Professional Documents
Culture Documents
INTRODUCTION
Over the last two decades important advance s in computational fluid dynamics (CFD), pertaining to the developme nt and maturity of solution algorithms, have
been achie ved. In this work, a solution algorithm, such as the SIMPLE w1, 2x
algorithm, denote s the procedure used to resolve the coupling that arise s in the
solution of Navie r-Stoke s equations between velocity, density, and pressure . Many
difficulties conne cted with these algorithms have been resolve d and bette r insight
gaine d. In specific, work has been directed toward settling a numbe r of pending
issues related to the choice of primitive variable s (density-based versus pressure based), the grid arrangeme nt (staggere d versus collocate d arrange ment ), and the
solution approach (semidirect versus segregate d approach ). Developments in these
areas are briefly reviewed next. In the choice of primitive variables, density-base d
algorithms, in which the continuity equation acts as an equation for density while
pressure is obtained from the energy (enthalpy ) and state equations, have been
used successfully in the simulation of highly compre ssible flows. For low Mach
numbe r flows, howeve r, as small disturbance s in density result in large variations in
the pressure field, density-based algorithms become unstable and their conve rgence rate gre atly diminishes. Moreover, despite the extension of this class of
algorithms to predicting incompressible flows through the use of the so-called
pseudo or artificial compressibility technique w3 ] 5x , the difficulties encountere d in
efficiently avoiding the stiff solution matrices generate d by these methods have led
Received 1 April 1999; accepted 2 June 1999.
The financial support provided by the University Rese arch Board of the American University of
Beirut is grate fully acknowledged.
Address correspondence to Dr. F. Moukalled, Me chanical Engineering Department, American
University of Beirut, P.O. Box 11-0236, Beirut, Lebanon. E-mail: memouk@aub.edu.lb
103
104
NOMENCLATURE
afP , a fE ,
5 a, b 5
b fP
Cp
d f
D wf x
D w f x
Dwf x
D w f x
a
b
d t
g
D wf x
f
f
underrelaxation factor
thermal expansion coefficient
time step
scaling factor
diffusion coefficient for f
the D operator
dependent variable
scalar value at cell face f
dissipation term in energy equation
space ve ctor equal to (n f y g d f ) S f
viscosity
Ff
ff
H wf x
Hwf x
H w f x
H w f x
i
j
P
Qf
R
RM
s f
Sf
T
t
u, v
Uf
V
v
x, y
density
Subscripts
e, w, . . .
f
F
P
Superscripts
C
D
(n)
a number of researche rs w6 ] 10x to work on exte nding the pressure -base d algorithms, originally developed to solve incompre ssible flows, to this class of flows,
thus encompassing the entire subsonic to hype rsonic spectrum.
With pressure-based solution algorithms, two different approaches have been
followed, which are denote d in the literature by the semidirect approach and the
segregate d approach, respectively. In the semidirect method, first proposed by
Caretto et al. w11x , the discretized forms of the momentum and continuity equations are linked toge ther and the resulting system of equations is solve d simultane ously. This technique guarante es a close connection between the velocity components and the pressure , which enhances the convergence rate and hence the
105
efficiency of the method. However, the memory required to store the various
coefficients at all grid node s is ofte n prohibitive , particularly for multidimensional
and multiphase flows. This storage problem can be alleviated by subdividing the
domain into parts and solving the syste m of equations ove r each subdomain
separate ly. These so-calle d subdomain solve rs, which degenerate to point solve rs
when the domain is confine d to one volume , suffer a conne ctivity proble m and
require iterations among the various parts of the domain, leading to a drastic loss
of the converge nce rate w12 ] 14x . In the second, much more popular segregate d
approach, the discretized forms of the various differential equations are solve d
separate ly, but over the whole domain. This has the advantage of requiring
conside rably less compute r storage than the semidirect method in addition to
providing the flexibility of easily solving additional partial diffe rential equations
(such as turbule nce kinetic energy, turbule nce dissipation rate, concentration of
chemical species, etc.) when needed. In this technique, while the velocity components are obtained from the corresponding momentum equations, there is no
apparent equation governing pressure . To derive a pressure or an equivalent
pressure -corre ction equation, the discretized forms of the continuity and momentum equations are combine d toge ther. More over, for compressible flows, density is
replaced by pressure through the equation of state . For an incompre ssible flow,
this pressure or pressure -correction equation is elliptic, while for a compre ssible
flow it is hyperbolic. Once the pressure (or pressure -correction ) field is calculate d,
the velocity and density fields are update d to satisfy the continuity equation.
Hence, in compre ssible flow simulations the pressure plays a dual role by acting on
both the density and the velocity through the equation of state and momentum
conservation, respective ly, so that mass conservation is satisfie d. Obviously, for
low-speed flow, the pressure gradient required to drive the velocities through
momentum conservation is of such magnitude that the density is not significantly
affe cted and the flow can be considered ne arly incompre ssible. In the hypersonic
limit where variations in velocity become relatively small as compare d to the
velocity itself, the changes in pressure do affe ct density significantly. In fact, in this
limit, the pressure can be viewed to act on density alone through the equation of
state so that mass conse rvation is satisfie d w6x . This dual role played by the pressure
field explains why pressure -base d algorithms have been applie d with success to
both incompre ssible and compressible flows.
The cell-centered or collocated variable scheme has been used quite successfully with density-base d algorithms. With pressure -based algorithms, howeve r, the
use of this grid layout is not as straightforward. While the momentum equations
link the velocity to the respective pressure gradients, the continuity equation,
apparently having no direct link to pressure , acts as a constraint on the velocity
field. Consequently, the conve rgence and stability of pressure-based algorithms
depend large ly on how the pressure gradients and velocities are evaluate d in the
continuity and momentum equations. Early attempts to use collocated variables in
pressure -base d algorithms failed to give converged solutions, producing spurious
oscillations, and resulting in the so-calle d red ] black checkerboard splitting of the
pressure field w1x . This unde sirable behavior stems from the linear interpolation
practice used in evaluating the velocity at the control-volume faces in the continuity equation, which cause s the velocity there to be related to the pressure differ-
106
ence between two alternating, rather than consecutive, nodes. The use of staggere d
grids, first propose d by Harlow and Welsh w15x , in which pressure is store d at the
control-volume center and velocities at the cell faces, removes the need for
interpolation of pressure in the momentum equations and of velocity in the
continuity equation. This is a conside rable advantage and hence stagge red grids
became very popular and have been used with gre at success to solve a wide range
of proble ms in Cartesian, polar, and also more general orthogonal coordinate s.
