JOI’RNAI. OF COhlPI~TATIONAI. PHYSICS 61. 499.

518 ( 1985)

Finite Difference Method to Solve

Incompressible Fluid Flow

A method to solve incomprcssihlc fluid flow is proposed. ‘fhe method uses primitive
variables (velocity and pressure of rhc fluid) and introduces an idea that the discretizcd
Navicr-Stokes equations hax an invariant implicitly in each iteration at any time step As
numcrlcal examples. transient two-dimensional Poiseuille flow and steady flow past a
backward-facing step arc calculated. It is shown that the method needs fwer iterations than
the MAC and the SMAC methods. and the accuracy of the plesent method is guaranteed by
comparison with the analytic solution and the existing methods. ( 19X5 Academc Prw. Ini

Numerical solutions of Navicr-Stokes equations are extremely valuable in study-

ing fluid dynamics. Finite difference methods have been applied (1) where stream
function and vorticity arc used, and (2) where primitive variables, velocity and
pressure, arc used. In case (1) the equation of continuity is satisfied exactly by
introducing the stream function and convergence is rapid. However. for flows with
free surfaces, the boundary conditions present difficulties and, in three dimensions,
lack of a scalar stream function inhibits almost the solution [ 11. In case (2) boun-
dary conditions and three-dimensional geometry are easily handled. Unfortunately.
the velocity components do not automatically satisfy the equation of continuity nor
is the convergence rapid [2 91.
This paper presents a new method, using primitive variables. based on the idea of
an invariant implicit in each iteration at any time step (unsteady flow): or at any
iteration (steady flow). Although applicable to three-dimensional flows, the paper
will be restricted to two-dimensional flows.

The Navicr Stokes equation of a viscous incompressible fluid is expressed in

non-dimensional form as
500 NOBUMASA TAKEMITSU

using a velocity vector v (Cartesian components U, v, w), time t, pressure p, and the
operator nabla V. Here, Re = UL/v is the Reynolds number, U the characteristic
speed, L the characteristic length of the flow field, and v the kinematic viscosity of
the fluid. The equation of continuity is

D=V.v=O, (2)

where D is dilatation.
When we take the rotation of Eq. (1) the vorticity transport equation

(3)

is obtained, where o (=V x v) is vorticity. Velocity is obtained by

v=vxyl, 0 = -v=yl. (da, b)

Here, v is a stream function, which is a vector potential for three-dimensional
flows.
To obtain numerical solutions of Eqs. (1) and (2), or of Eqs. (3) and (4), requires
initial and boundary conditions. In both cases, the solution process is iterative. In
two-dimensional flows, \~r is a scalar function (see next section) and Eq. (2) is
always satisfied exactly. Thus fewer iterations are needed to solve Eqs. (3) and (4)
than to solve Eqs. (1) and (2).
Fewer iterations are necessary if some conditions are added implicitly in each
iteration when Eqs. (1) and (2) are solved simultaneously. For this purpose add an
implicit invariant in each iteration at any time step t = t.
When Eqs. (3) and (4) are solved iteratively using the forward-time and centered-
space finite difference method at the reference time step t = t, the value of the vor-
ticity is unchanged and Eq. (2) is imposed. Moreover, the vorticity remains
unchanged when Eqs (1) and (2) are solved iteratively, i.e., the vorticity is invariant
at t = t. This is expressed as
u)’ = ou) = o(2) = . . . = m(k) = m(k+ 1) = .. .
> (5)
where t denotes a time step at t = t, and (k) the number of iterations. Hereafter, t is
deleted. Equation (5) is rewritten as
O=VXp)=VX”(k+l), (6)

