!

534

Comparisonof Alternating IXrection Explicit and Implicit

Procedures in Two-DimensionalFlow Calculations
K. H. COATS* ESSO PRODUCTION RESEARCH CO,
MEMBER AIME HOUSTON, TEX.
AMERICAN AIRLINES
M, H. TERHLINE
TUf_SAw OKLA,

ABSTRACT Eq. 1 at odd time steps by:

Ana !ys is and exarrrp[e applications have been
performed to compare the accuracy and computing Ax.: - ‘x
2n+l
‘i-14i
speed of alternating-direction explicit ad implicit L2M1 (U) =
procedures (ADEP and ADIP) in numerical solution AX2
of reservoir fluid flow problems. ADIP yields
significantly greater accuracy and requires about
60 per cent mere computing time than ADEP, not
300 or 500 per cent more as reported elsewhere, 1*2

INTRODUCTION
+q= o,.......(2)
SeveraI recent papers 1-3 discuss an altemating-
direct;on expIicit difference approximation (ADEP) and at even time steps by:
to the diffusion equation. Example applications
of ADE P an? ADIP4 were reported to support
conclusions that ADEP is comparable in accuracy
to ADIP and requires one-fifth to one-third the
computing time of ADIP. Applications of ADEP in
calculation of two-phase flow in reservoirs was
also proposed. 3
This study was performed to compare further the
relati-, e merits of ADEP and ADIP in simulation of
two-dimensional fIow of one and two tluid phases
in reservoirs. Since two-phase flow equations are
often essentially elliptic rather than parabolic, the +-q =0,.............(3)
efficiency of ADEP in solving the elliptic equation
was ako examined.
where
ADIP AND ADEP DIFFERENCE EQUATIONS
= U(ti> JAY, WM)
The diffusion equation: ‘; j

u ~+u+q=u . (1) Ax U“ n
YY t“””’’””” ij 5 ‘~1-l., j - ‘ij
governs heat conduction, molecular diffusion and
slightly compressib~e fluid flow through porous
n=n
media for the case of homogeneous, isotropic media.
~y ‘ij - ‘i, j+l - ‘;j
The ADEP procedure 1-3 involves replacement of

Original manuscriptreceived in Society ofPetroleum Engineers

office .luly 1, 1966. Revlaed manuscript received Nov. 22, 1966.
Paper (SPE 1534) was presented at SPE Gas Technology Sym- Sweeping a two-dimensional grid from southwest to
posium held in Omaha, Nebr., Sept. 15-16, 1966. @ Cop yri~ht northeast using Eq. 2 and from northeast to sou&-
1966 American Inst kute of Mining, Metallurgical, and Petroleum
Engineers, Inc. west using Eq. 3 allows direcr (explicit) calculation
*Presently associate profe seer of petroleum engineering at of u at the new time step at each grid point.
The U. of Texas, Austin, Tex, ADIP 4 repIaces Eq. 1 by:
~R~f=re~ceSgiven aten.i Of paper.

950 SOCIETY OF PETROLE’kS ENGINEERS JOURNAL

 Ay z are identical for both techniques and are not
included in Eqs. 6 and 7. The presence of Ax in
the denominator of the leading term in the ADEP
truncation error indicates inferior accuracy to
ADIr-” because Atz/Ax is an order of magnitude
2n+l- ~2n larger than At 2, the leading term in ADIP truncation
‘ii i. error.
~+q= o”””@)
At RE PRESENTATION OF INSULATED
BOUNDARIES WITH ADEP
at odd time steps, and by: Fig. 1 shows two types of spatial grids for
numerical simulation of flow in reservoirs. For
problems involving closed exterior boundaries, the
difference representation A %/ij of u%% + UYY in
Eq. 1 must satisfy

2rn-2 - u2n+l
IJ
‘ii i.j EA2U =().... . . (8)
. +q=o”(s) z
ij
i=l j=l

at even time steps, where to preserve the no-flow condition at the boundaries.
If a grid of type shown in Fig, lb is empIoyed, the
~2 Un n side points must be weighted in Eq. 8 by a factor
x ij = ‘i+l, j - ‘U;j + U;-l, j of one half and corner points by one fourth.
As shown in Appendix A, ADIP satisfies Eq.
8 exactly for either type of grid. ADEP, however,
~2 Un yields an error term of order (At)z on the right side
of Eq. 8 for grids of type shown in Fig. la. This
Y ij = ‘;,j+l - 2u~j + ‘;, j-l
error will cause the material balance in ADEP
,
Each of Eqs. 4 or 5 implicitly relates” three un-
known (new time step) u vaIues at each grid point
and leads to a set of simultaneous equations J ● ● ● ● ● ●
6
easily solved by Gaussian eli&ination.

TRUNCATION ERRORS ● ● ● ● ● ●

The truncation error of ADEP, defined as

[L ‘1 9
2*JU) + L2*2 (0
t ● ● ● ● ●

- 2(U= + Uyy - Ut)t . 2nA~ ],

1 ● ● ● ● ● ●

is: I s— t
-,

Al u + At2
+QA$ +OA$
AX Xtt Ay ‘ytt
()()
. . . . . . . . . (6)
The ADIP truncation error, defined as

[R2M1(u) + R2N2(u)

- 2(U Xx+u - Ut)t = 2nAt],

YY

is:

2
At2 U Uttt + u(At3) . (7)
yytt “ ~+ {81

Truncation error contribution of order Ax Q and FIG. 1 — TWO TYPES OF GRIDS.

