in Eq. (ll) are discretized using the wide, six-point difference stencil for 0_/_ and ..... Pig. 1. The shape of the lower wall between. 0 < z < 1 is y(z). = r sin 2 rrx.
AIAA
OO-2252
Multigrid Conservative
Relaxation
Discretization Flow Equations Thomas
W. Langley
David
Sidilkover
and
for
of the
Compressible
Roberts
NASA
Institute
of a Factorizable,
Research
Computer
Center
Applications
in Science
and Engineering
J. L. Thomas
NASA Hampton,
Langley
Research
Center
VA
Fluids
%
19-22
June
2000
2000/Denver,
CO
For permission to copy or republish, contact the American Institute of Aeronautics and Astronautics 370 L'Enfant Promenade, S.W., Washington, D.C. 20024
/
i l
MULTIGRID
RELAXATION
DISCRETIZATION
OF
OF
THE
A
Thomas
for
W.
Computer
J.
L.
Langley Hampton,
Abstract
the
factorizable
compressible
Euler
discretization
of
developed
by
equations
Sidilkover is extended to conservation era curvilinear body-fitted grids. equations
are solved
Seidel relaxation flow in a channel 0.0001
by symmetric
to a supercritical
Mach
uniform in the
lution
flow
the
semi-infinite
collective
and FAS multigrid. with Mach numbers
demonstrating loss of accuracy for
form on genThe discrete
around
parabolic
scheme
maintains
tiiining
a stagnation
number
convergence incompressible the
body
rapid
Gauss-
Solutions for ranging from are
leading
edge
demonstrates
convergence
shown,
that
for a flow
Steady
inviscid
which
subsystem tropy and bolic
can
The
operator,
hyperbolic
for supersonic
flow, space
marching
corresponding
culty well.
is true
As has
by
the
Euler
been
corresponds
to a full
for subsonic
flow and
flow.
For a purely
most
efficient
supersonic
way of solving
operator which
terms
liptic
correction Existing flow rely As such, tiveness
for certain multigrid
smooth
methods
only
gives
components for subsonic
part
of the
of the
error.
and
transonic
on the coarse grid to smooth the entire system. they are fundamentally limited by the ineffecof the
coarse
grid
in correcting
the
part
of the
Copyright Q2000 by the American Institute of Aeronautics and Astronautics, Inc. No copyright is asserted in the United States under Title 17, U.S. Code. The U.S. Government has a royalty-free license to exercise all rights under the copyright claimed herein for government purposes. All other rights are reserved by the copyright owner. * Research Scientist t Senior Staff Scientist t AIAA Fellow
1, to obtain
as
ideal
correspond
efficiently
are rapidly that
to
solved the
to be equal
The
advec-
by marching
and
the el-
with
by spfitting
parts,
a multigrid
the system
convergence
to the
and
behavior.
iteration. into
rate
slowest
its ellipof the
of the
full
two
sub-
systems. Using this approach, Brandt and Yavneh have demonstrated textbook multigrid for the incompressible equations and
of the
Ta'asan
compressible
Euler
of "canonical
Euler
equations 4.
equations
In a closely
equations.
This
variables"
which
method and that
ap-
for the
is based
express
of an elliptic
for two-dimensional
related solver
3 it. is shown
can be achieved
the
on
stead),
a hyperbolic ideal
multi-
for the compressible
Euler
subsonic
body-
flow
using
grids. work 5' 6, the
the
of the
advection
system
and unstructured the discretization.
dimensions
strated
ideal
gation collective
a multi-
operator
of equations.
from
Both
equawhich
the
elliptic
structured
grid
can achieve ideal multigrid convergence flows. The scheme has been extended
to three
This
presented
grid flow solvers were written using The results in Ref. 5 demonstrated
that the scheme rates for internal
grids--a equations
authors
for the steady, incompressible Euler on a pressure Poisson discretization
distinguishes
flows.
for a simple
a staggered-grid
a fast multigrid
in terms
In an earlier
part
are
using
equations.