from which we have

(7)
where 4 is a potential function. If we take the divergence of Eq. (7)
 501

is obtained. Since the dilatation D’k * ” is zero. we obtain

‘Thus an invariant is introduced at any time step, if Eq. (9) is solved for 4 in each
iteration. The new velocity field at k = k + I is obtained from llq. (7). and the
pressure field at t = t + At (nr = discrctizcd time incrcmcnt) is obtained from llq.
(1 1.
In the SMAC (simplified MAC) method [4]. the tentative velocities are modified
to their final values so as to preserve vorticity at every point. Hence. the basic idea
of the SMAC method is that it introduces a tentative velocity field. On the other
hand. the present method is based on the idea that the discretized Kavier Stokes
equation has an invariant in each iteration in the same way as in the stream
function and vorticity method. The author believes that the simplest invariant is
vorticity. If the vorticity is taken as an invariant: the formulation is somcwha:
similar to that of the SMAC method cxccpt for the calculation of pressure.
However. because of the difference of the basic idea bctwcen the proposed method
and the SMAC method, Eq. (9) plays an important role in the present method. but
in the SMAC method the time increment At is important. It should also bc noted
that a potential function 4 is calculated by
&V?Q (101

in the SMAC method. Here, ij is calculated from tentative velocity components.

and b remains unchanged until the converged solution is obtained.
This new formulation requires less calculation time than the SMAC method, and
the steady-state equation can be evaluated.

First, transient two-dimensional Poiscuille flow is studied. The initial and boun-
dary conditions of this flow arc

t <o: u=c=$=(~=o (everywhere), (Ila)

! > 0: * = (1)= 0 at j’ = 0 (center line ). (1 lb)
II/= I, u=2$/aJ=o at ): = 1 (upper wall ), (112)

where u and c are velocity components in x and )’ directions, respectively5 $ is the

scalar stream function, and o is the vorticity. The analytic solution for this flow.
obtained by the Laplace transform, is
502 NOBI;XI.ASA ‘I-AKI-MITSIJ

- 1 exp( -pi fl&), (12a)

(12b)

(12c)

2 sin p,, y
w=3y+ f cxp( - /?,;t;Re),
,, _ 1 sin B,,

3 + 2 f exp( -/I: t!Re)

,I - I
where p,, is the nth positive solution of [I,, = tan /In (0 -C/I, <PI < . ). Numerically,
convergence will require, at most, 40 terms in the above series.
In two-dimensional flows, Eqs. (1) and (2) are written in the form
^. :
c/” + 25 2 + -duv = -.!i!- + - 1 V$,, (13a)
dt c?x l3y tx Re

(13b)

These equations are discretizcd as

u,2+1:.2.j-4 ,;2.,+%, bl,Z”! 1.‘2.,- I -u r., l,.Z”,- I.‘?./
u;,,!J’=u;,,+ At -
i( h s

-Pt.,-I’, I./ 14 1
h +Rc ph l.,+ll, -1.,-2u,,,)

1
+ 2 (lli.,+ I +",~, -I- 2ui.,J > (144

21I, I,,-n”,. 1:2.,--ui,,-I~2”! 1,‘2./

t’:.
a./+1;2 - c;, I:2
I:;.; n’ = I(, + At - +
i ( i s )

-1),.,-11),.,-l 1 1
+Re jp’“,, i.,+“;-,,-2”,,,)
s i
I
+ f (CL, , I + Ii,,+ , -- 2ZJ,/) .
II
 FINITE DIFFERENCE METHOL~ 503

The indices i andi count cell-center, cell-boundary, or cell-corner positions in the

x and y directions, and h and s are mesh sizes in the x and y directions, respectively.
The superscript indices t and t + At (At = time increment) show time steps. The
velocity components u and o are defined at cell boundaries; pressure p and
dilatation B are defined at cell centers; the stream function + and the vorticity CO
are defined at cell corners (see Fig. 1). The quantities at all fractional indices are
obtained as simple averages of the two adjacent quantities, e.g.,

u I + lj2.I = t”i,j + u, + l.,)!2t etc.