DECEMBER, 1966 Ssl

calculations to differ from 1.0 and may give rise to u ~+u =Ut Osxsl, Osysl
extremely serious errors if weIls (sources or sinks) YY
are placed close to the insulated boundaries. This . . . . . . . . . . . . . . . . . . . . (1)
statement ia i llusttated in the following example
calcukions.
Repreaencation of insulated boundaries by ADEP
with grids of type shown in Fig. lb may be achieved U(x, y,o)= l””””””” “Ill)
in several ways. Four cases are described in
detail in Appendix A and are brief!y summarized
here, since consideration of the first example 2!? .&=oby X=o, o=ysl
problem discussed below requires a choice of one
ax
of these alternatives. This problem involves a and y = 0, c)<xsl, ~llt. .(l2)
unit square with two insulated and two constant
potential boundaries (Fig. 2). As shown in Eq. 2,
ADEP represents AX2 uof at odd time steps by
(with the j index suppressed): U(l, y,t) = U(x,l, t) = o “ “ “ “(13)

Ml
$ - ‘: - (Uo uql) -’ ‘ “ “ “ (g) The analytical solution to this problem is:

Case 1 insulates the boundary x =“r’ = O of Fig. lb

setting u-in + 1 equal to u In leaving Eq. 9 other-
wise unchanged. Case 2 is . ore consistent in
preserving the time IeveI of the first difference in
x by replacing Eq. 9 by:

‘:-”; -(u:-u:)””””””(IO) which gives 0.62177 at x = y = 0.5 and t = 0.08.

I.arkin compared the ADIP and ADEP numerical
solutions at rhis position and rime in the form of
Case 3 avoids the insulated boundary difficulty Table 1 which gives the difference between the
(with ADEP) by treating the square of side 2 with numerical so!utions and 0.62177. Spatial increments
zero potentiaI imposed on” all sides. The unit of 0.1 in each direction were used. On the basis of
square with two insulated boundaries is simply this comparison at t = 0.08, Larkin concluded the
the upper right-hand quadrant of this larger square. methods ‘were of roughly equivalent accuracy.
Case 4 preserves Eq. 9, reflecting U-ln+l to ADIP and ADEP were rmo~rammed in this studv
u In+l, but re5ul~s in 21 + 2] -2 equations requiring for Larkin’s prhblem, and” th~ differences betwee~
simultaneous solution, where 1 and ~ are the total numerical results and the analytical soIution (Eq.
numbers of grid points in the z and y directions, 14) are pIotted vs time in Figs. 3 through 8 for
respectively. None of Cases 1, 2 or 4 satisfies rime increments of 0,0025 and 0.02. The plotted
Eq, 8 (Appendix A). per cent error is defined by 100 x (u* - ‘Eq. 14/
1- uEaL . .. Id
. . which is actual error expressed as a
COMPARISON OF ADIP AND ADEP per cent of the tota~ change in u from the initial
USING LARKIN’S EXAMPLE value of 1.0; u* is the numerical ADIP or ADEP
solution.
Larkin 1 applied ADIP and ADEP to the diffusion
Figs. 3 and 4 compare ADIP and ADEP errors
equation in the unit square for conditions noted on
at the center point x = Y = 0.5 for Case 1 treat.
Fig. 2. Mathematical statement of the problem is: ment of ADEP differences at the insuIated
boundaries. Fig. 3 also shows Lne analytical
solution. These figures show the pronoun ced
Y superior accuracy of ADIP ac small (0.0025) and’
t
U=o more practical (0.02) time steps for Case 1 ADEP.
(0,1 ) Figs. 5 and 6 show that the Case 2 treatment of
the ADEP scheme at the insulated boundary is
/
superior to that of Case 1. At the 0.0025 (critical)
/ time step, ADEP is comparable or even slightly
“Sl
/ INITIALLY superior in accuracy to ADIP but this is of little

TASLE 1 _ LARKIN’S ERROR COMPARISON

Ai “
x O*O2 0001 0,005 000025
——
INSULATED ADEP - 0,02S5 0.0012 0,0000 -000013
FIG. 2 — LARKIN’S EXAMPLE PROBLEM. ADIP 0,0032 -0.0039 - 0s0021 -0,0019

S6B SOCIETY OF PETROLEUM ENGINEERS JOURNAL

practical interest. At the 0.02 time step, however,
the results again show the superior accuracy of Atc=tiL . . . . . . (15)
2

H
ADIP.
Figs. 7 and 8 correspond to Case 3 ADEP treat- 2[1+5 ]
ment where the question of difference form at
insulated boundaries is avoided by solving the or At== [(0.1 )2]/4 = 0.~02~ for this problem. Thus,
larger probIem of a 2 x 2 square with all sides the above comparisons of the two methods were
maintained at zero potential. These figures again made for time increments up to 0.02/0.0025 or
show the superior AIXP accuracy at the 0.02 time eight times rhe critictd.
step. Relative accuracies of merhods such as ADIP
Figs. 3 through 8 show the erroneous conclusions and ADEP at the critical time step are of little
rhat can be reached by simply comparing errors at practical interest since the sole advantage of
the single time t = 0.08. The critical time increment these techniques is their provision of stability
ia defined as the maximum At at which the normal at considerably larger than critical At’s and
explicit method (u%% + ~Y@ A:= (~ ‘+1 - ~n)/At attendant reduced computing time requirements.
is stable. This increment Is:
ANALYSIS OF QUON ET AL.. 2
RESERVOIR EXAMPLE

,*. Quon et al. 2 simulated a heterogeneous under-

,, ----- ----
~.- ti-- ‘*

. .
./’ ml.

- ,.
~

1
.,
\

! -.,
t!
* *
Y -
1’
OS

.,. % r“,

- M
: ,
-..- 1

I
-., I
-,

tad.!
FIG. 5 — COMPARISON OF ADIP AND ATIEP ERRORS
FIG, 3 — COMPARISON? OF ADIP AND ADEP ERRORS (CASE 2)0
(CASE 1).
10

,Q

Q
I

j-, ‘u I

- c

- 1. #

I
. .*W *,*

“i’.,
m I*RN

FIG. 4 — COMPARISON OF ADIP AND ADEP ERRORS FIG. 6 — COMPARISON OF ADIP AND ADEP ERRORS
(CASE 1). (CASE 2).