In Reference
grid efficiency
results
grid,
3 presented
a set
partition
2. Their
a Cartesian
discretization proach,
grid scheme tions based
grid
Brandt
to elliptic
are treated
tion
coarse
which
correspond
terms
fitted
as the
by
diffi-
flows
advection,
those
the Euler equations. For subsonic flow, the ellipticity of the full potential factor should be effectively handled by multigrid. However multigrid is not effective for advecoperators,
out
same
ferential
ought
is elliptic
This viscous
inviscid flows, parts of the dif-
system
One
factor.
Reynolds-number
pointed
the
equa-
to advection
for high
multigrid convergence rates for subsonic, the discretization nmst distinguish those
geometry
of as two snbsystems.
is the
error
advection
subsystem
which
Center
1 argams
con-
Engineering
VA
Navier-Stokes
is described
other
Research
tic and
corresponds to the equations governing envorticity advection. This subsystem is hyper-
in space.
potential
flow
and
t
Brandt
point.
be thought,
in Science
of a
Introduction
tions,
Center
Thomas
tion
rates and no limit. A so-
EQUATIONS
Roberts*
Applications
NASA
second-order
FLOW
Langleg Research D. Sidilkover t
and
The
CONSERVATIVE
COMPRESSIBLE
NASA
Institute
FACTORIZABLE,
by Sanchez
multigrid
approach
rates
has the advantage
collocated, vertex-based is used, simplifying the operations
7, who has
convergence
and
Gauss-Seidel
allowing relaxation.
also
demon-
for
internal
of non-staggered
discretization restriction and the
use
of the prolon-
of simple
Although
the
point pres-
surePoisson approach maybeextended tocompressibleseries about each grid vertex and considering the leadflows,it.is not conservative andis unsuited forsuper- ing terms of the truncation error. These terms are the criticalflowswithshocks.Furthermore, theextension artificial dissipation of the scheme. toviscous flowsislimitedto tileincompressible NavierThe starting point fox" the scheme is the twoStokes equations. dimensional Euler equations in non-conservation form. Recently, Sidilkover 8 hasdevised a discretization of Let p be the density, ff = _u + jr, be the velocity, and p thecompressible flowequations that.overcomes these be the pressure. The entropy s is defined as limitations.Thisdiscretization maybeappliedto the Eulerequations in conservative form,usingthemultidimension;d upwindscheme ofSidilkover 9' 10.In Ref.10 it is shownthatthisapproach should leadtoa scheme thatis factorizable, i.e.,thescheme distinguishes be- where p0 and po are a reference pressure and density, and "_ is the ratio of the specific heats. tweenthosepartsoftheoperator thatrepresent, advec- respectively, Then the Euler equations may be written in the varition,andthatpartof the operator that. represents poables (s, u, v,p): tential
flow.
In
Ref.
8, such
a factorizable
constructed for Cartesi,'m grids. Because tion is stable for Gauss-Seidel relaxation,
scheme
is
the diseretizait the conver-
gence rate does not depend upon a Courasat-FriedrichLewy number restriction, unlike standard discretizations of the Euler equations which must use time marching. For this reason, the same convergence rate is obtained for subsonic pressible
Mach
limit.
The
numbers
all
scheme
may
be written
plus
an
of a central-difference As
such,
it
upwind,
part
is very
similar
finite-vohune
discrete terms
must
he shows the the
dissipation.
scheme
In the
ix, the present
factorlzable nates
viscosity
scheme
presented,
artificial
the
terms
must
be
ability.
A discussion
of the next.
nel_
infetl _Iach
numbers
additional
of the
computation
how
collective
the
dissi-
factorlz-
Gauss-Seidel
for flow in a chan-
ranging
from
0.0001
flow around
the
leading
ficulties
for
the authors'
previous
point
scheme
Mathematical The can
artificial
tion,
or first
hy
first
differential
discrete
scheme.