However, the primary disadvantage of the stagge red grid arrangeme nt is the
greate r geometric, and related mathe matical, complexity associate d with the use of
different grid syste ms for the various variable s, which becomes ove rwhelming in
curviline ar coordinates. Another disadvantage of the staggere d-grid approach is
related to the choice of the contravariant, covariant, or even Cartesian velocity
components at the control-volume faces w16x . A better solution to avoid the
checkerboard splitting of the pressure field was proposed independently, at the
beginning of the 1980s, by several worke rs through the use of a special interpolation proce dure for evaluating the control-volume face velocities. Depending on the
path followed, two collocated methods have been devised, which are denoted in the
literature as the pressure -weighted interpolation method w17 ] 20 x (PWIM) and the
momentum-we ighte d interpolation method w21 ] 25x (MWIM), respective ly. Both
methods reconstruct the momentum equations at the cell faces by interpolating the
coefficients at the cell centers.
It is apparent that while decisions on the above-mentione d points at one time
seemed difficult, the direction followed by the CFD community (in conne ction with
the finite-volume method, FVM) is now much more distinct. Although the semidirect versus segregate d approach dichotomy is still not clearly resolve d, especially
with the incre asing use of multigrid techniques w8, 26 ] 29x , it is now widely believed
that collocated pressure -base d solution algorithms for predicting fluid flow at all
speeds are optimum in the use of resource s. More over, the segregated approach
when combined with multigrid acceleration techniques, at least for the near future,
still represents the most efficient approach.
Within this context, a large numbe r of solution algorithms w1, 2, 30 ] 37x have
been propose d. The first of these algorithms was the SIMPLE (Semi-Implicit
Method for the Pressure-Linked Equation ) algorithm of Patankar and Spalding w1,
2x , developed in the late 1960s and beginning 1970s. Over the years, a number of
modifications to the SIMPLE algorithm have been suggested with the aim of
improving the robustne ss and r or conve rgence rate . The SIMPLER (SIMPLE
Revise d) algorithm of Patankar w1x , propose d in 1981, the SIMPLEC (SIMPLE
Consistent) algorithm of Van Doormaal and Raithby w30x , proposed in 1984, the
SIMPLEST algorithm (SIMPLE ShorTened) of Spalding w31x , the SIMPLEX
algorithm of Van Doormal and Raithby w32x , the SIMPLEM algorithm (SIMPLEModified) of Acharya and Moukalled w33x . The PISO (Pressure Implicit with Split
Operator ) algorithm of Issa w34x , and the PRIME (PRessure Implicit Mome ntum
Explicit) algorithm of Maliska and Raithby w35x are the most important atte mpts to
improve on the original solution algorithm. Sometimes minor changes were introduced to these algorithms, such as in the MSIMPLEC w36x , MPISO w36 x , SIMPLESSEC w37x , and SIMPLESSE w37x algorithms.
These SIMPLE-like algorithms, scattere d in the literature using diffe rent
107
notation and appe aring to a nonversed CFD user as being unrelated are in fact
very similar and share the same essence in their derivations. Thus, the main
objective of this work is to derive these algorithms using a unified, compact, and
easy-to-unde rstand notation that can be expande d syste matically to yield the
coefficients of the pressure -corre ction equation, hence facilitating the implementation of these algorithms for new CFD users. In the proce ss, the philosophie s
behind these algorithms in addition to their similarities and diffe rences are
explained. Moreover, because they were originally developed for incompre ssibleflow calculations, the majority of these algorithms has not yet been extended to
simulate hypersonic flows (compre ssible ), even though such extension has been
successfully applie d to the original SIMPLE solution algorithm w6, 8, 10, 17 x .
Therefore , a second objective of this work is to exte nd all the above -mentioned
algorithms to solve for fluid flow at all speeds, and to demonstrate in so doing that
the incompre ssibility limitation is not inherent to these algorithms.
In what follows the gove rning equations for compre ssible flow are presented
and their discretization outline d so as to lay the ground for the derivation of the
pressure and pressure-correction equations, which are developed using a unified
notation and shown to be equivale nt mathematically. Moreove r, the use of a
unified notation also helps in understanding the importance of the diffe rent
influences affecting the conve rgence prope rties of the various algorithms. However, it should be stressed that the purpose of the article is not to compare the
relative performance of the different SIMPLE-like algorithms; rathe r, the aim is to
unify their formulation.
GOVERNING EQUATIONS
The equations gove rning the flow of a two-dime nsional compressible fluid are
the continuity equation, the momentum equations, and the energy equation. This
set of nonline ar, coupled equations is solve d for the unknowns r , v, T , and P. In
vector form, these equations may be written as
r
t
( r v)
t
(r T)
t
s
q = ? ( r v) s 0
( 1)
q = ? ( r vv) s y = P q = ? ( m = v) q
1
3
= ( m = ? v)
( 2)
q = ? ( r vT )
1
cp
= ? (k= T ) q b T
P
t
q = ? ( P v) y P = ? ( v) q F
q q
( 3)
where
F
sm
( ) () (
u
x
u
y
v
x
2
3
( = ? v) 2
( 4)
108
r s
RT
( 5)
s Cp P
(r f )
t
q = ? ( r vf ) s = ? ( G
= f ) q Qf
( 6)
where the expre ssions for G f and Q f can be deduced from the parent equations.
The four terms in the above equation describe successively unste adine ss, conve ction (or advection ), diffusion, and generation r dissipation effects. In fact, all terms
not explicitly accounte d for in the first thre e terms are included in the catchall
source term Q f .
a fNB f
NB
q b fP
( 7)
NB ( P )
In the above equation, a fNB are the coefficients multiplying the value of f at the
neighboring node s NB (P) s (E, W, N, S) surrounding the central node P, a fP is the
coefficient of f P , and b fP contains all terms that are not expressed through the
nodal value s of the dependent variable . An equation similar to Eq. (7) is obtaine d
at every grid point in the domain and the collection of these equations forms a
syste m of algebraic equations that is solve d to obtain the f field. Since the
coefficients in these equations are in general dependent on f , an iterative solution
scheme should be employed for handling this nonline arity. During the iterative
process, it is often desirable to slow the changes, from iteration to iteration, in the
109
sf
U
P
(
p
f
NB(P) a NB f NB
a fP
q b fP
yf
U
P
( 8)
110
In the above equation, the unde rlined term represents the change in f P produce d
by the current iteration. This change can be varie d by introducing and unde rrelaxation factor a (0 F a F 1), so that
sf
qa
U
p
f
NB(P) a NB f NB
a fP
q b fP
yf
U
P
( 9)
or
a fP
a fNB f
NB
q b fP q ( 1 y a )
NB ( P )
a fP
U
P
( 10 )
At the state of convergence , f PU and f P are equal and the original equation
is satisfie d. There are no general rules for choosing the optimum value of a , and a
suitable value for a given proble m is usually found from exploratory computations.
Equation (10) can be rewritten in the form of Eq. (7) by redefining a fP and b fP as
follows:
If
aP
a fP
b fP b fP q ( 1 y a )
a fP
( 11 )
U
P
111
simultane ously and conve rgence is achie ved by repe atedly applying the abovedescribed procedure .
