0% found this document useful (0 votes)
39 views21 pages871780
Uploaded by
Peter PedigrewCopyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF or read online on Scribd
0% found this document useful (0 votes)
39 views21 pages871780
Uploaded by
Peter PedigrewCopyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF or read online on Scribd
/ 21
INTERNATIONAL JOURNAL FOR NUMERICAL METHODS IN FLUIDS, VOL. 17, 301-321 (1993)
BENCHMARK SOLUTIONS FOR THE INCOMPRESSIBLE
NAVIER-STOKES EQUATIONS IN GENERAL
CO-ORDINATES ON STAGGERED GRIDS
C. W. OOSTERLEE, P. WESSELING, A. SEGAL AND E. BRAKKEE
Faculty of Techical Mathematics and Informatics, Delf Univesity, P.O. Bax $031, 2600 GA Delf, The Netherlands
SUMMARY
Benchmark problems are solved with the steady incompressible Navier-Stokes equations discretized with
4 finite volume method in general curvilinear co-ordinates on a staggered grid, The problems solved are
skewed driven cavity problems, ecently proposed 2s non-orthogonal grid benchmark problems. The system
of discretized equations is solved efficiently with a non-linear multigrid algorithm, in which @ robust line
smoother is implemented. Furthermore, another benchmark problem is introduced and solved in which
4 90° change in grid line direction occurs
KEYWORDS Benchmark solution Incompressible Navier-Stokes Staggered grid General co-ordinates Multigrid
1, INTRODUCTION
In recent years several methods have been proposed for the discretization and the solution of the
incompressible Navier-Stokes equations in arbitrarily shaped domains with a finite difference or
2 finite volume method, Most publications adopt a boundary-fitted curvilinear co-ordinate
system. The discretization methods which employ curvilinear co-ordinate systems differ in grid
arrangement (staggered or non-staggered grids), and in choice of velocity components (Cartesian
or so-called grid-oriented velocity unknowns, such as contravariant components). A brief over-
view of some important groups solving the incompressible Navier-Stokes equations is given here.
A distinction is made between groups solving the incompressible Navier-Stokes equations in
_general co-ordinates on staggered and colocated grids and researchers using a pseudo-compress-
ibility approach, First, investigations on colocated grids with Cartesian velocity components as,
dependent variables and transformed velocity components as independent variables are enu-
erated, In Reference | this choice is made, where further a comparison is given between the
several choices to be made to get accurate discretizations. The pioncering papers of the colocated
approach are by Rhie et al A lot of research on discretizations for non-staggered grids with
Cartesian velocity unknowns is also presented in References 4-6, where the set of benchmark
solutions discussed here is proposed for incompressible Navier-Stokes equations in non-ortho-
gonal domains, Other papers based on the colocated approach are References 7 and 8 where
results with advanced upwind schemes on colocated grids and fast solution methods are obtained,
In Reference 9 non-staggered grids are adopted to investigate the effec of several upwind schemes,
(0271-2091 /93/160301-21815.50 Received 10 September 1992
© 1993 by John Wiley & Sons, Ltd Revised 11 February 1993 302 CW. OOSTERLEE ET AL,
and turbulence models on the convergence of a multigrid solution technique for the incompress-
ible Navier-Stokes equations. Some commercial codes have adopted colocated grids for solutions
of complex flows in arbitarily shaped domains."° A second group of researchers adopt pseudo-
compressibility methods. With an artificial time-dependent pressure term in the continuity
equation the incompressible Navier-Stokes equations can be approached with methods from
compressible Navier-Stokes equations. All unknowns are also stored in the centres of grid cells;
Cartesian yelocity components and pressure are dependent variables. This method is, for
example, adopted for discretizations in curvilinear co-ordinates by Kwak etal!" "? An advanced
upwind scheme based on flux difference splitting is incorporated in Reference 12.
Discretizations on staggered grids are also applied on a large scale. Interesting papers have
been published by Rosenfeld et al," where on a staggered grid contravariant fluxes V* and
pressures are used as primary unknowns. Two- and three-dimensional results with moving grids
and multigrid acceleration are described in these papers. In Japan, staggered grids are widely
investigated in Reference 16 for two-dimensional and in Reference 17 for three-dimensional
problems with dependent variables contravariant fluxes V* and contravariant vorticities. In
Reference 18 contravariant physical components and pressure are used as unknowns. A similar
approach using the same unknowns can be found in Reference 19.