This
is found
by expmlding
were
observed
11
of the
it
way
without
The
advection
encing in coordinate
the
differential
the difference
weakly
coupled
equations
to
the form of equa-
the
through
momen-
the
equation
may be discretized independently of the and pressure equations in any appropriate affecting
the
operator
factorizability
ff.V uses
of the
simple
Eq. (2a). Let (,_, r/) be a general system and define the contravariant
of the
velocity,
(U, F') by the
scheme.
upwind
differ-
curvilinear compo-
transformation
(;: ;,9 In this
coordinate
tions
are
with
a grid
system
discretized
first-order
spacing upwind
g.Vt7 = (?O_ + f'O,.
on
a uniform
A¢
=
difference
ATI =
grid 1.
in The
equaspace,
FDA
of the
to tT.V is
and
the
entropy
1
equation
.=
is discretized
(4) as
qs = 0.
has
upwind also
particular
been
cases
(5)
discretization used
shown
of
in Eq. below
the
the
(5). use
order advection operator has an insignificant the computed results. This point, is discussed
(FDA),
of the
sults
equation
which
in a Taylor
The tions,
for
of a second-
equa-
schemc
adveetion
However,
modified
equation
The (¢,t/)
approximation
1
the
factorizable
approximation is the
Therefore, momentum
A second-order
presenting
scheme depends on added to this system
of state (1). In fact, the entropy equation corresponds to one of the advection factors of the Euler equations.
operator
Formulation
dissipation
be described
that
(2c)
+ ff.Vp
to
the dif-
a stagnation
= 0.
pressure
are
edge of a semi-infinite parabola demonstrates that current scheme does not suffer from the convergence near
(2b)
is only
the
are shown, demonstrating rates of the scheme. An
for the
= 0,
multidimensional
Solutions
Mach number and convergence
entropy and
coordi-
equations
to maintain
point
is presented
with
with
tum
nents
It. is shown
relaxation a supercritical the accuracy
In addition,
body-fitted
discretized
ff-Vz7 + 1Vp P
of Sidilkover's
governing
form
dissipation.
pation
for the
limit.
curviIinear,
the
way.
generalization
First.,
including
upwind
that
the factorizability, and thus h-ellipticity of the operator
the
to
can be a modi-
may be rescaled
incompressible
is presented.
and with
(2a)
The factorizability of the of the artificial dissipation tions. The
and
the dissipation
in a specific
work,
Euler
equations, 8 shows
the artificial
viscosity.
for the
factorizability
incom-
to standard
Sidilkover
Mach number such that accuracy, as well as the
is preserved
artificial
scheme
be discretized
the
= 0,
pc2V.a
in the form
upwind
to preserve
that
to
formulation
Navier-Stokes
as a conventional
fied artificial
way
discretizations
Reynolds-Averaged written
in
the
a.Vs
attect on in the Re-
section. dissipation Eqs.
(2b)
for the momentum
and
pressure
and (2c), is the umltidimensional
equaupwind
dissipation of Sidilkover pation
is written
9.
In vector
notation,
this dissi-
adding
the following
corrections:
as a-_"/O = qL/-4- 7l [(--?[05/at -'t- l [_='l 0_0._',
a.va+ 1-vp- "'ev (p 2v.a+a.vp) =0, p
2
(6)
2pc
pc2_7.ff+ff.Vp
--pc-_.(_._ff+
_Vp)
=0,
(7)
1
The
cross-derivative
second-order where
c is the
coefficients, grid tum
speed
and
spacing. equation
residual, the
the
divergence
of the
artificial
scheme.
ap
are
pressure
momentum
of the
and
the
dissipation. second
equation
that
ttle
that
curl
The
ilarly,
because
of the
momentum
part.
irrotational
of the
factor,
If the
and
advection terms
the
h-elliptic. This central difference
advection
8.
equation, operator
Replacing
first-order
tion term second-order
(2e)
is
to the
by Eq.
accurate.
upon
covariant
operator
Note
(7)
components
by
the
the
to be
accuracy
The
form
accuis not of the term
first-order
pressure
while
of these
of the
of the velocity to the
a discretization
by
maintaining nmst be that fac-
terms
is de-
physical
(8) to find (9a)
into
and
the
/at and
(9b)
are
physical
of the
necessary
factorizability
for at the
_', the
antidis-
evaluated,
and
components.