Before deriving the pressure or pressure -corre ction equation, the discretized
momentum equations are first written in the following notationally more suitable
form:
u
a NB
u NB q b Pu y V ( = P ) P ? i
a Pu u P s
p
NB ( P )
a vP v P
a vNB v NB
b vP
( 12 )
y V ( = P )P ? j
NB ( P )
w 5 w
uP
vP
H wuxP
H w vxP
5 w
D wu xP
0
s y
0
D w v xP
5w
( = P )P ? i
( = P )P ? j
( 13 )
where
( = P )P s
H
V V
= P dV
H wf xP s
f
NB (P) a NB f NB
a fP
q b fP
D wf xP s
V
a fP
( 14 )
In the above equations, V is the volume of cell P, and the subscripts e, w, n, and s
refer to value s at the east, west, north, and south faces of the control volume
(Figure 1). Defining the vector forms of the above operators as
H w vx P s
H wuxP
H w vxP
( = P )P s
( = P )P ? i
( = P )P
s
y
( = P )P ? j
( = P )P
DP s
D wuxP
0
0
D w v xP
( 15 )
( 16 )
For later use, modifie d forms of the H and D operators are defined as follows:
H w f x P s
f
NB (P) a NB f
a fP
NB
H wvx P s
H w u x P
H w v x P
D P s
DP
1 y H w1 x P
( 17 )
112
obtaine d as (Figure 1)
(r
t
P
dV q
H=
? ( r v) dV s 0
yr
o)
P
d t
V q
( 18 )
Es ( r
v) dS s 0
yr
o)
P
d t
(r
yr
d t
V q D w r v ? S xP s 0
o)
P
( 19 )
V q D w r U xP s 0
where
Uf s vf ? S f
and
D wf xP s f
qf
qf
qf
( 20 )
For the calculation of the mass fluxes across the control-volume face s and for
checking mass conse rvation, the value s of the velocity compone nts are needed
there. In orde r to avoid oscillations, which may result if a simple linear interpolation is used, a special interpolation practice has to be employed. In this work, the
MWIM is followed w23 x . The basis for the interpolation proce dure are the discretized momentum equations at the control-volume centers, as given by Eq. (16),
where the pressure source term has been taken out of the Q f term and shown
explicitly. To evaluate velocities at the control-volume face f, terms in Eq. (16) are
selectively interpolated and evaluate d at the f location to yield
vf y H wvx f s y D f ( = P ) f
( 21 )
where the ove rbar denotes a linear interpolation between the two control volumes
straddling the surface f. This equation, very similar to Eq. (16), may be viewed as a
pseudo-momentum equation at the control-volume face . Since the mass fluxes are
needed at the control-volume face s, rather than the velocity compone nts, it is the
value of Uf that should be calculate d there. Thus,
Uf s vf ? S f s H wvx f ? S f y D f ( = P ) f ? S f
( 22 )
( 23 )
H w vx f s f f H w vx P q ( 1 y f f ) H w vx F
where f f denote s the interpolation factor between the main grid points P and F.
More over, the term ( = P ) f ? S f is discretized using the method described by Zwart
113
( g d ) f
q ( = P ) f ? n f y ( g d ) f
)S
( 24 )
where (= P) f is the ave rage of the adjacent cell pressure gradie nts n f and d f (Figure
2) are the contravariant (surface vector ) and covariant (curvilinear coordinate ) unit
vectors, respectively, and g is a scaling factor. This factor is chose n such that it is
equal to 1 on orthogonal meshes in orde r for the method to collapse to classical
stencils w38, 39 x . With that constraint, the expression for g on structure d meshes is
1
n f ? d f
Sf d f
( 25 )
Sf ? d f
as
s n f y ( g d ) f S f s k
x
f i
qk
y
f j
( 26 )
114
( = P ) f ? S f s ( = P ) f ? ( d ) f
S f q ( = P) f ? ( k
Sf ? d f
x
f i
qk
y )
f j
( 27 )
In this form, the term ( = P ) f ? (d ) f represents the pressure gradient in the direction
of the coordinate line joining P and F (see Figure 2). Therefore, the above
equation can be rewritten as
( = P )f ? S f s
PF y PP S f d f
Sf ? d f
df
S f q (= P) f ? ( k
x
f i
qk
x
f i
qk
j)
( 28 )
j)
( 29 )
( = P ) f ? S f s ( PF y PP )
Sf ? d f
q (= P)f ? ( k
Substituting the various terms in Eq. (22) by their equivale nt expressions, the
following is obtaine d:
Uf s w H w u x f
H w v x f x ? w S fx
D w uxf
(k
(= P ) f ?
D w v xf
S fy x y
x
f i
D wuxf
D w v xf Sf ? d f
0
qk
Sf ? Sf
( PF y PP )
j)
( 30 )
or
Uf s ( H w u x f S fx q H w v x f S fy ) y
x
w Sf
S fy x
D wu xf
S fx
D w v xf
S fy
x
1
Sf ? df
( PF y PP ) y
D wuxf
D w v xf
(= P ) f
y
(= P ) f
? wk
x
f
y
f x
( 31 )
Upon expanding, the final form of the interface flux velocity is written as
2
Uf s ( H
w u x f S fx
qH
w v x f S fy
x
y D w u x f (= P) f k
x
f
)y
D w u x f ( S fx ) q D w v x f ( S fy )
S fx d fx q S fyd fy
y
q D w v x f ( = P) f k
y
f
( PF y PP )
( 32 )
( = P) f s f f ( =
P ) P q (1 y f f ) ( = P ) F
( 33 )
115
and the pressure gradient at the main grid point P (or F) is obtained from
= PP s
s
H=
V
P dV s
1
V
E P?
V s
dS s
1
V
( Pe S e q Pw S w q Pn S n q Ps S s )
( Pe S ex q Pw S wx q Pn S nx q Ps S sx ) i
1
( Pe S ey q Pw S wy q Pn S ny q Ps S sy ) j
s ( = P )P i q ( = P )P j
( 34 )
The cell face velocities are thus made dependent on the pressure difference
across the face , which helps avoiding the checkerboard proble m previously encountered in collocated variable algorithms (Patankar w1x ).
