lecture 11
TRANSCRIPT
Computational Fluid Dynamics IPPPPIIIIWWWW
Numerical Methods forElliptic Equations
Instructor: Hong G. ImUniversity of Michigan
Fall 2001
Computational Fluid Dynamics IPPPPIIIIWWWW
Direct Methods• Cramer’s rule• Gaussian elimination• Tridiagonal matrix algorithm (Thomas algorithm)• Block tridiagonal matrix algorithm
Iterative Methods• Jacobi, Gauss-Seidel iteration• Successive overrelaxation (SOR)• Multigrid method
Outline
Solution Methods for Elliptic Equations
Computational Fluid Dynamics IPPPPIIIIWWWW
SfSy
fx
f =∇=∂∂+
∂∂ 2
2
2
2
2
;
Poisson equation (Laplace equation if S = 0).
A model equation for elliptic problems
On the boundaries (BC)
),(
),(
0
0
yxgnf
yxff
=∂∂= Dirichlet
Neumann
Computational Fluid Dynamics IPPPPIIIIWWWW
ωψ −=∇ 2
Examples of Elliptic Equations
2-D stream function equation
Projection method(Step 2)
tjihjih t
P ,,2 1 u⋅∇
∆=∇
kqT&
−=∇ 2 Steady conduction equation
Computational Fluid Dynamics IPPPPIIIIWWWW
Sy
fxf =
∂∂+
∂∂
2
2
2
2
Applying central differencing
Solving the Poisson equation
for which the modified equation becomes
jijijijijijiji S
yfff
xfff
,21,,1,
2,1,,1 22
=∆
+−+
∆+− −+−+
[ ] L+∆+∆−=+ 22
121 yfxfSff yyyyxxxxyyxx
(x, y)
i−1 i i+1
j+1j
j−1
Computational Fluid Dynamics IPPPPIIIIWWWWIf ⇒=∆=∆ hyx jijijijijiji Shfffff ,
2,1,1,,1,1 4 =−+++ −+−+
=
−
−−
−
−−
JI
JI
J
JI
JI
J
SS
SSS
SS
h
ff
fff
ff
,
1,
2,2
1,2
,1
2,1
1,1
2
,
1,
2,2
1,2
,1
2,1
1,1
410
1001001
101410100141010014
M
M
M
M
M
M
M
M
LLLL
LLLL
LLLL
Sparse matrix: only 5 non-zero entries in each row
Computational Fluid Dynamics IPPPPIIIIWWWWUltimately, the difference form of the Poisson equationboils down to solving for
[ ] bf =AHence,
[ ] bf 1−= A
Direct method:- Solving inverse matrix directly (Cramer’s rule)- Inverting matrix more cleverly (Gaussian elimination)- Other (L-U decomposition, Thomas algorithm)
Computational Fluid Dynamics IPPPPIIIIWWWWCramer’s Rule
- Check elementary linear algebra book- Extremely time consuming for large matrices
Operation count:For example, 11×11 matrix:
Using 1 Gigaflops (109 flops) computer
Direct Method - 1
)!1( +n
221075.4)!199( ×=+×
!!!!!! years1020.3 100×
Computational Fluid Dynamics IPPPPIIIIWWWWGaussian Elimination
- Pivoting: rearranging equations to put the largestcoefficient on the main diagonal.