DECEMBER. 1966 ass

saturated oil reservoir about three miles wide by 0.00633 k Atc ~~z
—
five miles long. The reservoir was heterogeneous
p$bc =
with the following properties:
2 [[)7
I+t;
At = 15 days
Ax = Ay = 1,320 ft or At = 0.12 day. A rough guide for selection
pavg = 2.5 Cp of At & derived in Appendix B; this guide indicates
kb avg = 70,000 md-ft that time steps of about 30 Atc, or about four days
$rMavg = 7 ft in this case, are reasonable at times prior to the
c ~ 7 x 10~ psi-l onset of quasi-steady-state.
Actually, a nearly exact solution to the difference
Pinitial = 1,065 psia,
equation for times greater than 50 days could be
with six wells producing at 625 B/D and eight obtained by setting dp/13t in the diffusion equation
wells producing at 375 B/D. to a constant a which can be easily calculated
A we!l-known relationship giving the time from the total reservoir volume-compressibility
necessary for a wel! producing from a closed product and the total production rate. An elliptic
reservcir to reach quasi-steady state is: equation then resulis which need be solved only
once (i.e., not repetitively at successive time steps)
to obtain the pressure distribution which is then
t* =
4(%&daYs . . . . . . . . . (16) positioned for any given time about the average
reservoir pressure at that time.
Insertion of the above data with the maximum
possib!e distance of five miles used for re gives: COMPARISON OF METHODS
USING WELL PROBLEM
2
= 2.5 (7) (7 x 10-6) (5 X 5280) ADIP and ADE P were compared in this work for
t
4 (.00633) (70,000) the problem of a welI located in the cecter of a
square reservoir containing undersaturated oil
=48days . . . . . ..(I7) (Fig. 9). The governing equation is:

Thus, one time step of 15 days represents about P=+ Py>, =p~ . . . . . ..(W
30 per cent of the time necessa.y for a quasi-
steady - state regime to occur. A comparison with px=Oatx=Oandl, and py=Oaty=O and
between ADIP and ADEP at a time corresponding 1. Initially, p is zero and fluid Injection occurred
to 96 of these increments (1,440 days) thus has at x = y = O. The injection rare used in the numerical
quesriona ble solution was normalized so that the analytical
significance. A more meaningful
comparison would be one at times !ess than 48 soIucion:
days, using time increments considerably less

()
than 15 days. 2
, The critical time incremeni for this problem is P =J-Ei4:
41?
. . ..o .(19)
given by
m
I I I I I L<-
,/ ‘-+’ ‘-. ---, --
/
#
I ,,,,
i\
f

-t-
I

,,.
c+,, . ... . I
*4
OM
Ill
aon: m ** 1*
I
.*
I ,, , .* ,.,. ,
!,,4SI0 ,, *
..’:. ,
,,- . !

FIG, 7 — COMPARISON OF ADIP AND ADEP ERRORS FIG; 8 — COMPARISON OF ADIP AND ADEP ERRORS
(CASE 3). (CASE 3).

sad sOCIETY OF PETROLEUM ENGINEERS JOURNAL

 TABLE 2 - CASE 2 - ADEP RESULTS FOR FIG. lb
GRID WITH WELL AT i = ❑ O IN UNIT SQUAF?E.VARIABLE
~, CASE ‘
1,1)
iOOl) Material
PADEP - ‘Eg. 19 ~ ,00 Balance
for Unit
t pEq, 19 Square

0.005 224.4 1.261

0.013, “ 39*9 1.205
ALL S19CS 0.027 16,6 1.249
INSULATED
0.057 6,7 1.352
0.077 320 1.341
2,4 1,366

I
0.!37
0.217 2,9 1,308
0.257 3,5 1,283

(1,0 t ‘it
WELL
o -0.001 0.0005
FIG, 9 — WELL PROBLEM CONFI~JRATION. 0,001-0,005 0.001
0,005-0.017 0,002
P 0.o1- -0.037 0,005
wpii- each point r = ~ x 2 + yz for times
at at 0.037-0.097 0.01
which the effect of the exterior boundary is 0.097 -0.657 0.02
negligible. The effect of the exterior boundary is 0.657-0.857 0.05
negligible at the well for times up to: 0.857- 1.457 0.1
1.457 - 1.957 (),1>
2.957-5.457 0.25

ADIP calculations were performed with a grid of

which is the time at which a quasi- steady- s!ate type shown in Fig. lb and 20 itrcrements along each
regime begins. side of the quarter square with side = 1. ADEP
Calculations were performed for increasing time results were obtained by treating the square of side
steps as foIlcws: . 2 with the well at the center and with a grid of
twe
. . shown in Fig. la. Preservation of the 0.G5 Ax
value used in the unit-square AIXP calculation

$0
‘\. —

d r \
\
Is

‘\ &DC*

!0
\ \

.L-----LrLu-74- I ‘

.\
‘\}

““ET*. /“ , T- I
.15 .5
,01 .* .1
.m, .005

FIG. 10 — COMPARISON OF ADIP AND ADEP ERRORS (WELL PROBLEM).