This
form
/7.Vff
= qa
+ VD
(10)
where D= The
above
of the
gain
more
terms,
where is the
(iCrlo, a+
expression
coordinates tion
GI
the
.
is the generalization one given
insight FDA
in Ref.
about,
of Eq.
the
(10)
(11) to curvilinear
8. nature
may
of the
be rewritten
J
+ .7(c' I '1
correcas
e"lC'l
_o _ 0_t, - Ouu is the vorticity, J = av_y,7 - y{x, Jacobian of the coordinate transformation, mM
(4e, gn) are the contravariant coordinate directions.* Note t.ional,
then
identically accurate.
the first-order and
truncation
the approximation
It was pointed out upon the dissipation written
basis vectors in the ((, _/) that. if the flow is irrotaerror
terms
becomes
vanish
second-order
above thai factorizalfility depends of the momentum equation being
as the gradient
of the pressure
dissipation
of the
equation
divergence second-order
of
pressure
the momentum correction to the
equation, being
equation. adveetion
and
written
the
as the
Likewise, the terms must be
the curl operator first-order
of Eq. (6) yields q.0 = 0. Thus the acting on the vorticity is unchanged scheme.
To get. full use second-order operator elliptic,
second-order accurate
in the momentum fully
second-order
= C0{
+ _'0 n
accuracy discretization equation. upwind
advection from the
it. is necessary to of the advection The
advection
FDA
of an h-
operator
(_, Ii), and
contravariant vector.
(9) are
in the form of a gradient so that the vorticity equation is is unaffected. With the definition in Eq. (10), taking
advec-
approximated
coordinates
in terms
are related
dius-
restores h-ellipticity but it. now becomes
that
the computational
be written
ant. components
and
discretized
antidissipative term_ equation in such a way
is preserved.
must
are
is second-order such a scheme
in Eq. (4) factorizable,
second-order
torizability
gradient,
(7)
in Eq. (7) continues central differences.
obtain
the
Eq.
Eq. (6). Sidilkover shows corresponds to vorticity
this
h-ellipticity, appropriate added to the momentum pendent
pressure
back
+vD=
as the gradi-
described
the scheme However,
upwind approximation and the scheme remains
by
the solenoidal
lack of h-ellipticity is a result approximation to the advection
in the momentum that this advection
they
not
it corresponds
(6) and
gives
To
by tile Jr-
equation
FDA
operator, in Eqs.
ing central differences, rate and factorizable.
To
pressure
equation,
to Sim-
this property.
vergence
only
corresponds
may be written
As the
continuity
full potentiM preserves
part
to
(7) is the divergence
field but
rotated
equation of by the arti-
only
Eq.
in Eq.
then
in Eq.
and to retain
(6) is
equations.
it is affected
of the velocity
of a potential.
a form
of Eq.
equation,
part
The
equation
of the Euler
the dissipation
rotational ent
vortieity factor
Using
terms
It
leads
of Eq.
time.
sipative
terms
accuracy
is
residual.
dissipation
Note
adveetion
same
to the
equation
identical to the curl of Eq. (2b), i.e., vorticity the governing Euler equations is unaffected ficial
scaling
proportional
dissipation of the momenof the pressure equation
dissipation
of the
property
a factorizable
am scale
Note that the is tile gradient
and
is this
of sound,
( is a length
(9a)
1
The
and
quo
covari-
components '2 1
transformation
IglO_+2sgn_'f'OeO,+_O,,]
(12a)
when 1 '1> lv'l and *, W¥iting the
the
advection
covariant
containing scheme
velocity the
may
y,_
=
operator
" in
upgraded
(6)
and
ignoring
geometric to
(8)
Eq.
components
higher-order be
v
nearly
in terms
qHO ----_0_
"k-VO,7
of - 21 k(C20_ _
terms
derivatives,
the
second-order
by
*dg_
+ 2 sg,,t 7- c-o{o,,
= _Yn - dxn, ggn = __y¢ + jx_
+ lvl o,g )
(a2b)
The
when l_?[ < I_::[. With t.his advection operator in the momentum equation, the quantity D in Eq. (11) must be modified in order to preserve the factorizability of tile scheme.