Equations (3) and (4) become

(15a)

(15b. c)

in the two-dimensional case ($-CJ method). These equations are discrctized as

(16b)

(16c)

Pij
Ui.j”
Di,j
l “i+l,j
I

FIG. I. Placement of field variables

504 NOBUMASA TAKEMITSU

Hem, we introduce a relaxation factor RF,k for $. Velocity components ui.i and u,,,
are calculated from

ui,j= (Ic/i.j+ I - Ic/i.,):Lys, vi,, = -(tii f I./ - ll/i,,)lh. (16d)

Solutions arc obtained by four methods; the ICI-0 method, the MAC method, the
SMAC method, and the present method.
In the MAC method, the pressure field is obtained by solving

+ PI., -I)

p;,;.+ ‘) = p;.“,’ + RF,,(p;, - p;,;‘),

where RF,, is a relaxation factor for p.

In the SMAC method: the tentative velocity ?(ii. a) is calculated from

and velocity components at t = t + dt are corrected by

+ d,., - I) - iji,jT

4j.T.’ ‘) = qSr;:’+ RF,(#:i - #;,:‘)T

V ‘+Ar=3.-vq$,

where RF, is a relaxation factor for 4.

In the present method, v ’ +” is calculated from Eqs. (14a), ( 14b), and v’ +” is
corrected by
#k)=v.v’k’ (19a)

(19b)

/$I;,; ’ ‘) = #j,;.’ + RF,(+,T, - d;,;,),

v’+dr(k I I)=v’.4 nr(k)-vp 1 I),

where RF, is a relaxation factor for 4.

 FII\;ITE DIFFERENCE METHOD 505

Let Re = 5 and divide the flow field as shown in Fig. 2. Mesh sizes /I and .v are
h=s=O.l, and fC=61. JC= 13, and the time increment .4t= I/100.
The initial values for u, u, p, $, and (L) are all equal to zero, and the boundary
conditions are:

(a) Upstream boundary

(1) u, I/I: w: given by the analytic solution

(2) L.=o:
(3) p=o.
(4) 4,. l.l=4i., (i=2).

(b) Upper wall

(1) (1/: given by the analytic solution,

(2) wi,!: calculated by ~($,,-I/I~.,. ,)/!.Y’ (j=JC- I).
(3) I(,,,+~ = -u,,, (j=.lC--2),
(41 l:=ol
C5) Pl,j+ I = p,,, + (2/.~sRe)(v,., -- c,.!- , j3 where t:,., , ? = L‘,,, and D,,,, I = D,,,
(j=JC- 2, MAC method),
(6) 48.,, 1= +hi,,(j = Jc’- 2).

(c) Lower boundary (center lint)

(1) l//=0=0.
u , =u,, G=2),
151 c “’= 0,
14) tit., I = d,., (j = 21,
(5) p,,! -,=p,,, (j=3, MAC method),
(6) c,., I = --t: r.,.+I (j= 2, MAC method).

FIG. 2. Discretion of the flow field. B boundar) cell. open box - fluid cell
506 NORUMASA TAKEMITSU

(d) Downstream boundary

( 1) a211//ax2= 0,
(2) acu/ax=o:
(3) &/dx=O,
(4) u,,,~j=U;,j-(h~~)(t:,,+, -c,,j) (i=IC- l),
(5) 4=0,
(6) d2p/dx2 = 0 (MAC method).
In the SMAC and the present method if the condition

4, I., = di,,
is imposed at the upstream end: then

#i+ I,/
A+; (iv,=,, + f (d;.., I I + di,, I) - nt,,
( /)
is used instead of Eqs. ( 18~) or (19b).
The numerical procedures per cycle arc:
(i) I+-o method

v
t=t+nt
w’ 0, 1A/ - $k,\.fT I)’’ A’
V’ Eq. (16a) 1 Eq. jlhc)l--<:;;;‘-i;q. (20)

where suffix B.C. denotes boundary conditions.