- Eliminate the column below main diagonal.- Repeat until the last equation is reached.- Back-substitution
- Special case: tri-diagonal matrix - Thomas algorithm
Direct Method - 2
nnn cfafa
cfafacfafa
=++
=++=++
LL
MM
LL
LL
2211
2222121
1212111
nnnn cfa
cfacfafa
=′
=+′=++
M
LL
LL
2222
1212111
⇒
Computational Fluid Dynamics IPPPPIIIIWWWWDiscretized Poisson Equation
Iterative Method
jijijijijiji Shfffff ,2
,1,1,,1,1 4 =−+++ −+−+
Rearranging for jif ,
[ ]jijijijijiji Shfffff ,2
1,1,,1,1, 41 −+++= −+−+
[ ]jinji
nji
nji
nji
nji Shfffff ,
21,1,,1,1
1, 4
1 −+++= −+−++ Jacobi
[ ]jinji
nji
nji
nji
nji Shfffff ,
211,1,
1,1,1
1, 4
1 −+++= +−+
+−+
+ Gauss-Seidel
SOR
jijijijijiji Shfffff ,2
,1,1,,1,1 4 =−+++ −+−+
[ ] ( ) njiji
nji
nji
nji
nji
nji fShfffff ,,
211,1,
1,1,1
1, 1
4ββ −+−+++= +
−++
−++
Computational Fluid Dynamics IPPPPIIIIWWWWA One Dimensional Example
Convergence – 1-D Problem (1)
)(xFx =
for which convergence is achieved if
An equation in the form
can be solved by iterative procedure:
)(1 nn xFx =+
nn xx ≈+1 or ε<−+
11
n
n
xx
Computational Fluid Dynamics IPPPPIIIIWWWW
The above figure suggests that
Convergence – 1-D Problem (2)
1≤dxdF for convergence
nx
1+nx
1<dxdF
1>dxdF
)(xFx =
Computational Fluid Dynamics IPPPPIIIIWWWWFor multi-dimensional problems
Convergence – Multi Dimensional (1)
fIt can be shown that the space of can be “spanned”by eigenvectors
[ ] nn M ff =+1
[ ] NMiM iii ×== L,1,qq λsuch that
NMNMyyy ××+++= qqqf L2211
Hence, [ ] [ ] ( )NMNMyyyMM ××+++= qqqf L2211
[ ] [ ] L++= 2211 qq MyMy
L++= 222111 qq λλ yy
Computational Fluid Dynamics IPPPPIIIIWWWWTherefore, the problem
Convergence – Multi Dimensional (2)
is equivalent to
[ ] nn M ff =+1
or
LL ++=++ ++2221112
121
11 qqqq λλ nnnn yyyy
M
nn
nn
yyyy
221
2
111
1
λλ
==
+
+
yielding a convergence condition
1max ≤λ
Computational Fluid Dynamics IPPPPIIIIWWWWGeneral iterative procedure
Convergence – Multi Dimensional (3)
[ ] [ ] [ ]21 AAA −=Let
[ ] bf =A
[ ] [ ] bff += 21 AA
An iterative scheme is constructed as
[ ] [ ] bff +=+ nn AA 21
1
For example,
[ ] [ ] [ ] [ ] [ ] [ ] [ ]DABAIDA −==−== 121 ,41 Jacobi
[ ] [ ] [ ] [ ] [ ]UALDA =−= 21 , Gauss-Seidel
Computational Fluid Dynamics IPPPPIIIIWWWWConvergence – Multi Dimensional (4)
Requirements:
1. should be invertible.2. Iteration should converge, i.e.
[ ] [ ] bff +=+ nn AA 21
1
Define error at n-th iteration
ff =∞→
n
nlim
nn ffε −=
[ ]1A
[ ] [ ] nn AA εε 21
1 =+ [ ] [ ] nn AA εε 21
11 −+ =
[ ] [ ]( ) 02
11 εε
nn AA −=
Computational Fluid Dynamics IPPPPIIIIWWWWConvergence – Multi Dimensional (5)
Condition for convergence
which requires
This is achieved if the modulus of the eigenvalues are less than unity.
Therefore, the convergence condition becomes:
0lim =∞→
n
nε
1max
≤= iλρ
[ ] [ ]( ) 0lim 21
1 =−
∞→
n
nAA
iλρ Spectral radius of convergence
Eigenvalues of matrix [ ] [ ]211 AA −
Computational Fluid Dynamics IPPPPIIIIWWWWConvergence – Jacobi Method (1)
Jacobi Iteration Method for Poisson Equation(2nd order central difference with uniform mesh)
1,,1,1,,1,coscos21 −=−=
+= NnMm
Nn
Mm
mn LLππλ
and using a discrete analog of separation of variables, it can be shown that the eigenvalues of are
Therefore, and the Jacobi method converges.