IP s P ( n) q P9
v s v* q v9
r s r ( n) q r 9
( u s u* q u9 , v s v * q v 9 )
( 35 )
( 36 )
( 37 )
Subtracting the two sets of equations (37) and (36 ) from each other yields the
following equation involving the corre ction terms:
X
vP y H wv9 x P s y D P ( = P9 ) P
( 38 )
116
The velocity and density fields should be corrected to satisfy mass conse rvation. Therefore , the new density and velocity fields, r and v, will satisfy the
continuity equation if
(r
yr
o)
P
d t
V q D w r v ? Sx P s 0
( 39 )
( 40 )
( 41 )
yr
o)
P
d t
V q D w ( r v* q r *v y r *v* q r 9 v9 ) ? S x P s 0
( 42 )
or
V
d t
q D w ( r v* q r *v) ? S x P s
d t
o
P
q D w ( r *v* ) ? S x P y D w ( r 9 v9 ) ? S x P ( 43 )
d t
q D w U *r x P q D w r * ( H wvx y D ( = P )) ? S x P
d t
o
P
q D w r *U * x P y D w ( r 9 v9 ) ? S x P
( 44 )
d t
q D w U *r x P y D w r *D ( = P ) ? S x P
s y D w r * ( H w vx ) ? S x P q
d t
o
P
q D w r *U*x P y D w ( r 9 v9 ) ? S x P
( 45 )
Substituting for density using the equation of state , the pressure equation becomes
VC p
d t
PP q D w C p U *P x P y D w r *D ( = P ) ? S x P
s
d t
o
P
q D w r *U * x P y D w r * ( H wv* x ) ? S x P
y D w r * ( H wv9 x ) ? S x P y D w ( r 9 v9 ) ? S x P
( 46 )
117
yr
o)
P
d t
V q D w ( r *v* q r *v9 q r 9 v* q r 9 v9 ) ? S x P s 0
( 47 )
d t
(r
U
P
qr
V
X)
r
q D w ( r 9 v* q r *v9 ) ? S x P s
d t
o
P
y D w ( r *v* ) ? S x P y D w ( r 9 v9 ) ? S x P
( 48 )
d t
X
P
q D w U *r 9 q r * ( H wv9 x y D ( = P9 ) ) ? S x P
s y
(r
U
P
yr
d t
o)
P
V y D w ( r *v* ) ? S x P y D w ( r 9 v9 ) ? S x P
( 49 )
Finally, substituting density corre ction by pressure correction, as obtained from the
equation of state, the final form of the pressure -correction equation is
VC p
d t
X
PP q D w C p U *P9 x P y D w r *D ( = P9 ) ? S x P
s y
(r
U
P
yr
d t
o)
P
V y D w r *U * x P y D w r *H wv9 x ? S x P
y D w ( r 9 v9 ) ? S x P
( 50 )
From the above, the one-to-one corresponde nce between the pressure and
pressure -corre ction equations is obvious, and as such, the two equations are
interchange able . Thus, algorithms in which the pressure field is obtaine d using a
pressure equation (e.g., SIMPLER, PISO, SIMPLEM, . . . ), may equally be calculated from a pressure-corre ction equation. More ove r, the second-orde r correction
term r 9 v9 is usually neglected. This does not affe ct either the conve rgence rate
(i.e., it is considerably smaller than othe r terms ) or the final solution, since at the
state of conve rgence the correction fields vanish. For this reason, it is neglected in
all subse quent derivations. Furthe rmore , if the H wv9 x term in the above equation is
retaine d, there will result a pressure-corre ction equation relating the pressure correction value at a point to all value s in the domain. Such an equation ensures
that the corrected fields will satisfy the continuity and momentum equations, but is
undesirable because it becomes intractable. To facilitate implementation and
reduce cost, simplifying assumptions related to this term have been introduce d.
Depending on these assumptions, different algorithms are obtained.
118
yr
d t
o)
P
( 51 )
V q D w r v ? Sx P s 0
( 52 )
r s Cp P
( 53 )
One then may wonde r about the differences among the various algorithms. The
key answer is in the different approximations to the H wv9 x terms and the various
sequences of operations that take place.
( 54 )
Corrector :
( v 9 , P9 , r 9 ) ( v** s v* q v9 , P* s P ( n) q P9 , r * s r
( n)
qr 9 )
X
v P s H wv9 x P y D P ( = P9 ) P
r 9 s C p P9
( 55 )
( 56 )
( 57 )
119
Condition:
(r
U
P
yr
o)
P
d t
[
(r
(n)
P
qr
X
P
yr
o)
P
V q D w( r
d t
[
VC p
d t
V q D w r *v** ? S x P s 0
(n)
q r 9 ) ( v* q H wv9 x y D ( = P9 )) ? S x P s 0 ( 59 )
PPX q D w C p U *P9 x P y D w r
s y
(r
( n)
P
yr
d t
yD wr
( n)
( 58 )
o)
P
( n)
VyD wr
(n)
D ( = P9 ) ? S x P
U *x P
H wv9 x ? S x P y D w ( r 9 v9 ) ? S x P
( 60 )
( 61 )
VC p
d t
sy
PPX q D w C p U *P9 x P y D w r
(r
( n)
P
yr
d t
o)
P
VyD wr
( n)
( n)
D ( = P9 ) ? S x P
( 62 )
U *x P
Discussion. Since at the state of convergence the pressure- and velocitycorrection fields are zero, the approximation introduce d does not have any effect
on the final solution. Rathe r, it affects the convergence behavior. More ove r, had
this term been retaine d, the pressure correction at any point would have been
related to the pressure correction at all grid points in the domain. The resulting
exact pressure -corre ction equation would have had a full coefficient matrix, and its
solution would have required an undesirable, more expensive , direction solution
method. However, because the H wv9 x terms are droppe d, the predicted pressure correction field is overe stimate d and the corrected velocity field no longe r satisfie s
the momentum equations. Therefore, in orde r to avoid divergence , the pressure
field is underrelaxed according to:
P s P* q a
( 63 )
p P9
which slows down the conve rgence rate of the iteration process ( a
underrelaxation factor for pressure ).
A global SIMPLE iteration
Solve implicitly for u and v , using the old pressure and density fields.
Calculate the D field.
being the
120
( 64 )
Corrector :
( v9 , P9 , r 9 ) ( v** s v* q v9 , P* s P ( n) q P9 , r * s r
( n)
qr 9 )
( 65 )
( 66 )
X
[ vP s H wv9 x P y D P ( = P9 ) P
( 67 )
X
Subtracting H w1x P vP from both sides, one gets
( 68 )
X
X
[ ( 1 y H w1 x P ) vP s H wv9 y v P x P y D P ( = P9 ) P
( 69 )
IX
vP s
X
H wv9 y v P x P
( 1 y H w1 x P )
r 9 s C p P9
DP
( 1 y H w1 x P )
( = P9 ) P
( 70 )
Condition:
(r
U
P
yr
d t
o)
P
V q D w r *v** ? S x P s 0
( 71 )
(r
( n)
P
qr
yr
o)
P
d t
(r
qD
VC p
d t
q r 9 ) v* q
( n)
H wv9 y vPX x
(r
yr
( n)
P
o)
P
d t
yD
(n)
VyD wr
X
H wv9 y vP x
1 y H w1 x
1 y H w1 x
( n)
(n)
1 y H w1 x
X
PP q D w C p U *P9 x P q D
s y
121
1 y H w1 x
)
)
( = P9 ) ? S
s 0 ( 72 )
P
( = P9 ) ? S
P
U *x P
y D w ( r 9 v9 ) ? S x P
? S
( 73 )
X
Approx imation: Neglect: H wv9 y vP x , D w( r 9 v9 ) ? Sx P .