855
DECEMBER, 1966
required 41 spatial increments along each aide of Constant At simulations were also carried out on
the square of side = 2. a unit square (well at corner) with Fig, lb gri for
Fig. 10 compares the percentage errors in ADIP ADIP and on a square of side 2 (well at center)t
and ADEP sokrtions at the point x = 0.25, y = O up with Fig. la grid for ADEP. The criticaI ti~e
to a time of 0.25. At this time. exterior boundarv A%2 ~@13025\
increment for this problem is .&t= . — . —
effects appeared at rhis position and the analytical 4 47
solution (Eq. 19), no longer applied. The pIotted 0,000625 while quasi-steady-state is reached at
Pnumerical sol’n -P* where ~, is ~e2
errors are 100 x tQss = :-S 0.25. Quon et al. z used a constant M
“P*
the analytical solution (Eq. 19). The negligible about 30 per cent of their quasi-steady-stare time.
effect cn well wessure” of the exterior boundarv at Here, a At of 0.06, about 25 per cent of tQ~~, was
times prior to’ 0.25 was checked by numeric’slly employed. This increment is about 100 times the
solving Eq. 1 for a somewhat larger square and critical time step.
noticing the identity of *he unit and larger square
solutions (compared to the errors from Eq. IS) at Fig. 10 shows the variable time step ADIP
x = 0.25, y = O for z s 0.25. The analytical solution results to be closing within 1 per cent of the true
for pressure is also shown on Fig. 10. The results solution at time = 0,25. Error in the constant time
step calculations was therefore defined as 100 x
for this case show a pronounced superior accuracy
for ADIP. Attempts to use a grid of type shown in ~’ where p* is Eq. 19 for time <0.25 and is
Fig. lb with the Case 2 ADEP procedure gave P*
excessively large errors (Table 2). The reason for the ADIP solutior, using the variable time steps for
this error of ADEP is discussed in Table 2 and in t > 0.25.
Appendix A: ADEP fails to preserve the no-flow Fig. 12 compares the ADIP and ADEP errors for
condition ac the insulated boundaries, and the error the case of constant (k = 0.06. The results again
thus introduced is increased as the perturbing well show ADIP to be considerably more accurate. Fig.
is located closer to the boundary. 13 compares the 1.0000 ADIP rilateriaI balance to
The more reasonable ADEP resuIts shown in the ADEP balance which immediately deviates over
Fig. IO correspond to a well in the center of a 40 per cetit from uni:y since transients reach the
square of side 2; transients at the ifistiIated bound- insulated boundaries in one or two time steps of
aries were delayed in time and Iess severe than 0,06.
when the well was on rheqboundary. The attendant
The Case 2 ADEP results for the case of the
ADEP error induced by faifure to preserve insulation
well at the corner of the unit souare
. on a Fig. lb
was therefore reduced. Fig, 11 compares the 1,0000
grid with :\t = 0.06 were as follows:
ADIP materiaI balance with the ADEP baIance.
which deviates from unity when–transients reach the
insulated boundaries.

‘“’~ .

= ,mo - ,24

,01

TIMIK - t

FIG, I l—COMPARISON OF. ADIP AND ADEP MATERIAL BALANCES (WELL PROErLEbO.

SS6 SOCIETY OF PBTROLEIIM ENGINEERS JO[lRN AI,

 PADEP -P* ~ loo TABLE 3- ADIP ANG ADEP ERRORS FOR At= 0,012
Material
* BaIance
t (P - P~q. J x loO/f@+ ~g ADEP
Moterld
0.12 19,307 180.664 time ADIP ADEP Balance
0.36 11,469 61.084
0.024 48.96 34.28 0,9946
0.60 7,413 30.983 0.048 -17,39 -27,47 1,0020
1.2 3,380 11.381 0.072 7,27 -23,35 1.0039
2.4 1,482 4.587 0.096 -6.82 -19.79 1,0048
3.6 1,038 3.145 0,12 2,64 -16,07 1,0052
4.8 860 2.577 0.24 - O*O7 - 5.71 1.0033-
5.52 773 2.372

As previously mentioned, the 0.06 time increment approach will succeed only for a limited rime step
is about 100 times the criticai At. Since increments size; the Imitation on time increment is far less
of the order of 20 times the critical are more severe if iteration is employed. As shown below,
reasonable, a fairer comparison of ADEP and the limitation on time step in the noniterative
ADIP errors might be obtained by using a At approach is generally so severe that fireater
value of 20 (0.0006) or 0,012. These errors are’ computing efficiency in two-phase flow problems
given in Table 3 for the point z = 0.~5) Y = o. The is attained by iterating ac each time step.
ADEP results were again obtained from the square The question considered here is not whether
of side 2 with the well pIaced in the center. The .ADEP is superior or inferior to ADIP in noniterative
ADEP material balance is also given in Table 3; solution of two-phase fIow problems, The question
ADIP material balance was 1.00000 over all time. is whether the two-phase f!ow problem is essentially
parabolic (i. e,, subject to ncniterative solution
UTILITY OF NONI~ERATfVE ADIP OR with reasonably large time steps) or elliptic (i. e.,
ADEP IN TWO-PHASE FLOW PROBLEMS requiring iteration for use of a reasonable time
step). In either case, the use of ADIP is indicated
QuorI et al.3 proposed a noniterative application since, as shown above, ADIP is clearly superior to
of ADEP in solution of the two partial differential ADEP in the parabolic case and, as shown in
equations governing two-phase flow in reservoirs. Appendix C, in the elliptic case ADEP becomes
ADEP or ADIP can be employed in noniterative identical to the extrapolated Lkbrnann method that
solution of these equations only if fluid compress- has been proven inferior 4 to iterative ADIP for
ibility is not zero. If compressibility is zero, then unit k, ~ and c.
iteration is required at each time step. Even if In earlier work the authors attempted to apply
compressibility is not zero, the noniterative ADIP in noniterati ve solution of two- and three-

60.

.
40

, 20

~“
*
AK?
// ‘--
4 -

-*O . /--
OH

-4 0.
7/-
/ “
/ P

-s0 %Wml

hf. w K.es 7.0

TIME -t

FIG. 12— COMPARISON OF ADIP AND ADEP ERRORS (WELL PROBLEM),

DECEMBER, 1966 3s7

dimensional two-phase flow probIems. The approach Irr Appendix B, a terminal time increment of
succeeded only for the very limited class of large- the order of 50 Atc is shown to be reasonable in
block areal problems; iteration at each time step numerical solution of Eq. 21. Irr two-phase flow,
was found to be far more efficient in the majority however, perturbations are corrtinuaIly being
of probleims of interest. The reason for the advantage produced by saturation and mobiIiry changes; thus,
of iteration can be demonstrated by an analysis of a time increment more suitable to an early portion
the equarion governing pressure distribution. of the transient period should be employed. There-
Capillary pressure and saturation change terms fore, let 50 Atc be selected as rm upper bound on
may be dropped from the equation for simplification a suitable time increment for two-phase flow
of the analysis without detraction from the resuIts problems.
or conclusions. The resulting equation: Eqs. 22 through 24 allow calculation of Atc for
any given reservoir and fluid properties. Let T = 5 x
IO--S l/Psi; ~ = 0.20; and M = 200 md/cp. Then,
for a two-dimensional areal problem with lx = .fy,
Eq. 23 gives:
. . . . . (21)
where M = k (k ~/p 1 + k2/~42), subscripts refer to .2(5 X 10-5) ~X2 =
A~c =
to fluid phases Z = c1 S1 + C2 S2 is parabolic if 2(200) (.0063) (2) ●