Some other papers will be mentioned briefly. An analysis of the treatment of a pressure
equation on a curvilinear staggered grid is presented in Reference 20, In Reference 2! the stability
problems on staggered grids with Cartesian velocity components are investigated and remedies
are given. In Reference 22 the staggered approach is chosen for solving incompressible
Navier-Stokes equations in curvilinear co-ordinates, because of the merits in boundary treat-
‘ment, the compactness of eflective mesh spacing, and the insensitivity of solutions with respect to
the values of relaxation factors used in the iterative procedure. Approaches different from all
others are described in Reference 23, where on a staggered grid covariant velocity components are
used to solve the incompressible Navier-Stokes equations, and in Reference 24, where a rectangu-
Jar co-ordinate system is used with special treatment for the boundaries. In our group progress
has been made with the discretization of the steady and unsteady incompressible Navier-Stokes
equations in general co-ordinates using a finite volume method on a staggered grid.?~?7
Accurate results were obtained with Dirichlet boundary conditions**-* for the steady equations
with a multigrid solution method and also for the unsteady equations using (semi-) natural
boundary conditions?® with a pressure correction method and a GMRES solver. The incom-
pressible Navier-Stokes equations were discretized on a staggered grid with contravariant fluxes,
V* and pressure as primary unknowns.
Recently, two benchmark problems were proposed in Reference 6 for discretizations on
xnon-orthogonal grids in two dimensions in order to compare different discretization methods.
The Navier-Stokes equations were solved on a non-staggered grid with Cartesian velocity
unknowns; the discretization is described in Reference 5. The solution method is a SIMPLE-type
method.*"™! Discretization in general co-ordinates on a staggered grid is a far from trivial
‘matter, and may be very inaccurate, unless certain precautions are taken, discussed in References
25 and 26. In order to validate our approach, in the present paper the benchmark problems of
Reference 6 are solved with our steady variant. A steady-state multigrid solution method is used
in which velocity and pressure are solved in a coupled technique by means of the Symmetric,
Coupled Alternating Lines (SCAL) smoothing method.>? Furthermore, a third benchmark
problem is introduced and solved. In this problem a 90° change in grid line direction occurs,
4 situation which is difficult to handle with Cartesian velocity components as unknowns, and
brings out the advantage of the co-ordinate-invariant approach of References 25 and 26 and the
resent paper. BENCHMARK SOLUTIONS FOR INCOMPRESSIBLE NAVIER-STOKES EQUATIONS 303,
2. THE INCOMPRESSIBLE NAVIER-STOKES EQUATIONS IN GENERAL
CO-ORDINATES.
‘The fundamentals of tensor analysis, especially i relation to continuum mechanics are presented
in textbooks such as Aris? and Sokolnikoff. The discretization is described in more detail in
References 26, 29 and 35.
In general co-ordinates the steady incompressible Navier-Stokes equations are given by
us=0, w
TH =(pUU\g+(oP)g— 7h = PFs, @
where 1% represents the deviatoric stress tensor given By
eap(gU$+ 9? U5), ®
Here U* are contravariant velocity components, p is density, p is pressure and is the viscosit
coefficient, Unknowns V*=,/g U" are used as primary unknowns together with the pressure. The
arbitrarily shaped flow domain Q is mapped onto a rectangular block G, resulting in boundary.
fitted co-ordinates. The co-ordinate transformation is given by x=x(@), with x Cartesian
co-ordinates and & boundary conforming curvilinear co-ordinates. The covariant derivative
formula used for the continuity equation is
2a dfaut “
a Tg ae
Terms in the momentum equations of the type T7f are given by
6)
where {;5} represents the Christoffel symbol of the second kind.
The equations are discretized with a finite volume method on a uniform staggered grid in G.
We introduce local cell co-ordinates given by Figure 1, which shows part of the computational
arid in the E-plane. Taking a cell with centre at V'-point (1,0) as an example, finite volume
discretization gives, using (5)
[rv [ Panes] ce
ar
Ja TenssaereilaT iis set+( Jaf hh rs )attaee ©
02 2
I I '
blo gf ~ at = at = at
1 i |
Lo og 10 20 30
t 1 1
et = Ot Sat = at a =
Figure 1, Loca cell co-ordinates 304 ©. W, OOSTERLEE ET AL.