X
vP s y D P ( = P9 ) P
( 74 )
VC p
d t
s y
X
PP q D w C p U *P9 x P y D
(r
( n)
P
yr
d t
o)
P
VyD wr
r
( n)
( n)
D ( = P9 ) ? S
U *x P
( 75 )
122
contribution to the conve rgence of the entire flow field by the iterative sweeps
within each momentum equation. On the othe r hand, finding the correct solution
for the pressure field represents the most important factor in the overall conve rgence. Base d on this argument, the PRIME algorithm can be summarize d as
follows.
The PRIME algorithm: Symbolic form
Predictor :
vPU s H wv ( n) x P y D P ( = P ( n) ) P
Corrector :
( v9 , P9 , r 9 ) ( v** s v* q v9 , P* s P ( n) q P9 , r * s r
( n)
qr 9 )
( 76 )
( 77 )
vPU s H wv ( n) x P y D P ( = P ( n) ) P
( 78 )
w
[
X
vP s H wv* y v ( n) x P q H wv9 x P y D P ( = P9 ) P
r 9 s C p P9
( 79 )
Condition:
(r
U
P
yr
o)
P
d t
(r
[
( n)
P
qr
X
P
yr
o)
P
d t
V q D w r *v** ? S x P s 0
V q D w( r
( n)
q r 9 ) ( v* q H wv* y v ( n) x
qH wv9 x y D ( = P9 ) ) ? S x P s 0
VC p
d t
X
PP q D w C p U *P9 x P y = w r
sy
(r
( n)
P
yr
o)
P
d t
y D w( r
( n) ) (
VyD wr
( 80 )
(n)
( n)
( 81 )
D ( = P9 ) ? S x P
U *x P
H wv* y v ( n) x q H wv9 x ) ? S x P y D w ( r 9 v9 ) ? S x P
( 82 )
VC p
d t
sy
X
PP q D w C p U *P9 x P y D w r
(r
( n)
P
yr
d t
o)
P
VyD wr
( n)
( n)
D ( = P9 ) ? S x P
U *x P
( 83 )
123
( 84 )
vP s H D wvx P q H C wvx P y D P ( = P ) P
( 85 )
Thus,
( 86 )
124
Corrector :
( v9 , P9 , r 9 ) ( v** s v* q v9 , P* s P ( n ) q P9 , r * s r
(n)
qr 9 )
( 87 )
vPU U s H D wv U U x P q H C wv U U x P y D P ( = P* ) P
[ s H D wv* q v9 x P q H C wv* q v9 x P y D P w = ( P ( n) q P9 ) x P
s H D wv* x P q H D wv9 x P q H C wv* x P q H C wv9 x P y D P ( = P ( n) ) P y D P ( = P9 ) P ( 88 )
w
[
X
vP s H wv9 x P q H C wv* y v ( n) x P y D P ( = P9 ) P
r 9 s C p P9
( 89 )
Condition:
(r
U
P
yr
o)
P
D t
(r
[
qr
( n)
P
X
P
yr
o)
P
V q D w( r
d t
( n)
V q D w r *v** ? S x P s 0
( 90 )
q r 9 ) ( v* q H wv9 x
qH C wv* y v ( n) x y D ( = P9 )) ? S x P s 0 ( 91 )
VC p
d t
X
PP q D w C p U *P9 x P y D w r
s y
(r
(n)
P
yr
o)
P
d t
yD wr
( n) (
VyD wr
( n)
( n)
D ( = P9 ) ? S x P
U *x P
H wv9 x q H C wv* y v ( n ) x ) ? S x P y D w ( r 9 v9 ) ? S x P
( 92 )
( 93 )
VC p
d t
sy
X
PP q D w C p U *P9 x P y D w r
(r
( n)
P
yr
d t
o)
P
VyD wr
( n)
( n)
D ( = P9 ) ? S x P
U *x P
( 94 )
125
A good understanding of the SIMPLEST algorithm is obtained by conside ring the following two limiting situations. The first is when the Reynolds numbe r
approaches zero, in which case convection becomes negligible, i.e.,
H C wvx P 0 H D wvx P H wvx P
( 95 )
and the method becomes identical to SIMPLE. The second limiting situation is
when the value of the Reynolds numbe r is very large so that diffusion is negligible
in comparison with convection, i.e.,
H D wvx P 0 H C wvx P H wvx P
( 96 )
In this case, the momentum equations are solve d, as in PRIME, via the explicit
Jacobi method w40x , using old information from the previous iteration and eliminating the need for any unde rrelaxation. Thus, the SIMPLEST algorithm can be seen
to be a compromise between SIMPLE and PRIME.
A global SIMPLEST iteration
Solve for u and v , tre ating H D w v x implicitly and H c w v x explicitly.
Calculate the D field.
Solve the pressure -correction equation.
Correct u, v , P, and r .
Solve implicitly the energy equation and update the density field.
Return to the first step and iterate until conve rgence.
126
First corrector
( v9 , P9 , r 9 ) ( v* s v ( n ) q v9 , P* s P ( n) q P9 , r * s r
( n)
qr 9 )
( 97 )
[ vPU s H wv* x P y D P ( = P* ) P
( 98 )
vP( n) s H wv ( n ) x P y D P ( = P ( n) ) P
( 99 )
w
[
X
vP s H wv9 x P y D P ( = P9 ) P
r 9 s C p P9
( 100 )
Condition:
(r
yr
U
P
o)
P
d t
[
(r
( n)
P
qr
X
P
yr
o)
P
d t
[
V q D w( r
VC p
d t
X
PP q D
sy
(n)
V q D w r *v* ? S x P s 0
q r 9 ) ( v ( n) q H wv9 x y D ( = P9 ) ) ? S x P s 0 ( 102 )
C p U ( n ) P9
(r
( n)
P
yr
o)
P
d t
yD wr
(n)
( 101 )
yD wr
VyD wr
( n)
( n)
D ( = P9 ) ? S x P
U ( n) x P
H wv9 x ? S x P y D w ( r 9 v9 ) ? S x P
( 103 )
VC p
d t
s y
X
PP q D
(r
(n)
P
C p U ( n) P9
yr
o)
P
yD wr
VyD wr
D ( = P9 ) ? S x P
U (n) x P
( 104 )
vPU s H wv* x P y D P ( = P* ) P
( 105 )
d t
(n)
(n)
127
Second corrector :
( v 0 , P0 , r 0 ) ( v** s v* q v0 , P** s P* q P0 , r ** s r * q r 0 )
( 106 )
w
[
vPY s H wv0 x P y D P ( = P0 ) P
r 0 s C p P0
( 108 )
Condition:
(r
UU
P
yr
d t
(r
[
[
U
P
qr
Y
P
yr
o)
P
d t
VC p
d t
o)
P
V q D w r **v** ? S x P s 0
( 109 )
V q D w ( r * q r 0 ) ( v* q H wv0 x y D ( = P0 ) ) ? S x P s 0 ( 110 )
Y
PP q D wU *C p P0 x P y D w r *D ( = P0 ) ? S x P
sy
(r
U
P
yr
d t
o)
P
V y D w r *U * x P y D w r *H wv0 x ? S x P y D w ( r 0 v0 ) ? S x P ( 111 )
( 112 )
VC p
d t
PPY q D w U *C p P0 x P y D w r *D ( = P0 ) ? S x P s y
(r
U
P
yr
d t
o)
P
V y D w r *U * x P
( 113 )
128
Solve implicitly for u and v using the new pressure and density fields.