@ # O and is elliptic otherwise. in the elliptic

case, the difference form of Eq. 21 is augmented by -6
a term Hk (p ‘+1 - p $ on the right-hand side and 2X1O AX2 days . . . . . (25)
solved iteratively by the irerative ADIP method. A
The critical time increments for the explicit For a large reservoir with lx = 1,000 fr, /\t= =
difference equivalent of Eq. 21, for one-, two- and two days and an upper limit of 50 h= or 100 days
three-dimensional fIow, are: per time step is estimated. Use of this or even
smaller time steps in noniterative solution of the
fi; &2 twG-phase flow equations completes in efficiency
ID flow: Atc = ~M . . . (22)
with the iterative method where time steps of 360
days h~ve typically been successfully employed
. in such large reservoirs.
21) flow: Atc = A2L4.ixT . (23) Simulation
10 increments
of a quarter 20:acre five-spot
on each of the 467-ft sides gi~es
using

2Mp +&) 1 a i\~ of about 50 ft which gives, from Eq. 25, ,ltc =
- 0.005 days. In this case, 50 Atr is only 0.25 da~s, . .
@: Ax’ and a noniterative sioIution using this At is far more
3D flow: At. = expensive than an actually performed iterative
(&}2] solution which employed a 30-day time step.
The above two cases indicate the applicability
.. . . . . . .(24) of the noniterative
areal cases with sufficiently
method in two-dimensional
large Ax. The iterative
,,
method is far more efficient even for reservoirs of
large areaI extent, however; if a two-dimensional
4 cross-section or three-dimensional simulation is
‘a f \\ performed. For the three-dimensional case with
.\z << Ax, Eq. 24 gives Atc S 4 x 10-6 & 2 and
/\ for a h of 10 ft, AtC = 0.0004 day. The tolerable
!s- I increment of 50 Atc or 0.02 day is so small that
j ‘\, the iterative method is two orders of magni tu~ie
~,, I \,
cheaper than the noniterative.
To repeat, the above analysis simply indicates
;
:
1
1
I
“&,, why in the writers’ opinions, the noniterative
is generally inferior to the iterative solution,
method
The
\\ analysis in no sense constitutes a proof of this
,!
‘\ conclusion. The validity of the conclusion rests
I ‘\\ on the writers’ experience in a ctuaIly soIving
-.
.O,* ‘-. ___ _ two- and three-dimensional two-phase flow problems
,*
by both techniques.

,! . ac COMPUTING TI!,lE REQUIREMENTS

o*O , . ●
,,.[., The numbers of arithmetic operations per grid
point per time step for ADEP and ADIP ire:
FIG. 13—COMPARISON OF ADIP AND ADEP MATERIAL
BALANCES (WELL PROBLEM).

358 sOCIETY OF PET ROLEIIM ENGINEERs JOURNAL

 Additions or Multi- NOMENCLATURE
Subtractions placations Divisions
ADIP = alternating-direction implicit procedure
ADEPAX . Ay 4 2 0
ADEP = alternating-direction explicit procedure
ADIP 5 3 1 c = compressibility of fluid and rock, psia-l
4 3 0 h = reservoir thickness, ft
‘DEPAx + Ay
4 1 i, j = grid ~ndices, x = iAx8 y = jAy
ADIP 5
I = total number of grid points in x direction
Relative computing times for these floating-point,
single-precision operations far the Burroughs 205 ~ = total number of grid points in y direction
computer used by Larkin, the IBM 7040 used by k = permeability, md
Quon et al. and the IBM 7044 used here, are as n = time index, t = tn or ~ = KU3t for constant At
follows:” p = pressure, psi
Addition or Multi- pw = well pressure
Subtraction plication Division p= = pressure at exterior boundary
Burroughs 205 2.5 10.1 13.5 q = injection rate, volume fluid /volume of
IBM 7040 3 4.5 7 reservoir-unit time
IBM 7044 5.5 10 18
qw = well injection rate, B/D
Weighting the mix of operations for each method by r = radius
these relative computing times gives the following r~ = well radius, ft
ADIP:ADEP ccmputing time ratios: r= = radius of closed exterior boundary
Burroughs 205 IBM 7040 IBM 7044 Sj = saturation of fluid phase i
Ax = Ay 1.87 1.69 1.8 t = time days, where units are implied
Ax & .Iy 1.65 1.57 1,65 .Vc = critical time increment for normal explicit
If, as Peacemans suggested, certain intermediate difference scheme
data are stored rather than recalculated each time tD = 0.00633 kt/p ~ crw2
step in the ADIP procedure, then one Iess multi- x, Y, 2 = spatial coordinates
plication eld division are required. The above ~ = porosity
ADIP:ADEP ratios then become:
p = viscosity, cp
Burroughs 205 IBM
.— 7040 IB,M 7044

Ax = Ay 1.08 1.14 1.13 ACKNOWLEDGhiENT

Ax + ,Ay 1.o6 1.12 1.11 The authors benefited from suggestions by D. W.
“Ilrese ADIP:ADEP ratios indicate that the ratios Peaceman and appreciate permission of the Esso
of five (Larkin) and three (Quon et al. ) are excessive. Production Research Co. to publish this paper.