The second component of 2) is similarly approximated with a cell centred at a V-point. With
T7 from (2) this is the discretization used for the momentum equations. It is found that the
variable V*=,/gU* appears naturally in many places in (6). The convection tensor is linearized
using a Picard iteration
(BUA) g=(pU= TPM, 0
Where the superscript n is an iteration index.
‘The convection term is discretized with the so-called hybrid discretization schem
Depending on the mesh Reynolds number Re" (ie. the ratio between the absolute magnitudes
of the flux part of the convection term and the viscous term in point (i j)), the flux part of the
convection term is discretized with a central difference scheme (when Re/< 1) or with a firs-
order upwind scheme (when Re“> 1), There is a smooth switch between the two schemes using.
a smooth switching function (Re,
‘The total number of variables linked together in a momentum equation is 19. The discretized
equations are solved with the standard non-linear multigrid method. *?” Details are presented in
References 38 and 29. Here the smoothing method called SCAL™* is described briefly.
All velocity components and pressures in a line of cells are updated simultaneously. This is
a line-by-line version of the cell-by-cell smoother SCGS introduced in Reference 39, a so-called
coupled smoothing method, because different unknowns (velocity components and pressure) are
solved simultaneously. SCGS is used as smoothing method for discretizations in curvilinear
co-ordinates in References 40 and 29. The smoother SCAL is much more robust than SCGS.
arbitrary domains, where cells with large aspect ratio occur have been solved
1030131
successfully.25
For each line a banded system is solved and intermediate values (V"*, ¥%, p*) are obtained,
With underrelaxation the new values (72*, 72", p*0) are found, After an update of
unknowns along horizontal lines, unknowns are updated again using vertical lines. SCAL is
a zebra-type smoother: first all odd (white) lines are visited, then all even (black) lines are visited.
With special ordering strategies acceleration can be obtained on vector computers.
3, BENCHMARK SOLUTIONS
In Reference 6 two benchmark problems are defined for discretizations on non-orthogonal grids.
The flow problems arc skewed-driven cavity problems. The domain, depicted in Figure 2, is
parallelogram with boundary length L= 1. Angle fis 45° for case 1, and 30° for case 2, so that
highly non-orthogonal cells occur in the x-image of the grid. The top wall is moving with
Cartesian velocity components u! = 1, 1? =0.On all other boundaries w= is prescribed. The flow
problems are calculated for Reynolds numbers (Re) 100 and 1000. The discretization is
Figure2, Domain forthe skewed-driven envity problem BENCHMARK SOLUTIONS FOR INCOMPRESSIBLE NAVIER-STOKES EQUATIONS 305,
ated on four grids, consisting of 32x32, 64x 64, 128x 128 and 256x256 cells, respectively.
The streamline patterns obtained agree closely with those in Reference 6. They are presented in
Figure 3 for =45°, in Figure 4 for =30°. The prescribed streamline values in these figures are
identical to those in Reference 6. The maximum and minimum values of the stream function
Wao and Pain are determined for all grids together with their co-ordinates (xin, Xan) and
(shi. XB) and an error measure ¢, defined as in Reference 6:
oo [ Paget +100, ®)
Figure 4. Siecanines for Re=100 and Res S000 for testeate 2 306 ©. W. OOSTERLEE ET AL,
‘Table I. Minimum and maximum stream function values in vortex eentres and their postion for all grids
tand for both Reynolds numbers, skewed cavity, B=45"
Re= 1000
Min Max
32%32
¥ =70260% 10-2 S61S0x10"* = 46914x10-# 73515 x 10-9
x 11149 03422 12933 7415
y 05524 01587 0574s 03978
64x64
¥ = 7026x102 1088 10"$ = SUTTBXIO-F 846275 10-8
x 11039 03468 13089 07682
y 544 01436 05745 408k
128% 128
¥ = 7025310"? 37725 10"$ = 58456 10°F 1.0024 10-#
x 11094 03401 13089 07783
y 05496 01436 574s 04033
256x256
¥ = 7023810"? 6932 10-$ 53523 10-F 1.0039 10-F
x 11400 03390 13128 7775
y 05469 01409 5745 04005
e oat 21472 1282 1404
‘Table II. Minimum and maximum stream function values in vortex centres and their postion forall grids
and for both Reynolds numbers, skewed cavity, )=30"
Re=100 Re= 1000
Min Max Min Max
2x32
¥ = 53045 x 107? LSS97x1O-# 397309 10-4221 10-9