[ ] [ ]IA =−11
[ ]jinji
nji
nji
nji
nji Shfffff ,
21,1,,1,1
1, 4
1 −+++= −+−++
[ ]2A
1≤mnλ
Computational Fluid Dynamics IPPPPIIIIWWWWConvergence – Jacobi Method (2)
For a large matrix
L+
+−≅
+=
2
2
2
2
max,
411
coscos21
NM
NMmn
ππ
ππλ largest eigenvalues for
Thus, for a large matrix, is only slightly lessthan unity.
⇒ Very slow convergence
1,1 == nm
max,mnλ
Computational Fluid Dynamics IPPPPIIIIWWWWConvergence – Gauss-Seidel (1)
Gauss-Seidel Iteration Method for Poisson Equation(2nd order central difference with uniform mesh)
[ ]jinji
nji
nji
nji
nji Shfffff ,
211,1,
1,1,1
1, 4
1 −+++= +−+
+−+
+
[ ] [ ] [ ] [ ] [ ]UALDA =−= 21 ,
−
=
−
−−
−
+
+
+
NMn
NM
n
n
nJI
n
n
S
SS
h
f
ff
f
ff
,
2,1
1,1
2
,
2,1
1,1
1,
12,1
11,1
000000
010000010000010
4100
04100004100004100004
M
M
M
M
M
M
L
MMM
MMMMM
L
L
L
M
M
M
LL
MLMMM
L
L
L
L
A1 A2
Computational Fluid Dynamics IPPPPIIIIWWWWConvergence – Gauss-Seidel (2)
It can be shown that the eigenvalues of matrixAre simply square of the eigenvalues of Jacobi method
[ ] [ ]211 AA −
2
max, coscos41
+=
NMmnππλ
Thus, Gauss-Seidel method is twice as fast as the Jacobi method.
Computational Fluid Dynamics IPPPPIIIIWWWWSuccessive Overrelaxation - 1
Successive Overrelaxation
[ ] [ ]( ) [ ] bff +=− + nn ULD 1
Consider the Gauss-Seidel method
If Gauss-Seidel is an attempt to change the solution as
dff +=+ nn 1
Can we accelerate the change by a parameter?
1,1 >+=+ ββdff nn
Computational Fluid Dynamics IPPPPIIIIWWWWSuccessive Overrelaxation - 2
Hence, SOR first uses Gauss-Seidel to compute intermediate solution,
[ ] [ ]( ) [ ] bff +=− nULD ~or
( )nnn ffff −+=+ ~1 β
Then accelerate the next iteration solution
( ) nff ββ −+= 1~
f~
[ ] [ ] [ ] bfff ++= nULD ~~
[ ] [ ]( ) ( ) nnn ULc
fbfff ββ −+++=+ 1~1
[ ] [ ]IcD =
Combining the two steps
Note that [ ] [ ] 1~ += nLL ff
Computational Fluid Dynamics IPPPPIIIIWWWWSuccessive Overrelaxation - 3
Example: Poisson Equation (2nd order CS, uniform mesh)
[ ] [ ] [ ] bfff ++= nULD ~~
[ ] [ ]( ) ( ) nnn ULc
fbfff ββ −+++=+ 1~1
jijijijijiji Shfffff ,2
,1,1,,1,1 4 =−+++ −+−+
[ ] ( ) njiji
nji
nji
nji
nji
nji fShfffff ,,
21,,1
11,
1,1
1, 1
4ββ −+−+++= ++
+−
+−
+
[ ]jinji
nji
nji
nji
nji Shfffff ,
21,,1
11,
1,1
1, 4
1 −+++= +++−
+−
+
G-S
Combining
Computational Fluid Dynamics IPPPPIIIIWWWWSuccessive Overrelaxation - 4
Convergence of SOR
[ ] [ ] [ ] bfff ++= nULD ~~
[ ] [ ] [ ]( ) ( )[ ] [ ] [ ]{ }[ ] [ ] [ ]( ) [ ] b
ff111
1111 1−−−
−−−+
−+
+−−=
DLDI
UDILDI nn
ββ
βββ
From
Eliminating and solving for
( )nnn ffff −+=+ ~1 β1+nff~
[ ]SORM
Convergence depends on the eigenvalues of [ ]SORM
Computational Fluid Dynamics IPPPPIIIIWWWWSuccessive Overrelaxation - 5
For the discretized Poisson operator, it can be shown that
where is an eigenvalue of the Jacobi matrix
Note that if (Gauss-Seidel)
( )[ ]1421 222/1 −−+= ββλλβµ
λ[ ] [ ] [ ] [ ]( )ULDM J += −1
2λµ = 1=β
Minimum occurs atµ
2max
opt112λ
β−+
=Typically
9.1~7.1opt ≈β1max ≈λ
Computational Fluid Dynamics IPPPPIIIIWWWWSuccessive Overrelaxation - 6
For problems with irregular geometry and non-uniform mesh, must be found by trial and error.