CONCLUSIONS REFERENCES
1. Larkin, B, K.: Math. of Corn@ (1964) vol. 1S, No, 86,
1. ADEP is nonconservative in that it fails to 196.
preserve no-flow conditions at exterior boundaries.
2, Quon, D., Dranchuk, P. M., Allads, S. R. and Lmuw,
This causes errors in peter.tial and in material P. K,: ~~Astable, Explicit, Computationally Efficient
balance which can become extremely severe if wells Method for Solving Two - Dimensional Mathematical
are near the insulated bounda::ies. hlodels of Petroleum Reservoirs”, paper presented at
Petroleum end Natural Gas Div., CIM, Calgary, Alberta,
2. ADIP accuracy is cor,siderab!y superior to
Canada (May, 1965),
ADEP for Larkin’s example problem of fluid flow or
3. Quon, D., Dranchuk, P, M., Allada, S. R. and Leung,
diffusion in a unit square. P, K,: ‘ ‘Application of the Alternating Direction Explicit
3. ADIP accuracy was found to be considerably Procedure to Two-Dimensional Natural Gas and Reaer-
superior to that of ADiZP in a closed-reservoir type voira}}, Sot. Pet. Eng, ,Jow. (June, 1966) 137-142.
of problem, even when the well was located as far 4, Peaceman, D. W. and Rachford, H. H., Jr.: “The
from the boundary as possible in the ADEP case. Numerical Solution of Parabolic and Elliptic Differen-
tial Equations”, Jaur. SOC. [ndust. Appi. Math (1955)
4. Two-phase flow problems are mcme efficiently
Vol. 3, 28-41.
treated by iteration except for cases of very large
s. Peaceman, D. W.: Computitrg Reviews (Nov. -Dee., 1964)
two-dimensional areal cases. The iterative adapta-
vol. 5, No. 6, 389.
tion of ADEP is identical to tbe well-known
6, Richtmyer, R, D.: Dif/eretme Methods for Initial-Value
extrapolated Liebmann technique which has been Problems, Interscience Publishers, Inc., New York,
proven inferior to ite~ative ADIP. N. Y. (1957) 103.
5. ADIP requires about 60 per cent more comput- 7. Craft, B, C. and Hawkins, M. F., Jr,: Applied Petroleum
ing time than ADEP. Resetvoir Etrgineering, Prentice-Hall, Inc., Englewood
Cliffs, N, J, (1959).
*The f tgures for the different computers bear no relation to 8. young, D,; ~{Iterative Methods fOr solving Pafiial
each other; i.e,, a multiplication on the IBM 7040does not require
Diffe&tial Equations of Elliptic Type $~, Tmtzs.,
4.5/ 10the time for a multiplication on the IBM 7044.
American Mathematics Society (1954) Vol. 76, 92-111.
DECEMBER, 1966 a69
 APPENDIX A 2n+2 2n+l
- (Ul - 2U1 -t up)
REPRESENTATION OF INSULATED
BOUNDARY WITH ADEP

Consider solution of Eq. linarectan@e withaIl

four sides insulated. Fig. la shows the rectangle
with a grid placing points )f’-grid spacing in from the
boundaries. A difference representation of A~iiof
u xx + Uyy in Eq. I must satisfy the equation:
. (A-4)
IJ ADEP therefore faiIs to satisfy Eq. A-1 by an error
z zA2uij=o””””” @l) of order At 2. The Utt multiplier in Eq. A-4 implies
i=l j=l that this error introduced by the ADEP procedure
will increase as wells are placed cIoser to rhe
if the difference form preserves the no-flow condition boundaries. The failure of ADEP to satisfy Eq. A-1
at the boundaries. Satisfaction of Eq. A-1 by ADEP wiIl be refIected in material balances differing from
and ADIP can be examined wirk increased clarity 1.0.
and no loss in validity by considering satisfaction Problems in which boundar,y values of u are
of the relation: specified are more saris factoriIy treated by a grid
. of type shown in Fig. lb which places points on
the boundaries. If a problem is of mixed type with
buij=o . .. . . . (A-2) some insulated and some specified boundaries, then
i“= 1 x several variations of ADEP may be employed when
the grid of Fig. lb is used. Larkin’s example
Tie insuIated boundary is represented by the problem is of this mixed condition type involving a
difference boundary conditions: unit square with two adjacent insulated sides with
n
the two opposite sides held at zero potentiaI. Four
u (A-3) cases will be defined here for applying ADEP to
o,j ‘U; ,j ”-o-o-.””
such a problem with the grid of Fig. lb.
forj==I,2, ..., j. The ADIP differencing given in Simplicity is served, with no loss in validity, if
Eqs. 4 and 5 satisfies Eq. A-2 (and, therefore, Eq. only .4%2u.. portion of rhe ADEP difference form is
A-1) exactly, since: discussed % reIation to treatment of the insulated
boundary, At odd time steps at the insulated
I boundary x = z’= 0, with j suppressed:
n
X ‘“~+l, j - 2u~j + ‘f_-l, j) = 0
i=l

provided conditions in Eq. A-3 are imposed. ., (A-5)

Substitution of the ADEP differencing into Eq. No implicit u~~+l value is available; thus, one
A-2 gives possibility is reflecting u~~+l to u?. At even time
steps x = i = O.

and an explicit u~~+l value is available as u~+l.

This procedure is Iabeled as Case 1 in the treatment
at odd time steps and of J.arkin’s example problem. Case 2 arises from
rhe observation that a somewhat more consistent
I 2rl+2 2rtf-2 2*1 2n+l
treatment of rhe difference form, Eq. A-5 results if
- u. ) - (Uij - ‘i-l,j) the time level of the first difference u~n+l - u<~+l
z ‘Ui+l, j l~j
i= 1 is preserved in both terms of the difference. Thus,
if u_l2n+l IS
. set equal to u~n, then the Un
2n+l term
2rt?-2 2rl+l 2*2 2tl+l
shoul~ be replaced by u&? Thus, Case- 2 treats
=U -u -(U1-U1)
I I A 2 Uij in Eq. A-5 as:
“’x
at even time steps, For a full cycle:
A2UOj=U~-U$.
x ($ - U12n ). (A-6)