Solve the pressure -corre ction equation using the new velocity field to obtain a new
pressure-corre ction field.
Correct u and v using the new pressure-correction field.
Solve implicitly the energy equation and update the density field.
Return to the first step and iterate until conve rgence.
( 114 )
First corrector :
( v9 , P9 , r 9 ) ( v** s v* q v9 , P* s P ( n) q P9 , r * s r
( n)
qr 9 )
w
[
X
vP s H wv9 x P y D P ( = P9 ) P
r 9 s C p P9
( 115 )
( 116 )
( 117 )
Condition:
(r
U
P
yr
o)
P
d t
(r
(n)
P
qr
X
P
yr
o)
P
d t
VC P
d t
V q D w( r
X
PP q D w C p U *P9 x P y D w r
sy
(r
n
P
yr
d t
o)
P
VyD wr
V q D w r *v** ? S x P s 0
( n)
( n)
( n)
( 118 )
q r 9 ) ( v* q H wv9 x y D ( = P9 ) ) ? S x P s 0 ( 119 )
D ( = P9 ) ? S x P
U *x P y D w r
(n)
H wv9 x ? S x P y D w ( r 9 v9 ) ? S x P ( 120 )
( 121 )
129
VC p
X
PP q D w C p U *P9 x P y D w r
d t
sy
(r
yr
n
P
o)
P
d t
VyD wr
( n)
( n)
D ( = P9 ) ? S x P
U *x P
( 122 )
Second predictor :
vPU U U s H** wv** x P y D P** ( = P* ) P
( 123 )
Second corrector :
( v0 , P0 , r 0 ) ( v**** s v*** q v0 , P** s P* q P0 , r ** s r * q r 0 )
( 124 )
( 125 )
( 126 )
P0 ) P
D PU U ( =
P0 ) P
( 127 )
Condition:
[
(r
U
P
qr
Y
P
yr
o)
P
d t
(r
UU
P
yr
d t
o)
P
V q D w r **v**** ? S x P s 0
( 128 )
( 129 )
VC p
d t
Y
PP q D w C p U ***P0 x P y D w r *D PU U ( = P0 ) ? S x P
s y
(r
U
P
yr
d t
o)
P
y D w ( r 0 v0 ) ? S x P
( 130 )
130
( 131 )
VC p
d t
s y
Y
PP q D w C p U ***P0 x P y D w r *D PU U ( = P0 ) ? S x P
(r
U
P
yr
d t
o)
P
V y D w r *U *** x P
( 132 )
( 133 )
Y
s H** wv* x y D PU U ( = P* ) P q H** wv9 x P q vP
If v* is assume d to be ne arly equal to H** wv*x y D PU U ( = P*) P then, the H wv9 x term
that was neglected in the first corrector stage is partially recovere d as
Y
H** w yD ( = P9 )x P in the second corrector stage in addition to vP . Therefore, the
second pressure -correction brings the velocity and pressure fields closer to satisfying both the momentum and continuity equations without the need to unde rrelax
the pressure correction. Because of this fact, PISO does provide considerable
savings in computational time when compare d to SIMPLE. More over, by following
the sequence of events, it can be easily seen that PISO may be conside red to be a
combination of one SIMPLE step and one PRIME step, hence combining the
implicitness of the SIMPLE algorithm with the stability of the PRIME algorithm.
Furthe rmore, improving the time accuracy of PISO through incre asing the numbe r
of corrector iterations is akin to adding furthe r PRIME steps.
Finally, it is common in the literature to re ad that PISO is essentially
equivale nt to SIMPLER (e.g., w42x ). From the above presentation, it is obvious that
the two algorithms are diffe rent. In SIMPLER the pressure is update d once afte r
solving a pressure or a pressure-correction equation in which the coefficients are
based on the old velocity field. In PISO the pressure field is update d twice using
the latest available velocity fields. Consequently, the pressure and velocity fields at
the end of a SIMPLER iteration are diffe rent from those obtained at the end of a
PISO iteration, and users of SIMPLER and PISO should be aware of that.
A global PISO iteration
Solve implicitly for u and v , using the old pressure and density fields.
Calculate the D field.
Solve the pressure -correction equation.
Correct u, v , P, and r .
131
Solve implicitly the energy equation and update the density field.
Solve the momentum equations explicitly and calculate the D field.
Solve the pressure -correction equation.
Correct u, v , P, and r .
Return to the first step and iterate until conve rgence
( 134 )
Corrector :
( v9 ,P9 , r 9 ) ( v** s v* q v9 , P* s P ( n) q P9 , r * s r
( n)
qr 9 )
w
[
vPX s H wv9 x P y D P ( = P9 ) P
r 9 s C p P9
( 135 )
( 136 )
( 137 )
Condition:
(r
yr
U
P
o)
P
d t
[
(r
( n)
P
qr
X
P
yr
o)
P
d t
(r
( n)
P
qr
X
P
yr
d t
s yD w r
(n)
V q D w r *v** ? S x P s 0
V q D w( r
o)
P
( n)
( 138 )
q r 9 ) ( v* q v 9 ) ? S x P s 0
V q D w r 9 v* ? S x P q D w r
v* ? S x P y D w ( r 9 v9 ) ? S x P
( n)
( 139 )
v 9 ? Sx P
( 140 )
132
( 141 )
sx (
(
)
y D sx
= P9 ) x P y D P ( = P9 ) P
P = P9 P s H w y D
( 142 )
Assume that the pressure diffe rence local to the velocity can be approximate d to
be re presentative of all pressure diffe rences, i.e., H w y D sx ( = P9 ) x P s
( = P9 ) P H w yD sx x P . Thus,
sx
(
)
(
)
(
)
D sx
P = P9 P s = P9 P H w D x P q D P = P9 P
( 143 )
sx
D sx
P s H wD x P q D P
( 144 )
VC p
d t
s y
PPX q D w C p U *P9 x P y D w r
(r
(n)
P
yr
d t
o)
P
VyD wr
( n)
(n)
D sx ( = P9 ) ? S x P
U *x P
( 145 )
133
( n)
qr 9 )
( 146 )
[ vPU s H wv* x P y D P ( = P* ) P
( 147 )
vPn s H wv ( n) x P y D P ( = P ( n) ) P
( 148 )
w
[
X
vP s H wv9 x P y D P ( = P9 ) P
r 9 s C p P9
( 149 )
Condition:
(r
U
P
yr
o)
P
d t
(r
[
(n)
P
qr
X
P
yr
o)
P
d t
VC p
d t
PP 9 q D
sy
(r
V q D w( r
C p U ( n) P9
(n)
P
yr
o)
P
d t
( n)
V q D w r *v* ? S x P s 0
q r 9 ) ( v ( n) q H wv9 x P y D ( = P9 ) ) ? S x P s 0 ( 151 )
yD wr
VyD wr
( 150 )
( n)
(n)
D ( = P9 ) ? S x p
U n xP y D w r
(n)
H wv9 x P ? S x P y D w ( r 9 v9 ) ? S x P
( 152 )
VC p
d t
X
PP q D
s y
(r
(n)
P
C p U ( n) P9
yr
d t
o)
P
yD wr
VyD wr
(n)
(n)
D ( = P9 ) ? S x P
U (n) x P
( 153 )
Second predictor :
vPU U s H* wv** x P y D PU ( = P* ) P
( 154 )
134
( 155 )
where R M denotes the momentum residual. As state d in w37x , this method has a
better theoretical smoothing rate , suitable for the high-fre quency error reduction
that is important in multigrid techniques. In w37, 43x , the technique was applie d to
the SIMPLE and SIMPLEC algorithms to yield two modifie d versions denote d by
SIMPLESSE and SIMPLESSEC, respective ly. However, the approach is equally
applicable to all previously described algorithms. Moreove r, when using this
technique with the SIMPLEC algorithm, the H and D operators should be replaced
by their equivalent expre ssions as derive d earlier.