This
modified
D is given
-= D + lsgn_"
Dm_
following
difference
_" (O,L? + c3¢9)
g
=
0
o.=_
-4
sponding
operator,
dissipation
Eqs. (6), (7) and (9) is related to the standard upwind difference scheme, consider a uniform scaling the
equal
spacing
coefficients
grid
dissipation standard first-order replaces
in the
a,,
spacing,
and
and
x and
#v are
the
taken
terms
upwind
scheme
on
a nonmfiform
where
the
addition given
expressions
a six-point vertices,
may then
taking
-4
,
the
operator
of Eq.
cross-derivative
upwind
(9) on terms
a Cartesian
yields
grid.
the standard
-
2
the subscript h to the corre-
of the
overbar The
in flux
a two-point
( )
secondform
on an edge
by
between
difference
of
on a vertex. The subscripts e difference operators centered
respectively. as
f is
-
above.
centered
centered to denote
on an edge or a vertex, scheme is then written
,
be expressed
difference
and
-20 -
qhs
The
fully
discrete
= 0,
(14at
the The
further
p
the grid = O,
spacings in the two coordinate directions. The multidimensional upwind scheme uses a single length scale, and retains the cross-derivative terms. Consider the advection
condition:
-
scheme, difference
differences
in the
grid
f_ and (u are
and
"wide"
those expressions and v are used
If the
terms of Eqs. (6) and (7) are ignored, first-order upwind sdmme is obtained. (_" by g_.0_ + lyon,
two
first-order Cartesian
to be one,
cross-derivative
taking
of
g directions.
the
difference
when IUI < I% Taking tit(" curl of the momentum equation now gives qno w = 0, i. e., the vorticity equation is now second-order accurate. upwind
this
0-
write the complete discrete is used to denote a standaxd denotes
with
_
To
f-7 (O,7_ -4- O I1"1and DHC) = D -4- lsgn_>
satisfy
' '[i' i] , ,[!,!] '[! !] _
by
( 13at
To see how the
operators
Dropping
p c2
_Z h . ff
___ ff
._
(14b)
h p
first-order
discretization.
.
h-
1
(V__F,_)p)
=0
'
(1@)
Discretization The The factorizability of the FDA is a necessary condition for the factorizability of the difference scheme, but. it is not torizat,le, same
sufficient.
the
the difference
way as the
difference
For
difference
operators
differential
must
opera!ors
approximations
scheme
to the partial
8.
Introducing derivative
in the
advection
the oper-
part
of DH-_ given
wide,
six-point
way,
qHo
and
must
depend
be
upon
stencils
for the general
the appendix The
0_,,=0_+...,
to this
scaling
in Eq.
difference
discretized
the
curvilinear
(ll)
stencil
terms of DHO and for the second-order flow
grids
in a very
direction
the relative magnitudes of Cr and V. The described in Ref. 12 for uniform Cartesian
0,7 =0,_+---,
4', =0_ +...,
the
operator
precise
at, ors, =0¢+..-,
in that using
for 0_/_ and 0_V. The additional the corresponding dissipation terms
to be fac-
commute
derivatives
are discretized
and
details grids.
are The
are presented
in
paper.
coefficients
are 1
0-,_ = max(._/, where the following
conditions
must
Mach
hold
_I
chosen • _h
.-,h
, h
• h
stagnation
cJ_cJ,m = OCnO_n h
h
,h
is the
number
local
to be
O(h),
points.
skewed
4
Laplacian
_rv -
Mach
to prevent
pressure equation tic factor in the
,h
Me),
max(M,
number,
division
and
essentially
The
purpose
M_)
and
/ff_ is a cutoff
by zero.
The
becomes of the
(15)
cutoff
active
rescaling
dissipation, crp, is to prevent the discrete equations from becoming operator
in the
incompressible
is
near of the ellipthe limit.
Currently, we take Mc and AM is the two-point
= v IAM[, where difference in Mach
on an edge.
flow cases
For
tile
channel
u = 5 number
shox_m below,
never becomes larger than M. For the leading-edge the cutoff does become active at the stagnation When
Me
patton
> 3/I, it is necessary
to the
dissipation
advection
where tion.
momentum
coordinate
to both
as_d is cast
maintain
conservation.
optimize
either
ator
of tile
is added
equation, the
coefficient
(16)
transforma-
the entropy
in flux form
No at tenq_t
wilt.ten, the Eq. This is because
artificial
dissipation
in order
to
made
to
ha._ been
v or the form
and the
of the oper-
(14) is valid tile pressure
terms
are
only for subsonic differences in the
not. fully
upwinded
in a
supersonic zone. A simple modification to Eq. (14) can be made by rescaling the gradients of the pressure when the flow becomes defined as
sonic.