Case 3 avoids the problem of treating A 2 u ii on

.950 SOCIETY OF PETROLEUM ENGINEERS JOURNAL

the insulated boundaries by soIving Eq. 1 in a The well pressure is given approximately by the
square with aides 2 units long and with all stales infinite reservoir solution:
held at zero potential. The upper right-hand quadrant
is then identical to the original problem of a unit 70.7 qwp
square with sides z=O and y =0 insulated. ~W ‘ kb [ln t~ -f 0.8091, . . . .(9-2)
Case 4 treats the unit square with no alteration
of the ADEP scheme but requires simultaneous for 25 < tDS Rz/4 = N 2 and by the quasi-steady-
solution of a set of e-uations at points along the state solution:
insulated boundaries. Thus, if the time indexing of
Eq. A-5 is preserved, then u~~+l = u ~n+l and:
Pw = 141.4~&@D + +)
2
A = up+ U:*1 - u;” - $+1
3R4_ 4R4 lnR _ 2R2-1
x ‘O j
9* . .(B-3)
. . . . . .,0 . . . . . . . . . . . (A-7) 4(R2 - 1)2
Eq. 2 then contains two unknowns at each point
(O,j), j=o, 1..., ) and (itO), i = 1,2,...,1. Addi- for tD2 N2. For R = 2N of the order of 20 or larger,
tional equations at (1, ~), j = O, 1, . . . , ) anti (i, 1), Eq. B-3 can be approximated by:
i=l,2 , . ...1 introduce no additional unknowns End
provide a set of 2~ + 2) -2 equations in the same 141.4 qwp 2
number of unknowns. Simultaneous solution of this PW = kh m (tD+;) “n’-+
set of equations increases the computing time
requirements of ADEP while still failing to preserve .,, . . . . . . . . . . . . . . . . .(B-4)
the insulated condition at the boundaries. In this
case for t D ~ N 2. After quasi-steady-state occurs, the
difference between well and exterior radius pressures
is constant at:
:-1 AX2 Uij
i=O
should contain no contribution
fori=O,l,.., , 1-2. Howeverjn~~s
from values of ui,
su~tifor a fuI f
pw - p, = 141.4*
EnR -+1
~--(B-5)
cycle, contains the terms 2 ~ 1 2n-t2
- ~o -Uo The maximum race of pressure change occurs at
which is approximately 2( Ax) u% - At 2 Utt. Thus, the well and is given by:
an error of order (Ax + At2) is incurred in failing to
preserve insulation at the boundary. apw 70.7 qwp 1
— 25 < tD < N2 . .( B-@
~D= kb tD .
APPENDIX B

TIME STEP SIZE FOR Let a dimensionless time increment AtD for the
SINGLE-PHASE FLOW CALCULATIONS numerical -solution be chosen so that the maximum
pressure change in one rime step is a fraction /
A useful criterion for time step size in two- of the ultimate pressure difference P ~ - Pe. Then
dimensional, single-phaae flow calculations might for 25 ~ tD < N2, Eq. B-6 gives
be the ratio between the critical rime increment and
a At sufficient to give a desired maximum change
in pressure over the grid.
An estimate of a practical time increment for
42=
fkD
Ju.HM4L 70.7
‘“p L
kb tD
two-dimensional single-phase reservoir problems is
developed here by analyzing the case of a single
well producing from a bounded reservoir. Consider or
the case of a weII producing from a well in the
center of a square reservoir of side 2L. If 2N
AtD ~ 2f (lnR - ~) CD . . . . . .(B-7)
spatial increments on a side are employed and the
well is Iocated at x = y = 0, then the critical
and from Eq. B-i
dimensionless time increment is:

AtD = 1,...,... (B-1) AtD

crit. — = 2f(ln 2N - & ~D . . .(B-8)
Atn
where tD = 0.00633 kt/p#crw2. This two-dimensional
‘cri?.
rectangular grid approximates the case of a well of
radius rw = —A%L= —
situated in the center of a Thus, for / = 0.02, N = 20, AtD/AtD = 0.128 tD
2 2N
bounded reservoir of exterior radius re = L. ‘l%us, and ranges from 3.2 :tt tD = 25 to 51c&\t ~D = N2=
R = re/rw = 2N. 400 when quasi-ste~dy-state is achieved.

DECEMBER. 1966 S61

 APPENDIX C

EQUIVALENCE OF ADEP AND

EXTRAPOLATED LIEBMANN
. . . . . . . . . . . . . . . . . . .(C-3)
TECHNIQUE IN SOLUTION OF ELLI?TIC
EQUATIONS
ac odd iterations and:
The extrapolated Liebmann or successive over-
relaxation technique8 treats the elliptic equation: 2k+2 2k+l + B U;~j + ‘i, 2k-t-l
j+l
‘ij - ‘ij (
u ~+u=o . . . . . ..(c.l)
YY

by the iterative sequence: + u2k+2 2k+2

- 2U7’ - 2u~l
i.=.l, j + Ui, j-1 )
k

(
k+l k ~k
‘ij = ‘lij + a ii-l, j + ‘i., j+l . . . . . . . . .(C-4)

at even iterations, Identifying a in Eq. C-2 with

k+l k-tl k B/(1 +2B) causes Eq, C-2 to assume a form
+ ‘i-l, j + ‘i, j-l - 4uij “ “(C-2) identical with Eq. C-3. Thus, the convergence
) rates of the extrapolated Liebmann and ADEP
techniques in iterative solution of the elliptic
where u ~ik = the kth iterate at grid pcint (i, j), Eq. C-1 are identical. This convergence rate
a . relaxation factor. Application of the ADE!? is independent of the direction of calculations sn
differencing scheme to Eq. C-1 yields: that the use of Eq. C-4 in place of Eq. c-3 or
alternate use of the two equations is immaterial.
2k+l ~ ~2k 2k 2k ***
ij+B
‘i.j ( ‘i.+l, j+ ‘i, j+l

362
SOCIETY OF PETROLEUM ENGINEERs JOURNAL
 Further Discussion of a Paper Published in %ciety of
Petroleum Engineers Journd9 December, 1966

Comparison of Alternating Direction Explicit and Implicit

Procedures in Two-Dimensional Flow Calculations

I
K. H. COATS ESSO PRODUCTION RESEARCH CO,
MEMBER AIME HOUSTON, TEX.
AMERICAN AIRLINES, INC.
M. H. TERHUNE TULSA, OKLA,

(Published on Page 350)

DISCUSSION

R, D, CARTER
MEMBER AIME
I PAN AMERICAN
TULSA, OKLA.
PETROLEUM CORP.

