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (A-13c)
We seek the maximum value of |␭|2 over the range of eigen-
values, or equivalently, the range of cos␤ values, −1<cos␤<1. The ␧22 = 1 − d22, . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (A-13d)
maximum value of |␭|2 occurs at a ␤ value which satisfies d|␭|2/
d␤⳱0. Taking this derivative from Eq. A-4, setting it to zero, and and, for rs⳱11,12,21,22,
solving for ␤ gives two roots. Using the first root, sin␤⳱0, in Eq.
A-3, we find that stability (|␭|<1) requires the two conditions (for drs = −共ars + crs兲cos␤x + brs − î共ars − crs兲sin␤x , . . . . . . (A-14)
cos␤⳱−1 and +1, respectively) The spectral radius ␳E of E is the eigenvalue of E of maximum
absolute value. We impose the stability requirement ␳E<1. The
a + b + c ⬍ 2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (A-5a) value of ␳E obviously depends upon the eigenvalues ␤1, ␤2, which
a + c ⬍ b. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (A-5b) assume all values corresponding to the ranges −1<cos␤1<1 and
−1<cos␤2<1. In the two-phase case above, the most restrictive
Adding these two conditions gives a+c<1. The second root leads to stability condition |␭|<1 resulted for the eigenvalue cos␤⳱−1,
nothing of consequence. For N>1, all difference forms of type Eq. sin␤⳱0. Assuming that is also true for this three-phase case gives
A-1 in this paper satisfy a+c⳱b so that the Condition A-5b is the eigenvalues cos␤1⳱cos␤2⳱−1, sin␤1⳱sin␤2⳱0. Using these
always satisfied. Therefore, the stability condition for Eq. A-1 is eigenvalues in Eq. A-14 gives real drs values and Eqs. A-13 give
A-5a or, using a+c⳱b, the (real) ␧rs elements of the amplification matrix E. The two
eigenvalues of the E matrix are then easily calculated as the roots
a + c = b ⬍ 1. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (A-6) of E ‘s quadratic characteristic equation. The stability condition
that ␳E<1 is obtained as
Discussion of the result b<2 for the case N⳱1 is dropped for
brevity. d11 + d22 + 公共d11 + d22兲2 − 4det共D兲 ⬍ 4, . . . . . . . . . (A-15)
Many authors point out that the stability condition for two or
three dimensions is simply the sum of two or three terms, each of where det(D)⳱d11d22 – d12d21. The drs satisfy
which is identical in form to that derived for the 1D case. For
example, consider Eq. A-1 written for three dimensions drs = 2brs rs = 11, 12, 21, 22 . . . . . . . . . . . . . . . . . . . . . (A-16a)

␧ijk, n+1 − ␧ijk, n = ax␧i+1, jkn − bx␧ijkn + cx␧i−1, jkn d11 ⬎ 0 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (A-16b)

+ ay␧i, j+1, kn − by␧ijkn + cy␧i, j−1, kn d22 ⬎ 0 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (A-16c)


+ az␧ij, k+1, n − bz␧ijkn + cz␧ij, k−1, n. . . . . . . . . . . . . . . . . . (A-7) det共D兲 ⬎ 0. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (A-16d)

186 June 2003 SPE Journal


These drs are related to the coefficients in the original Eqs. 33 by numerical tests indicate the condition is both sufficient and nec-
essary in the three-phase case.
d11 = 共q⬘wwi, i − q⬘wwi, i+1兲 2⌬t Ⲑ Vpi . . . . . . . . . . . . . . . . . . . (A-17a)
d12 = 共q⬘wgi, i − q⬘wgi, i+1兲 2⌬t Ⲑ Vpi . . . . . . . . . . . . . . . . . . . (A-17b)
SI Metric Conversion Factors
d21 = 共q⬘gwi, i − q⬘gwi, i+1兲 2⌬t Ⲑ Vpi . . . . . . . . . . . . . . . . . . . (A-17c)
bbl × 1.589 873 E–01 ⳱ m3
d22 = 共q⬘ggi, i − q⬘ggi, i+1兲 2⌬t Ⲑ Vpi. . . . . . . . . . . . . . . . . . . . (A-17d) 3
ft × 2.831 685 E–02 ⳱ m3
As in the two-phase case discussed above, it might appear that lbm × 4.535 924 E–01 ⳱ kg
the form of each of these drs terms would depend upon which of psi × 6.894 757 E+00 ⳱ kPa
the 12 possible three-phase cocurrent/countercurrent flow configu-
rations is assumed. Again, however, that is not the case. Each drs Keith H. Coats develops reservoir simulation software at Coats
is independent of the flow configuration. Eq. A-17, together with Engineering, Inc. e-mail: keith@coatsengineering.com. He pre-
differentiation of Eq. 30, gives the drs values. That tedious pro- viously worked in simulation software development for Inter-
cess, together with the Condition A-15, gives the final stability comp, which he cofounded in 1968. He was elected to the
Condition 2 with Eqs. 3 through 5. National Academy of Engineering in 1988 and was selected by
Hart’s E&P as one of the 100 most influential people of the
For each of the three two-phase cases, the stability Condition 2 petroleum century in March 2000. Coats holds BS, MS, and PhD
is both necessary and sufficient. For the three-phase case, the degrees in chemical engineering and an MS degree in math-
condition is a necessary one, valid for the above described choice ematics, all from the U. of Michigan. He has served as an SPE
of eigenvalues ␤1 and ␤2. However, there is no proof here that the Distinguished Lecturer and is a recipient of the 1984 Lester C.
condition is also sufficient for the three-phase case. Numerous Uren award and the 1988 Anthony F. Lucas Gold Medal.

June 2003 SPE Journal 187

Вам также может понравиться