Introducing
the
parameter
n,
This
=max(l, final
form
of the
M2),
scheme q
h
artificial
dissipation
in
as
1
h .
(17)
is a conservative
terms
of the
square
bracket
form
primitive fluxes.
(20)
_
h
,
s - d_ps = O,
dissipation
The
and
These
terms
the
antidissipative
recast
terms
in conservation
are
The
length
scale
ing min((e¢,
/_h(/_,9)
form
discretized
( is evaluated
In addition (14b) terms
to the and
on the
are
discretized
(20)
o-v are
that.
using
the
a term pressure
be
by tak-
grid direction. evaluated
using
examination
the
pressure
wide
part of the conservation to account for these
may
cell face
terms,
shows
differ@ and
on cell faces.
in each
dissipation
(14e)
central-difference must be corrected
in Eq.
length
The scaling coefficients a,,, and the Mach number on the face. Eqs.
in the
on cell
or narrow operator
and evaluated
g,1), the shortest
cast inside
(21) are now interpreted
terms
faces according to the appropriate wide ences. The first-order upwind advection
This is done by adding for the momentum and
is
of the
variables.
in Eqs.
as dissipative
ent
the
The
can be rewritten
of this
d_p.
As flows.
(14).
(14c)
+V,_
dissi-
form
Eq.
Me
c( (0_ + 0_), 7
Mc-M)
Jacobian
operator
Tile
and
(14b)
pseudo-Laplacian,
1 _max(0,
J is the This
qHo.
in
Eqs.
flow, point.
additional
h
operator
is a five-point
d_p =
to add
operators
of gradi-
stencils.
The
equations (19) wide differences.
to the dissipative fluxes equations as follows:
(18a) p
qhl(,a
-- d_p_ + V,,D_,,
+ _lvhp P
O'm(_Th " (pc2_72"_-I-
pC2_Th.
([
+
and
o_+_
the
(14c),
_"u'_rep)2pc
can
+v_
difference be
)
(18e)
(14a),
(14b)
p =o.
equations
written
as
a central
equations
finite-volume
are diseretized
using
difference
by the
transformation
+
vhp
The
e = dissipative
in terms
a.ff/2
+ P/(O(_ fluxes
of (s, u,v,p)
on
each
using
the
conservation
p o
o
eval-
variables
,,/d] ,,lal
,_. _,, "
h/cV \41
0,,
where
h = e + p/p
is the
total
Solution The
solution
Procedure
of the discrete point found
ent.halpy.
equations
is performed
collective Gauss-Seidel t.o be a very effective
us-
iteration, smoother.
= 0,
(19a)
The grid point.s are ordered such that. the forward GaussSeidel sweep is in the streamwise direction. This means
=
(19b)
forward
(19e)
relaxed
0 ....
v_.[(p_ +v)a] = 0
where
to the
-P"I" -p,,Is
ing a symmetric which has been
a central-difference
(Ss, 5u, Jr, 5p) have been
converted
(24)
that l_rh.(pff_)
are
fluxes
they
part a conof the
approximation, Vh-(pa)
the dissipative
uated,
I L_o_'i = I
plus a dissipation, it is straightforward to obtain servative discretization. The conservation form Euler
(lSb)
ff._rhp
-peT,,,.._u. Because
Once
= o,
1)) cell
is the face
appropriate
total are
the
entropy
equation
sweep. with
The a factor
of the
conservation
puted
at each
is marched
reverse of 0.5 using
computed
in the previous formed to the
section. residuals
difference
inverse
transformation
energy.
of the
sweep
during
The
(r¢,, ro,,
ro,,, rp_) are
the
discretization
These residuals of the primitive in Eq.
(24).
the
is under-
for stability.
equations
vertex
in space
Gauss-Seidel
residuals com-
presented
are then transvariables by the
flow
ff
[_
p
.-ighl
=
I
1 1
Ic,_gth
Figure
At each
point,
the
is the
matrix
collecting
the contributions
entropy this
equation
is a block
decouples
is easily
The
of the
coefficients
froln
matrix
entry, and multiplying
found
relaxation
faces.