(ii) MAC method
 FINITE DIFFERENCE METHOD 507

(iv) Present method

V’

t=t+nt

The main calculations proceed from left to right, and the notation +- denotes
iterations. In the above numerical procedures, Eqs. (16b), (17a). ( l&z), and (19b)
are all solved by the SLOR method [ 101, the relaxation factors arc RF, = RF,, =
RF, = 1.8 in the I(/-0 method, the MAC method, and the SMAC method, and
RF, = 1 in the present method. In the I/-W method and the present method, the
pressure field at new time step is calculated by

r’v r
(VI)1I I Al_-- --v (vv)’ $ ; WV)‘, (2Oj
05
where (?v/Zt)’ is evaluated by simple forward differences. In this problem, pressure
is integrated from left to right at .r = I - s:2 and from top to bottom at s.
Figure 3 shows the convergence rate by various methods. In the figure. ;:, is
defined by

FIG. 3. Typical demeanor of convergence by various methods.

508 NORUYASA TAKEMITSU

where I Imaxdenotes maximum value of ( (, and J‘stands for I/ (I/-W method). p

(MAC method), or 4 (SMAC and the present method).
The velocity profiles at t = 2~tt are shown in Fig. 4. Only the solution by the
MAC method is quite different from the analytic solution. The MAC method
represents fluid driven by a pressure gradient (see Eq. (17a)). Therefore, when the
fluid starts impulsively from rest to a finite velocity, the solution by the MAC
method is not accurate in the whole flow field. In the MAC method, maximum
error of dilatation Ed was 10.5, where cLj is defined by

(22)

The solution by the SMAC method, while accurate (8,) = 1.1 x 10 2), required
many iterations to obtain convcrgcncc (see Fig. 3). The solutions by the present
method are very close to the analytic solution. When we stop iterations with the
condition that sr, is 0( 10V2), the velocity profile is very close to the one by the I,-(,>
method, and if we stop iterations when so is O(lO-“), the solution agrees almost
precisely with the solution by the I+-(9 method.
However, the convergent process is not easy to analyze and is not always stable
in the present method. Therefore 4 is set to zero when ~$1 becomes larger then
cj$ ‘1, or when cr,tk) becomes larger than I;{:-‘) and cy) becomes larger than ~2. I).
In this problem, previous values of 4 are used (cb < 10 -2), or 4 is set equal to zero,
when EE) becomes larger than cg ‘) (or) < 5 x 10 “). (See also Section B, Steady
Flow.)
Figure 5 shows velocity profiles by the four methods at t = 0.5 (at downstream
end). Again only the solution by the MAC method is not close to the analytic
solution. More time steps are necessary to make E,) small; E,, is still 5.2 in the MAC
method.

Y
Present (E0=4.5xlO-~)
1.0

0.8

0.6

0.4
P j
MAC
(Eo’10.5) SMAC kD=l.lxlO-*)

0.2

0 0.25 0.5 0.75 1.0 ”

FIG. 4. Vclocit;y profiles by various methods at downstream end (IL 0.02).

 FINITE DIFFERENCE LfWHOD

1.0

0.8

0.6

0.4

0.2

FIG. 5. Velocity profiles by various methods at down stream end (t = 0.5).

In Fig. 6, global error of convective terms, defined by

(23)
/=Z.J<‘ 2 ,= J,JC 2

is shown. The error by the I+!-o method, the SMAC method, and the present
method dccrcascs as time increases. The magnitude of the error by the SMAC
method is the same as that of the $ (9 method. Although I:~ is shown increasing
with time in the MAC method, it will eventually decrease.
Figure 7a shows the pressure distribution along the center line at t = 0.5. Only
the MAC method is quite different from the analytic solution.
Figures 4, 5, 6, and 7a indicate that the SMAC method is more accurate than the
present method since cp is small. However, this is not true in general. Figure 7b

lo2 - MAC
cn -...I-. ---
w .I--
- .A

103V 0 0.2 t

FIG. 6. Temporal global error cg by various methods.