x 11808 o5519 14537 08976
y 03750 01563 04063 02656
64x64
y =53126x10-2 74524 10-F 3883310"? 44938 109
= Ti6si 03750 14537 09997
y 05384 o14es 4063 02578
128% 128
w 5315x1072 $9527 «10-# 3869810"? 4.2446 10"F
x Lins 05316 14526 0.8997
y 03789 o-t4ss 4102 02578
256x256
y =53149% 10-2 5629810 3 8600x10-F 41657 10"*
x 11680 05291 456s 09036
y 03789 01426 4102 02559
@ 00060 58672 02539 8940 BENCHMARK SOLUTIONS FOR INCOMPRESSIBLE NAVIER-STOKES EQUATIONS 307
‘These values are presented in Table I for f=45° and in Table If for f=30", Results agree well
with Reference 6, The local error is less than 003 per cent of the maximum of the absolute values
of ¥ over the domain for B=45° and less than 0+! per cent for
all grids are accurate, The values for the 128 x 128. and the 256 x 256-grid do not differ much for
Re= 1000. This indicates that the exact solution is closely approximated on these grids. This
is also true for velocity profiles along the centrelines CL and CL; (Figure 2). Figure 5{a) shows
the Cartesian velocity component u? along line CLy for Re=100, f-=45°; Figure 5(b) gives u?
along CL. Figure 6(a) shows u! along line CL, for Re~= 1000, 6=45%; Figure 6(b) presents u
along CL. Figure (a) shows u? along line CL, for Re=1000, 8=30%; Figure 7(b) presents
w@ along CL. For Re=100 all curves are identical; therefore, the velocity profiles for testcase
2are not shown. This indicates that we have an accurate solver; for a small number of unknowns
accurate results are obtained for low Reynolds numbers.
‘Again for Re=1000 the finest two grids show identical results. On the other grids the effect of
the upwind part of the hybrid difference scheme can be observed; differences can be seen mainly in
the high-velocity regions. Due to these satisfactory results the discretization is called accurate;
similar results were found when other benchmark problems in rectangular domains? were
solved,
Next, the results for the multigrid solution method are presented for these testcases. Average
reduction factors Hi are calculated, defined as
ya 88-2
a(S ty 5 o
Tres,
with [the Euclidean norm, and res, the residual after n iterations. We also give
(19)
In many cases », is found to be approximately constant for n close to nit, in which case we have
found the asymptotic convergence factor jin2¥a4- The number of MG iterations that were
performed depended on the reduction factor ini, a8 follows:
015 03, nit 25,
‘The multigrid cycle used is the F-cycle, because it showed reduction factors comparable to the
\Weayele, while itis much cheaper than the W-cycle, The even cheaper V-cyel sometimes showed
bad reduction factors for Re= 1000 on very fine grids. The number of pre-smoothing iterations
and the numberof post-smoothing iterations is 1, The coarsest ri always is the x 2 ari, where
10 smoothing iterations are performed. The number of multigrid levels differed for the different
grids from § to & Table II presents 4, and vqi for testcases 1 and 2, The CPU times for one
Feycle on a Convex 3820 computer are presented in Table 1V. No parallelization statements or
special ordering strategies for vectorization purposes are implemented in the code.
‘We introduce another benchmark problem, because the grids oftescases | and 2 mainly test
the effect of non-orthogonalty of co-ordinates, In order to also include the effect of curvature of
co-ordinate lines (sch testases are also presented in Reference 6, but only for buoyancy driven
flows), we propose the problem of low in an L-shaped- driven cavity Ths test problem i not only
interesting for testing discretizations, but also for domain decomposition techniques, a used in
Reference 43. The domain is depicted in Figure 8, It can be seen in Figure 8 that boundaries 05}
os
07;
0
os
oa
03}
03}
on
d
“a2 0 02 or 06 08 1
2 sresaly, 2561 256- si,
fi results, 128 12% gi
ruts, 64x64 er a)
= sults, 32932 i
— 0 0a 0 reference rel from Demir ea,
Ww
03
03}
on
|
ul
-02|
03 ed
0 Or 02 03. «Od 0506 7 OD
2
»
fi
x
Figure 5, Velocity profes for 945": Re~ 100: (a)! along CL; (8)4* along CL 2 tet, 286 256
ests, 128 12 ri
res, 64 x 6: a
rel, 32432 fr
refonce rele fom Dei a
|
01
o4
oal-
se ee ee eee
‘001020304 03 06 07 On 05
2
»
a
Figure 6, Velocity profiles for B= 45% Re-= 100; a) u! along CL,s(b)x? slong CLy
1 as)
oa
o7
06
as
oa
a3]
a2]
on
“02 o ‘oz Os 08 os 1
revue, (28x (28- ri.