d t
V( r
X
PP q D w C p UP 9 x P y D w r D ( = P9 ) ? S x P s y
yr
d t
o)
P
y D w r U x P ( 156 )
where, depending on the algorithm used, U and r represent value s from the
previous iteration or from a previous corrector step. When discretizing this equation, careful atte ntion should be paid to the second term on the left-hand side,
which is similar to a conve ction term and for which any convective scheme may be
135
used. Adopting the UPWIND scheme w1x , the discretized form of the conve ction-like
term D ? C p UP9 @ P is
D ? C p UP9 @ P
s ( C p UP9 ) e q ( C p UP9 ) w q ( C p UP9 ) n q ( C p UP 9 ) s
X
X
X
X
s ( C p ) e w 5 Ue , 0 5 PP y 5 y Ue , 0 5 PE x q ( C p ) w 5 Uw , 0 5 Pp y 5 y Uw , 0 5 PW
X
X
X
X
q ( C p ) n w 5 Un , 0 5 PP y 5 y Un , 0 5 PN x q ( C p ) s w 5 Us , 0 5 PP y 5 y Us , 0 5 PS x
( 157 )
Re arranging, one obtains:
X
D w C p UP 9 x P s ( C p ) e 5 Ue , 0 5 q ( C p ) w 5 Uw , 0 5 q ( C p ) n 5 Un , 0 5 q ( C p ) s 5 Us , 0 5 PP
X
X
y ( C p ) e 5 y Ue , 0 5 PE y ( C p ) w 5 y Uw , 0 5 P W
X
X
y ( C p ) n 5 y Un , 0 5 PN y ( C p ) s 5 y Us , 0 5 PS
( 158 )
sr
D w u x e ( S ex ) q D w v x e ( S ey )
Se ? d e
qr
(D w u x (=
e
)e k
P9
( PEX y PPX )
q D w v x e (= P9
x
e
qr
qr
D w u x w ( S wx ) q D w v x w ( S wy )
(=
P9
)wk
x
w
q D w v x w ( = P9
D w u x n ( S nx ) q D w v x n ( S ny )
qr
qr
Sn ? d n
(D w u x (=
n
P9
)n k
x
n
qr
Ss ? d s
(D w u x (=
s
P9
)s k
x
s
X
X
( PW
y PP )
)w k
y
w
( PNX y PPX )
q D w v x n ( = P9
D w u x s ( S sx ) q D w v x s ( S sy )
qr
y
e
Sw ? dw
(D w u x
)e k
)n k
y
n
( PSX y PPX )
q D w v x s (= P9
)s k
y
s
( 159 )
136
where the unde rlined terms account for the nonorthogonal factors. They are
usually neglected since their contribution is small in comparison with other terms
and vanish when the grid is orthogonal. However, they could be accounte d for by
moving them to the right-hand side, adding them to the source term, and modifying
the solve r to explicitly update their value s after a solve r (not global ) iteration.
Neglecting these terms, Eq. (159 ) becomes
D w r D ( = P9 ) ? S x P
2
sr
D w u x e ( S ex ) q D w v x e ( S ey )
S ex d ex q S ey d ey
( PEX y PPX )
qr
D w u x w ( S wx ) q D w v x w ( S wy )
S wx d wx
S wy d wy
qr
D w u x n ( S nx ) q D w v x n ( S ny )
S nx d nx q S ny d ny
qr
sG
P9
e
s y (G
D w u x s ( S sx ) q D w v x s ( S sy )
S sx d sx q S syd sy
( PEX y PPX ) q G
P9
e
qG
P9
w
qG
P9
w
P9
n
X
X
( PW
y PP )
( PNX y PPX )
( PSX y PPX )
X
X
( PW
y PP ) q G
qG
P9
s
) PPX q G
P9
n
P9
e
( PNX y PPX ) q G
X
PE q G
P9
w
X
PW q G
P9
s
( PSX y PPX )
P9
n
X
PN q G
P9
s
X
PS ( 160 )
w Uw
q r n Un q r s Us
( 161 )
( 162 )
where
a EP 9 s G
P9
e
q ( C p ) e 5 y Ue , 0 5
P9
aW
sG
P9
w
q ( C p ) w 5 y Uw , 0 5
a NP 9 s G
P9
n
q ( C p ) n 5 y Un , 0 5
a SP 9 s G
P9
s
q ( C p ) s 5 y Us , 0 5
( 163 )
( a PP 9 ) s
137
V ( C p )P
d t
P9
a PP 9 s a EP 9 q a W
q a NP 9 q a SP 9 q ( a Pp9 )
q ( C p ) e Ue q ( C p ) w Uw q ( C p ) n Un q ( C p ) s Us
b PP 9 s y
(r
yr
d t
o)
P
V y ( r e Ue q r
w Uw
q r n Un q r s Us )
( 163 )
( Cont.)
CLOSING R EMARKS
The SIMPLE algorithm and its variants were reformulate d using a unified,
compact, and easy-to-understand notation. The new notation allowed the essence
share d in their derivations to be transpare nt. More over, the philosophie s behind
these algorithms and the differences among them were pointe d out. Furthe rmore,
the formulation of all these algorithms was exte nded to predict fluid flow at all
speeds.