510 NOBUMASA TAKEMITSL

FIG. 7a Pressure distribufions by various methods along cc~llerline (I :: 0.5).

-1

-2

FIG. 7b. Temporal pressure gradient at .x = 3 on center line by various methods.

 FINITE DIFFERENCE WITHOD 511

TABLE I

Comparison of the Number of Iterations by Various Methods

Criterion of Number of (‘PU time (appros. i

Methods convergence iterations (xc)

WJ El < 2.0 x 10 ‘l 232 4

MAC E,,<2.0X IO 4 6678 34
SMAC Eg<2.0X 10. J 5869 26
Present i:,) < I.0 x IO- z 314 5
I:,) < 5.0 x 10 j 506 6

shows temporal pressure gradients at x = 3 on center line by various methods. Only

in the present method, with zr, < 5 x 10 3, and in the $-a method is the pressure
gradient calculated accurately even for small time values. The SMAC method shows
severe oscillation, and the MAC method is inaccurate (see also Fig. 7a).
Table I shows number of iterations and CPU times by various methods. The
solution by the t+!+o method is very fast and is accurate. The number of iterations
required for convergence by the present method is about 1.5 times that by the I/,- CL)
method, and about l/20 times smaller compared with the MAC and the SMAC
methods. Moreover? the solutions by the present method are accurate.

As a simple example which has a corner in fluid, a steady flow past a backward-
facing step is studied (see Fig. 8).
In steady flow problems, the Navier-Stokes equations are written in the form

(24b)

(24~)

FIG. 8. Backward-facing step flow

512 NOBUMASA TAKEYITSU

and
i?uo

2X
+ 2 = ; v20, (25a)

o= -V'$, u = aI+bji?y, v = -al// jax. (25b, c)

Since the iterative solution of vorticity often diverges at higher Reynolds num-
bers, methods to stabilize the calculation of vorticity should be used. ln the present
calculation, the author’s method [ 1 l] is modified so that the vorticity obtained
from Eq. (25) is the same as that obtained from Eqs. (24). This method has second-
order accuracy in the whole flow ticld and avoids iteration divergence. Similar
stabilized finite difference equations, which arc different from second-order
upstream difference equations (see, for examples, Ref. [ 12]), have been previously
published [ 13-153. In this problem we use the finite difference equations

ui,j + I;2 vi- l!2. j - 1 - u!,J I/Z’1 - 1:2,1

)+P(k+f)Ui,,
s

-pr*J-p’- ‘.J+k $ tu;; I,j + ui- I./)+ f t”i,J I 1 + ui.j- I )}* Pa)
h
#.+ 1) = Il(k) + RF (u? - U(kj)
1. I 1.J ” ‘.I r,, ’ (26b)

and

1 1
( >
- v’, .
I,2
+P ;+; fii.,
s >

-Pi.j- Pi,j- 1
+A ~CL:i+l,)+O, ,,J)+$.j+~+C.i.J-,) (26~1
s i

I:,,;- ‘1 = c!(J’ + RF,(v,T, - vj,;‘), Wd)

(k+ 1)
D(k + 1) = ui 8 I.j- ut.j + ci.~-I I - vi.,

1.J (26e)
i h s > ’

where ,u, a relation factor, stabilizes the iterative process, and RF, and RF, arc
relaxation factors for u and v, respectively (present method).
 FINITE DIFFERENCE METHOD 513

In the I,-u method, Eq. (25) is discretizcd as

(27b)

where RF, is a relaxation factor for w.