resi 66x 6 a a
e sul, 32 32-
= sooo reference rls rm Demi eta
a
03
034
ou
or 0203 04 05 06 07 08 09 1
»
Figure 7. Velocity profiles for f=30°: Re
(000 (a) uéalong CL; (6)u# along CLy BENCHMARK SOLUTIONS FOR INCOMPRESSIBLE NAVIER-STOKES EQUATIONS 311
Table II, Average and asymptotic reduction factors for skewed-driven cavity problems
Angle pas pam"
Reynolds number 100 100 1000
Hai ee
32x32 grid O19 0221 0617 0695 03040372 0629 ass
64x64 grid O14 02480712 O718 0336 0412 0533 0589
128% 128 grid 0198 0253 O718 07910345428 ©0530 0607
256256 grid 0194 0250 0604 07190352048] O44 0555
‘Table 1V. CPU times for an F-cyele on a Conver 3820
Grid 32x32 64x64 I2BI2B 256x286 512x512
CPU times 134s 4675 i145 752s 3500s
Figure &. The domain for an L-shaped cavity problem
T, and I’, consist of two parts. On the upper part of P; w= 1, u?=0 is prescribed, on all other
boundary parts we have w=0. The Reynolds numbers investigated are again 100 and 1000. If the
tid is non-smooth, like the one shown in Figure 9, our discretization is not accurate for high
Reynolds numbers (Re=1000). Figure 10 shows for this grid streamlines for Re=100 and
Re= 1000, It can be seen that the non-smoothness of the 128 x 128 grid causes artificial non-
smoothness in the streamlines near the kink in the grid lines for Re=1000. Inaccuracy of the
discretization on the grid of Figure 9 is tobe expected with the method chosen dus to the presence
‘of Christoffel symbols, which involve second derivatives of the co-ordinate mapping. More
smooth grids are needed, like the grid in Figure 11, constructed with a biharmonic grid
generator.“*5 A generating system of higher order, like the biharmonic grid generator allows 312 CW. OOSTERLEE ET AL
Figure 9. A non-smooth grid in an L-shaped channel
Figure 10, Streamlines for Re= 100 and Re= 1000 for an L-shaped cavity with the grid of Figure 9, 128 128 cells
Figure 1. A smooth grid, in an L-shaped channel obtained witha biharmonic grid generator [BENCHMARK SOLUTIONS FOR INCOMPRESSIBLE NAVIER-STOKES EQUATIONS 313
‘more boundary conditions, Co-ordinate line angles are specified to be orthogonal at the
boundary. (With a Laplace grid generator we also obtained smooth grids) Figures 12 and 13,
present streamlines and isobars for Re=100 and Re==1000, respectively. Local maximum and
minimum values of the stream function Wu atid Wig are presented for all grids together with
their co-ordinates (xijay X2iq) ANd (Xhayy X4ax) in Table V. From Figure 13 it can be seen that the
1 azeee
2 4meos
3 8506-0
42020
5 200802
6 s10802
7 43060
4 s896.02
9 sa0800
1 8006-29
1 2906.05
12 (4806-05
13 34406-06
Figure 12. sober and streamlines an L-shaped cavity for Re= 100 with the grid of Figure 11; 256% 256 cells
1 42000
2 as060
3 1680602
4an0en
5 s9aovon
6 some.