The diffusion equation dealt with by K, H, Coats the material balance error was sufficiently smail
and M. H. Terhune is similar to the equation (less than or equal to about 0, 3 per cent of the
governing isothermal flow of gas in porous media. original gas conrenr after 75 per cent of the gas
Because of the similarity of diffusion problems and had been removed) that the results could be of
gas flow problems a discussion of the writer’s practical value. The writer has found that the
experience ~: aPPIYing ADEP (or Saul’ev) 1 and material balance error tends to be reduced as the
ADIP 2 methods to the solution of reservoir-type combination of reservoir condition S, time srep
gas flow problems seems appropriate. In summary, size and mesh spacing approach that in which a
this experience indicates: (1) that ADEP extends conventional explicit-type difference equation
the range of conditions for which explicit methods could be employed. Useful ADEP solutions to gas
may be used in practice to obtain gas flow solutions; flow problems have been obtained for heterogeneous
(2) ADIP is a method of more general utility than reservoir problems in which rr;~ch of the reservoir
ADEP; and (3) the additional storage and com- has a sufficiently large critical time step size
puter time required by ADIP over ADEP should that a conventional explicit difference equation
normally be of little concern with present could be used, but that also contained some areas
day computers. These latter two conclusions of high permeability in which the conventional
are at variance with an earlier speculation by expIicit equation (Eq. l-A, Ref. 1) could not be
Carter 1 regarding the relative merits of ADIP and employed because the critical time step for the
ADEP for two-dimensional gas reservoir calcula- high permeability area was too small to be practical.
tions. Some problems of considerable practical interest
The gas flow equation can be put in a form have been encountered for which the ADEP mcrhod
which is the same as the diffusion equation has proven unusable. These are problems that are
except for a coefficient of the rime derivative characterized in general by a requirement of small
which varies slowly with the dependent variable grid point spacing coupled with large areas of
(Eq. 5-A, Ref. 1). If this coefficient is evaluated high permeability and/or reservoir pressure. This
at the beginning of each time step in a numerical results in a very small critical time step size for
solution, a non-iterative diffusion equation approach these large areas. Use of ADEP with practical
can be employed to obtain solutions to gas flow time step sizes in such cases has resulted in
problems. unacceptably high material balance errors, For
Coats and Terhune point out that the ADEP example, a material balance error of about 1 psr
method does not preserve material balance in cent of the original gas content has resulted after
diffusion equation, closed boundary, reservoir- 12,5 per cenr of the gas had been produced in a
type problems. The writer has obtained several case in which the ratio of critical time step size
solutions to two-dimensional gas reservoir problems to practical time step size was about 0,02 over
using an ADEP method (Eqs. 1O-A and 11-A, Ref. the entire region, A non-iterative ADIP method
1). In none of these solutions was materiai balance has been applied to some of these problems with
strictly preserved, but in many of these solutions success, and ADIP marerial balance errors have
been negligible. This ADIP method has also been
l~eferences given at end of dlscussimh
successfully applied to some of the problems for
MARCH, 1967 \’
 which usable ADEP solutions were obtained, and gas content for a case in which the ratio of
( could probabIy have been applied to alI of the critical time step size to actuaI time step size
problems for which usable ADEP solutions were was about 0.02 and more than 31 per cent of the
obtained. (
original gas had been produced.
The ADIP equations referred to above are, This discusser has found that, with the ADIP

I
[
Mtn+l/2, n (@tn+I,ti - @n,z) - &-I/2, n @m,
n method, it is convenient
additional dependent
to maintain storage of an
variable array over that
required by ADEP. With present day computers,
I - @rIJ-l,
J1[~+ ‘m,n+l/2 (Qtn, n+l - ‘fa, tz) this will not normally constitute an objection to
I the use of ADIP. The ADIP program required about
-M tm,n-1/2 (Qm,n - ‘m,n-l fli+l five fourths as much computing time per time step
as dte ADEP program, but neither be ADEP nor
d~ the ADIP program had been optimized. Both programs
(L) make repeated use of interpolation subrouti~.es.
= Cm,w~–—
~~ ~,n,i (@i+l - @i)m,n . . (1-A)
All of the problems discussed above were for
)
regions which had irregular (non-rectangular)
boundaries.
Mm+l/2, n (@m+l, z - C$m,n) - Mm-1/2, n(@m, n
[
- @m#n)
NOMENCLATURE
-@ m-l, n)] i+l + ‘m, n+l/2 (“m, tz+l
[
AdA
(0 m,n -@m,n-~ )] i @ (p) = flow potential defined as ]
- ‘m, n+l/2 PC p(A)z(A)

p = gas viscosity, function of pressure

d(-)
= Cm, n —- z = gas compressibility factor
(@ii-l - @i)m,n . . . (I-B)
d~ m,n, i M = function of position proportional to
()
permeability-thickness product
Eqs, 1 -A and 1- B are employed alternately. C = function of position proportional to porosity-
These equations are an obvious extension of the thickness product, square of mesh
ADIP equations first presented by Douglas, spacing and reciprocal of time interval
Peaceman and Rachford 2 for the solution of a p = pressure
! problem in ideal gas flow.
A = variable of integration corresponding to
Because the non-constant coefficient of the
pressure
I ~(+) m,n, i = subscripts denoting spatial and time
) — is evaluated at the beginning position
1 ‘ime ‘erivative d4r
()
I of the time step, a source of material balance
REFERENCES
error exists in Eqs. 1-A and 1-B (as well as
I
I in the corresponding ADEP equations) that does L Carter, R. D.: ~{Per fomance Predictions for Gas
not exist for the diffusion case. However, as Reservoirs Considering Two-Dimensional Unstesdy-
State FIOWt~, SOC. Pet, Eng. Jot/t.(March, Ig@ 35-43.
! previously stated, materiaI balance errors using
2 Douglas, Jim, Jr., Peaceman, D, W. and Rachford,
these ADIP equations have thus far proven to be
H. H,: I Icalculetion of Unsteady-State Gas F1OW
~ negligible, the !argest ADIP error encountered thus Within a Square Drainage Area, ” Trans., AIME ( 19S5)
far being less than 0.04 per cent of the original 190-195. ***

VI SOCIETY OF PETROLEUM ENGINEERS JOURNAL