The case of Re = 50 is studied and mesh sizes h and s are chosen as h = s = 0.1. As
is shown in Fig. 8, the length of the channel is taken as 6, and the height as 1.4, and
the backward-facing step is located at x = 2 and its height Sh is 0.4, and L., denotes
the length of recirculating region.
Initial values are set equal to those from steady Poiseuille flow,

u= -6$+6(2~++1)~-6~&+1). c SE0, (28a)

II/= -2y3+3(2y, + l)y’-6~,(~, + 1) y+y:(2yI+3), (28b)

co= 12y-6(23’, + 1), (=I

where y, = 0.4. Pressure equals zero in the whole flow field. Boundary conditions
are imposed at the upstream end in the same form as Eqs. (28a), (28b), (28~):

ui., - I = -.ui,, (j= 2) or u ,.,-, = -u,.~ (.i=JC--21,

c = 0, (28d)

$=O or *=1 (28e)

on the wall.
In the present method, 4 is calculated by the equations

instead of Eq. (19b) at the corner C, since 4 is a multi-valued function (see Fig. 9).
514 NJBUMASA TAKEMITSU

Wa .II Wall

FIG. 9. Corner cell. R = Boudary cell, C = corner cell, open box = fluid cell.

Two calculation procedures follow:

(i) 3 -cr)method

* (k’
C!jCk’ $‘k i I’
k=k+l
__ #k+l’ (k-I 1)
P
“(k’ LL)(k+ 1)
I
Eq. (16d) Eqs. (24a), (24b)
(k’ Eqs. ( J~c), (27b)
p I i

(ii) Present method

$k’
k=k+ 1
#k+ 1) -4 (k-t 2) __ #kt-l’ -
P Eq. (26e) Eq. (19~) Eq. (19d)

The relaxation factors RF,, RF,,,, RF,, RF,., and p are all taken to be 1.
Table II shows the comparison of number of iterations and errors ,sf (.f-
tj, CO,U, v), cr, (Eq. (22)), and cM which is defined by

E M = /left-hand side of Eqs. (24a), (24b)

-right-hand side of Eqs. (24a), (24b)i,,,,
or

= Jlcft-hand side of Eq. (25a)

-right-hand side of Eq. (25a)i,,,.
 TABLE II
a
1’.
Comparison of the Number of Iterations, Errors c, (.f $, (1) or U, I;),
E,,, and cM, and Vortex Length I,, by Various Methods :
- _-- w
Numhcr of 5
Methods Iterations E, C” CM -LX Comments z
TIT
.- -. .- . _ -. - - -. - - - --
- 3
* (IJ 328 1.979 x IO -4 2.169x 10 3 0.765 - T
1x32 2.665 x 10.. 3 3.328 x IO.-’ 0.770 Time-marching method (500 steps)
Present I 600 3.363 x 10 .J 7.915 x IO ’ 3.576x IO-’ 0.580 4 - in each 5
0 iteration
11 553 4.718x IO 4 9.938 x lo- 6.287 x 10. ’ 0.746 rrt:p>f:i:-“.q5 2
-0
III 347 2.884 x IO / 9.875 x IO-.’ 3.885 x 10. z 0.724 Ifcl:l>(:~-‘)andc~k’>r:::-“.cb-O 8
.___.- --. .-.--- _.- .- _..
__- .-..___- -..- -..- -. .~___..
516 NOBUMASA TAKEMITSU

FIG. 10a. Velocity vectors obtained by the present method III. Unit velocity vector = + .

The $-o method is the fastest one to obtain a solution in steady flow. But the time-
marching method by the $-w method (calculated by Eqs. (16a), (16~) and At=
l/100) needs more time to obtain a converged solution, because E, is larger by one
order compared with the I,&u method (see also Fig. 11). Too many iterations are
needed to obtain convergence in the present method I, where 4 =0 in each
iteration; but the iteration process is the most stable. If we put 4 = 0 when .sg) >
&g-i) (present method II), the iteration process is less stable but .sD is smaller by
one order compared with the present method I with almost the same iteration num-
ber. If we put 4 = 0 when sg) > EK- l) and sr) >&g-i) (present method III), the
number of iterations is almost the same as that by the I+-u method, and the
solution obtained is accurate if the iteration process remains stable.
Velocity vectors and stream functions obtained by the present method III are
shown in Figs. 10a and b, respectively.
Figure 11 shows vorticity distributions on the wall by various methods. Solution
by the time-marching method has not yet converged in steady state because the
absolute values of the vorticity on the wall at the downstream end are not nearly
equal. Solution by the present method III is very close to the one by the t/-o
method.
From the table and figures, it is considered that the present method is accurate
and needs less time to obtain solution for steady flow, too.