REFERENCES
1. S. V. Patankar, Numerical Heat Transfer and Fluid Flow, Hemisphere, New York, 1981.
2. S. V. Patankar and D. B. Spalding, A Calculation Procedure for Heat, Mass and
Momentum Transfer in Three-Dimensional Parabolic Flows, Int. J. Heat Mass Transfer,
vol. 15, pp. 1787 ] 1806, 1972.
3. A. J. Chorin, A Numerical Method for Solving Incompressible Viscous Flow Problems,
J. Com put. Phys., vol. 2, pp. 12 ] 26, 1967.
4. A. Rizzi and L. E. Eriksson, Computation of Inviscid Incompressible Flow with Rotation, J. Fluid Mech., vol. 153, no. 3, pp. 275 ] 312, 1985.
5. D. Choi and C. L. Merkle, Application of Time-Iterative Schemes to Incompressible
Flow, AIAA J., vol. 23, pp. 1518 ] 1524, 1985.
6. K. C. Karki, A Calculation Procedure for Viscous Flows at All Speeds in Complex
Geometries, Ph.D. thesis, University of Minnesota, June 1986.
7. W. Shyy and M. E. Braaten, Adaptive Grid Computation for Inviscid Compressible
Flows Using a Pressure Correction Method, AIAA Paper 88-3566-CP, 1988.
8. C. M. Rhie. A Pressure Based Navier-Stokes Solver Using the Multigrid Method, AIAA
Paper 86-0207, 1986.
9. F. S. Lien and M. A. Leschziner, A General Non-orthogonal Collocated Finite Volume
Algorithm for Turbulent Flow at All Speeds Incorporating Second-Moment
Turbulence-Transport Closure, Part 1: Computational Implementation, Com put. Meth.
Appl. Mech. Eng., vol. 114, pp. 123 ] 148, 1994.
10. E. S. Politis and K. C. Giannakoglou, A Pressure-Based Algorithm for High-Speed
Turbomachinery Flows, Int. J. Numer. Meth. Fluids, vol. 25, pp. 63 ] 80, 1997.
11. L. S. Caretto, R. M. Curr, and D. B. Spalding, Two Numerical Methods for Three
Dimensional Boundary Layers, Meth. Appl. Mech. Eng., vol. 1, p. 39, 1972.
12. S. P. Vanka, Fully Coupled Calculation of Fluid Flows with Limited Use of Computer
Storage, Argonne Nat. Lab. Rep. ANL-83-87, 1983.
138
13. S. P. Vanka, and G. K. Le af, Fully Coupled Solution of Pressure Linked Fluid Flow
Equations, Argonne Nat. Lab. Rep. ANL-83-73, 1983.
14. M. E. Braaten, Development and Evaluation of Iterative and Direct Methods for the
Solution of the Equations Governing Recirculating Flows, Ph.D. thesis, University of
Minnesota, 1985.
15. F. H. Harlow and J. E. Welch, Numerical Calculation of Time-Dependent Viscous
Incompressible Flow of Fluid with Free Surface, Phys. Fluids, vol. 8, pp. 2182 ] 2189,
1965.
16. W. Rodi, S. Majumdar, and B. Schonung, Finite Volume Methods for Two-Dimensional
Incompressible Flows with Complex Boundaries, Com put. Meth. Appl. Mech. Eng, vol.
75, pp. 369 ] 392, 1989.
17. C. M. Rhie, A Numerical Study of the Flow Past an Isolated Airfoil with Separation,
Ph.D. thesis, Department of Mechanical and Industrial Engineering, University of
Illinois at Urbana-Champaign, 1981.
18. C. M. Rhie and W. L. Chow, A Numerical Study of the Turbulent Flow Past an Isolated
Airfoil with Trailing Edge Separation, AIAA J., vol. 21, pp. 1525 ] 1532, 1983.
19. M. Peric, R. Kessler, and G. Scheuerer, Comparison of Finite Volume Numerical
Methods with Staggered and Collocated Grids, Com put. Fluids, vol. 16, no. 4, pp.
289 ] 403, 1988.
20. S. K. Choi, H. Y. Nam, and M. Cho, A Calculation Procedure for Incompressible Flow
in Complex Geometries Using Momentum Interpolation Method, 7th Int. Conf . on
Numerical Methods in L am inar & Turbulent Flow, vol. VII, part 2, pp. 1634 ] 1644, 1991.
21. C. Hsu, A Curviline ar-Coordinate Method for Momentum, Heat and Mass Transfer in
Domains of Irregular Geometry, Ph.D. thesis, Univeristy of Minnesota, 1981.
22. C. Prakash, A Finite Element Method for Predicting Flow through Ducts with Arbitrary
Cross Sections, Ph.D. thesis, University of Minnesota, 1981.
23. M. Peric, A Finite Volume Method for the Prediction of Three Dimensional Fluid Flow
in Complex Ducts, Ph.D. thesis, Imperial College, Mechanical Engineering Department,
London, 1985.
24. H. Aksoy and C. J. Chen, Numerical Solution of Navie r-Stokes Equations with NonStagge red Grids Using the Finite Analytical Method, Num er. Heat Transfer B, vol. 21,
pp. 287 ] 306, 1992.
25. T. F. Miller and F. W. Schmidt, Use of a Pressure-Weighted Interpolation Method for
the Solution of the Incompressible Navier-Stokes Equations on a Non-staggered Grid
System, Numer. Heat Transfer, vol. 14, pp. 213 ] 233, 1988.
26. F. V. Poussin, An Accelerated Relaxation Algorithm for Iterative Solution of Elliptic
Equations, SIAM. J. Num er. Anal., vol. 5, pp. 340 ] 351, 1968.
27. A. Brandt, Multi-level Adaptive Solutions to Boundary-Value Problems, Math. Com put.,
vol. 31, pp. 333 ] 390, 1977.
28. A. Brandt, G uide to Multigrid Dev elopm ent, Springer-Verlag, 1982.
29. B. R. Hutchinson and G. D. Raithby, A Multigrid Based on the Additive Correction
Strategy, Num . Heat Transfer vol. 9, pp. 511 ] 537, 1986.
30. J. P. Van Doormaal and G. D. Raithby, Enhancement of the SIMPLE Method for
Predicting Incompressible Fluid Flows, Numer. Heat Transfer, vol. 7, pp. 147 ] 163, 1984.
31. D. B. Spalding, Mathematical Modelling of Fluid Mechanics, Heat Transfer and Mass
Transfer Processes, Mech. Eng. Dept., Rep. HTS r 80 r 1, Imperial College of Science,
Technology and Medicine, London, 1980.
32. J. P. Van Doormaal and G. D. Raithby, An Evaluation of the Segregated Approach for
Predicting Incompressible Fluid Flows, ASME Paper 85-HT-9, presented at the Natl.
Heat Transfer Conf., Denver, CO, 4 ] 7 August 1985.
139