7 6500-02
539060,
9 1000829
10 4250504
11 14s06.03
12 3440603
1 $2306.03
Figure 13 Isbers and sueamlies for an L-shaped cavity for Re=1000 with the grid of Figure U1; 256. 256 ells 314 CW, OOSTERLEE B7 AL
‘Table V. Minimum and maximum stream function values in vortex centres and their position forall grids,
L-shaped cavity, for both Reynolds numbers; grids similar to Figure I1 (for Re = 100 the finest grid consists
‘of 256 x 256 calls)
L-Shape Re= 1000
Max Min Max (05)
32x32
¥ 81183107? 39308% 10 61571 x10"? 43821 «10-9 35184x 10-9
x 06737 07518 07326 01904 06589
y ogie4 01720 07326 06895 02521
64x64
¥ 81073 x10"? 29953 10-* —73215x10-* 2854x109 6:3619x 10-9
x 06679 07599 07026 onrsas 06608
y o8i34 077 07503 07289 0.2886
128 128
¥ —80915% 10-2 26312x10~4 83419 10"F 6700510"? 58084 x 10?
x 06763 07643 06983 182s 6855
y 08092 0147 07464 07525 03009
256x256
¥ =80859x 10"? 25814 107F —85392x 10"? 62021 x10"? 2503x1079
x 6734 07688 06947 01819 06877
y 08127 173 07488 07505 03069
e 00626 1928 = = =
512xsi2
y = 8542510"? 6402210" 62712 10-
x = = 06938 1s 06868
y - = 07509 o7sI5 03089
: — : 00386 3126 0333
flow at Re= 1000 has two stream function extrema, one for x4 <0'5 and one for x! > 0:5, Both are
given in Table V. In order to judge accuracy as well as possible a 512x512 grid is needed for
Re= 1000. One should keep in mind that the moving part of boundary is then discretized with
256 cells, like in the first two testeases. The results obtained on the 256x256 grid and the
512x512 grid are identical to a satisfactory extent. For Re= 100 the finest grid is a 256 x 256 grid.
The error measure ¢ is defined for the finest grid as in (8).
Figure 14(a) shows the u! velocity profile along line CL, for Re= 100; Figure 14(b) gives the u?
velocity profile along CL. Figure 15(a) shows u! along line CL for Re= 1000, Figure 15(b)
presents u? along CL2. The velocity profiles predicted on the finest grid are given in tabulated
form in Table VI for Re=100 and in Table VII for Re= 1000, in which 14 reference points for
reproduction of the profiles are given. At(x!, x?)=(0'5, 0:5) singular point occurs. For Re= 1000
itis inspected by giving, forall grids the pressure along the line x! =0'5 and the pressure along the
line x? ==0-5 in Figure 16. A large change in pressure, found to be similar on the three finest grids,
can be observed at the singular point
Finally, Table VIII presents nu and vay for the L-shape. Figure 17 presents the La-norm of the
residual versus the number of iterations for the L-shaped cavity for Re= 100 and Re= 1000. As
expected the slopes of all lines in this figure are almost identical. This indicates the level-
independent convergence rates of the multigrid method used. The average reduction factors for
Re=100 are very good, only 15 iterations are needed to reduce the residual by 8 orders of as)
asl
a]
as
os
oa
03
i
|
baa
a et, 2563256 wi
sess, 1281 18 ri
= ress, 64x64 id a
rss, 32 Spi
03;
ou
02]
a4]
ar 0203 04 05 06 O7 08 os 1
»
7
x
Figure 14, Velocity profiles for Re=100 inthe L-shaped cavity: (au! slong CL, (0) 4# along CL, cee ene ee
rests, 512 512g
eo ‘resuli, 286 286 gd.
reels 128128 rd,
1 resus 4 64 i a
rel, 3232. gr,_ (very fe dts)
a
2a}
“Saray ea oS da 7s
»
a
Figure 15. Velocity profiles for Re~ 1000, L-shaped cavity: (2)u" along CL: (8)? along CLy oa}
|
03]
06
05}
oa
03]
02}
ou
ia ate aa
‘results, 512 x 512- grid, D)
== results, 256 x 256- grid.
results, 128 x 12%- arid (very fine dots).
a0,
0.02
0.04]
-o.0|
a
a5 04 03 02 01 0 OF 02 a3 4 os
b)
x!
Figure 16, Pressure profiles along the line x05 (a, and line x0 (b, both through the singular point 32) —
(05, 05) for Re= 1000. amber of MO- trations
¢-———
0 3 10 15 20 5 30
Oy
rests 512x512 gra,
— eres, 256 x256- ri.
results, 128 x 128 rid,
rel, 68 x 64 grid,
I) results, 32.432 grd,
Figure 17. The Z,-norm ofthe residual versus the numberof iterations for the F(L,1}