FIG. lob. Streamless obtained by the present method III.

 6
3
4

-6

FIG. 1 I. Vorticity distributions on the wall obtained by various methods. 1 L rj (1). 2 - i-o (time-
marching method). 3 = present method III.

CONCLUSIONS

A new method, in which an idea that the Navicr- Stokes equations have an
invariant implicitly in each iteration at any time step or at any iteration is applied,
is presented. Using this method, some numerical experiments were performed. As a
result of this study, we conclude:
(1) Navier-Stokes equations with an invariant implicitly in each iteration
produce cfticient and accurate numerical solutions.
(2) The proposed method is also applicable to steady flow and three-dimen-
sional flow problems.
The method proposed here has also been extended to the Crank--Nicolson
method. a non-iterative implicit method, and a two-step method [16].

The author wishes to express his sincere gratitude to Professor M. Kawaguti and Prof. Y. Matunobu
of Kcio University, and to Professor S. Mizuki and assistant engineer N. Shinya of Hosei University.
Examples of unsteady flow were calculated by FACOM M-380R at Keio Unjversity and the steady flows
were calculated by ACOS 6 computer at Hosei C’nivcrsity.

REFERENCES

I. A. THOhl, hoc. R. Sot. London, Ser. A. 141 (1933) 651.

2. F. H. L~AKLOW ANI) J. E. WIXCH, Ph~s. Fluids 8 (1965), 2182; J. E. WELCH, P. H. HAKLOW. J. P.
SHANNON, ANI) B. J. DALY, Los Alamos Scientific Report LA-3425, 1966.
3. A. J. CHOR19, Murh. c‘omp. 22 (1968). 745.
4. A. A. AMSDES AZD F. II. HARLOW, Los Alamos Scientific Report LA-4370, 1970.
518 NOBUMASA TAKFJMITSU

5. W. E. PRACHT, J. Comput. Phys. 7 (1971), 46.

6. J. E. VIECELLI, J. Comput. Phys. 8 (1971), 119.
7. H. TAKAMI AND K. KUWAHARA, J. Phys. Sac. Jpn. 37 (1974), 1695.
8. C. W. HIRT, B. D. NICHOLS, AND N. C. ROMERO,Los Alamos Scientific Report LA-5852, 1975.
9. C. W. HIRT AND B. D. NICHOLS, J. Comput. Phys. 34 (1980), 390.
10. R. S. VARGA, “Matrix Iterative Analysis,” p. 199, Prentice-Hall, Englewood Cliffs, N.J., 1962.
11. N. TAKEMITSU, J. Comput. Phys. 36 (1980), 236.
12. B. P. LEONARD, in “Computer Methods in Fluids” (K. Morgan, C. Taylor, and C. A. Brebbia, Eds.),
p. 159, Pentech Press, London/Plymouth, 1980.
13. S. C. R. DENNIS AND G.-Z. CHANG, Phys. Fluids, Suppl. IZ, (1969), 88.
14. S. OZAWA, J. Phys. Sot. Jpn. 38 (1975), 889.
15. H. FASEL, in “Computational Fluid Dynamics” (W. Kollman, Ed.), p. 167, Hemisphere Publishing
Corporation, Washington/New York/London, 1980.
16. N. TAKEMITSU, submitted for publication.