Because
At
the
the
rest
of the system,
the
upper
is accelerated Scheme
(FAS)
block
is
using
dissipation
a standard
multigrid.
Full-
A sequence
ator, u be the vector be the fine-to-coarse
of the conservation variables, grid restriction operator, and
be the coarse-to-fine
grid
tion
equations,
to the
of
grid
solving
This
procedure
=
(2G)
equation
for Uk-!,
the fine-
Multigrid tion.
performed pletely.
the
recursively
coarsest
to
to solve
insure
grid, that
A conventional
to the
many the
same
relaxation
fine-grid
equation.
coarse-grid
relaxation
equation
Iz cycle
Solutions
are
shown
channel,
is solved
are com-
is used.
Pig.
shape
1. The
of the
lower
wall
is shown
between
in
0 < z
O.
.h-
l 1
Sidilkover,
D., optimally
the inviseid vol.
Roberts,
approaches
efficient
4-5,
pp.
551-557,
Multigrid
Paper
99-3337,
for
and Fluids,
1999.
R. C.,
to
1
( _/'i+
take
Ideally
Airfoil
(Pi+2,j + "9,+2,j-1 - P,+l,_ - 13',+l,_-i ),
o2v = 97 Fourth,
"Extending
Methods
,
con-
solvers
Computers
T. W., Swanson,
Converging
towards
multigrid
flow equations,"
28, nos.
AL4A
"Some
- 1Ai+to-_)
O,U = 97(lgi+,o
1994.
structing
11.
Equations,"
Third,
2,3q_
l
--
"171__2,3__1)
.
(r < 0, 17"< 0.
1
Flows,"
(90h/_
=
_
(/_/+l,j+2
--
a/+l,3
),
1999. ;, c, c, 0_'P = 1_ (L+2,_+_ + L+2,_ - I;,+_,_+_ -
12.
Sidilkover,
D.,
Steady-State in Two
"Factorizable
Equations
and Three
Sdaemes
of Compressible
Dimensions,",
for Fluid
ICASE
the Flow
Report
The
Discretization
diseretization
for DHO, given
of the
in Eqs.
(13)
was so as to preserve scheme. In Ref. 12, sions
for a uniform
in
(13a),
derivatives must
Cartesian
in the expression
repeated
here
Ohol_ = 21 (lli,j
- ll,,j__
(i,j + 1/2}. the following
+ _,__.a
First we difference
-/_,-_o-1)
,
be done in a very. precise
grid.
here.
Now consider the o-cell edge, take (Tr > 0, 1-"> 0, in which case fornmlas are used.
DH¢_
the factorizability Sidilkover presents
of I/.771> ]_7"1is presented by Eq.
of
1
a._ = _ ff"+_,,+' - _,+_,,__).
preparation. Appendix:
9i+1,3)
of the these
The
discrete expres-
particular
In this case,
1 (_i,j
case
- "13',__,j) ,
-,_ah_'-= 971('_i-l,j+l--_i--l,3+_)i-";+l--_)i--2,3)
DHo is given
for convenience.
Second,
take
(r > 0, {-: < 0.
1 O_a
Duo
=-- O + lsgn(;
_" (OJ2
_-
2
(l_i'Jq-2
--
U"3"_I
-{- l_i--l'jq'2
=
1 (V_,,+_ - P,-_j+_),
--
/_t--l'3-}-I
)'
+ _¢_')
1
_,_,_ = -_(v,_,,.+_ - _',_,,. + 9___,,+, - P___,.) The
term
tion
section,
D has already and
been
what
discussed
remains
above.
Each
of these
a _-cell
edge,
(i + 1/2, j),
terms
is discretized
and
where i and j are the indices directions, respectively.
in the Discretiza-
is 0,1/_,
an ,t-cell
0_'
and
differently edge,
of the vertices
Third,
0_' on
in the _ and
*1
(o,,, -
1
a29 = ,hr,
_ f._,,.
l
- V,_l,,_, ).
_ 9_-_,_-,),
take
U > 0, V < 0.
h-
Fourth,
take
r,
(r < 0, 17"< 0.
|
1
r.'__,,._,)
0,+_,_+_ +0,o+z-
&o+,)
-,
4"v = 97(v,+2,,+1 - _',,,+,), 1 0._ ' = 97(_,+_,,+, - _',+_,. + 9,+_,.+, - _+,,,) above
definition
= 7 (&,j+_ - &,,),
1 (>,_,
- >,,,),
1
Analogous
a.u
+ a,,, - a,,,_,),
O,V = _ (_'_÷_,_÷_- _,'_+_,_+ V_÷_,,÷_- _;_+_,_)
The Second,
1
0,,_ _ - = 971 (&+_o+_-
1 (_.,. + _. • h -,
U" < 0, 17_> 0.
= (a,+,,, - u,+
(i, j + 1/2),
Consider the _-cell edge, (i+1/2, j). First. we take U > 0, 17_> 0, in which case the following difference formulas are used.
o,, v = _ (_',_,,,+,
takc
corresponding
differences
are
of the second-order expressions to lf:l
may >
[/?[.
also
the
advection be developed
ones
used
operator for Eq.
in the qno. (lab),
1,1E-11
1.0E-
11
doe_
-0.0002
1
9.0E-12
8.0E-12
_5
--o
Entropy
t-0.0004
Drag
-I
-0.0006
7,0E-12 -4
-_ -0.0008 o _
6.0E-12
v
5,0E-12
-0.0m 0
j_
-0.0012 -6
4.0E-12
-....
L,(rho) L,(r_o-u)
.......
L2(rho-v
)
L2(rbo-e) I @oe_ 3.0E-12
2.0E-12
iI
-0.0014
o _° t
I
' ii
-0,0016
O.OE+O00
-8o
-O.O01B
1.0E-12
.... ,'o.... 4 .... 4 .... 4'0.... ;0 V(1,1)cycles
10
20
30
40
5O
V(1,1) cycles Figure 7. Convergence of t.he L2 entropy error drag on the lower wall for a M = 10 -4 inlet using a FMG cycle with a 385 × 129 fine grid.
Figure 9. flow using
and flow
Convergence rates a FMG cycle with
for a M = 0.73 inlet a 385 × 129 fine grid.
0.0024
5.0E-04
0.0022 4.5E-04 0,002 4.0E-04
0.0018 _
--o
Entropy Drag
] 0.0016
3.5E-04 0.0014
\
3.0E-04
00012 I
_
2.5E-04
_r'
2.0E-04
Do.o
0.001 0 0.0008
-, %g/
0.0006
1,5E-04
0.0004
1,0E-04
o
I p _I
5.0E-05
I i O.OE+O0
0.0002
,Z' g _ r
0
i l 10
i/
$ i
-0.0002 i
,
i 20
....
V(1,1
Figure 8. Mach number contours for a ,_I = 0.73 inlet, contour increment _,"ll = 0.025, for a 385 × 129
l 30
, , ,
, I .... 40
Figure 10. Convergence of the L.. entropy drag on the lower wall for a M = 0.7a
grid.
using
]0
a FMG
cycle
with
0,0004 50"
) cycles
error and inlet flow
a 385 × 129 fine grid.
m
p v
-4
o
-k,(rho) .... k,(rho-u) ....... k,(rho-v) ...................... k,(rho-e)
10
20
30
V(1,1
Figure edge
Figure
11.
Semi-infinite
parabola,
13. using
Convergence a FMG
rates
cycle
40
50
) cycles
for a M
with
= 0.1
leading-
a 129 x 129 fine
grid.
33 × 33 grid.
7.0E-04 6,5E-04 6.0E-04 5,5E-04
i
--o
Entropy
]
5.0E-04 4.5E-04 A 4.0E-04
E
3.5E-04
Q 3.0E-04 2,5E-04 2.0E-04
C;7
1,5E-04 1.0E-04 50E-05
O.OE+O0
o
10
20 V(1,1
Figure Figure M
=
a 129
12. 0.1
Pressure leading-edge,
coefiicient increment
contours AC' v =
around 0.05,
a
M
for
=
14. 0.1
Convergence leading-edge
a 129 x 129 fine
× 129 grid.
ll
grid.
30
40
50
) cycles
of the flow
L2 entropy
using
a FMG
error
for
cycle
with
a