Typical Comparison Chart
optβ
290.7906SOR
6250.9890Gauss-Seidel
12500.9945Jacobi
Iterations( )maxmax µλ
Ferziger, J. H., Numerical Method for Engineering Application (1981)
Computational Fluid Dynamics IPPPPIIIIWWWW
[ 1,,11,,1, 4 ++−− +++= jijijijiji fffff β
] ( ) jiji fSh ,,2 1 β−+−
Coloring Scheme (Red & Black)
In large computer application (vector or parallel platform),SOR faces difficulties in using constantly updated values. Remedy: Two separate grid system (red & black)
[ 1,,11,,1, 4 ++−− +++= jijijijiji fffff βdo i =1, nx, 2
do i=2, nx, 2
] ( ) jiji fSh ,,2 1 β−+−
enddo
enddo
Computational Fluid Dynamics IPPPPIIIIWWWWSuccessive Line Overrelaxation (SLOR) - 1
Line Relaxation Method (Line Gauss-Seidel Method)
Adding one more coupling (E)
⇒ Thomas algorithm
[ ]jinji
nji
nji
nji
nji Shfffff ,
21,,1
11,
1,1
1, 4
1 −+++= +++−
+−
+
[ ]jinji
nji
nji
nji
nji Shfffff ,
21,
1,1
11,
1,1
1, 4
1 −+++= ++
++−
+−
+
[ ]jinji
nji
nji
nji
nji Shfffff ,
21,
11,
1,1
1,
1,1 4
141
41 −+=−+− +
+−
++
++−
New
New New
Old
Old
New
New New
Old
New
Computational Fluid Dynamics IPPPPIIIIWWWWSuccessive Line Overrelaxation (SLOR) - 2
SLOR = Line Relaxation + Overrelaxation
Apply line relaxation for intermediate solution
and then overrelax
which is no more complicated than line relaxation.
( ) njiji
nji fff ,,
1, 1~ ββ −+=+
[ ]jinji
njijijiji Shfffff ,
21,
11,,1,,1 4
1~41~~
41 −+=−+− +
+−+−
Computational Fluid Dynamics IPPPPIIIIWWWWSuccessive Line Overrelaxation (SLOR) - 3
Notes on SLOR
- Exact eigenvalues are unknown.- To ensure convergence, - Converges approximately twice as fast as Gauss-Seidel.- May be faster than pointwise SOR, but each iterationtakes longer with Thomas algorithm.
- Improved convergence is due to the direct effect of theboundary condition in each row.
2≤β
Computational Fluid Dynamics IPPPPIIIIWWWWAlternating-Direction Implicit - 1
ADI for elliptic equation is analogous to ADI in parabolicequation
Sy
fx
ftf −
∂∂+
∂∂=
∂∂
2
2
2
2
α
[ ]Sfftff yyxxnji
nji −+∆=−+ δδα,
1,
In discrete form
and take it to the limit to obtain the steady solution.
=∂∂ 0
tf
( ) ( ) ( ) ( )[ ]jijijixx x ,1,,12 21+− +−
∆=δ
( ) ( ) ( ) ( )[ ]1,,1,2 21+− +−
∆= jijijiyy y
δ
Computational Fluid Dynamics IPPPPIIIIWWWWAlternating-Direction Implicit - 2
ADI for [ ] Sfftff yyxxnji
nji −+∆=−+ δδα,
1,
is written as
( ) ( )[ ]jinji
nji
nji
nji
nji
nji
nn Sffffffh
tff ,1,,1,2/1
,12/1
,2/1
,122/1 22
2−+−++−∆=− −+
+−
+++
+ α
( ) ( )[ ]jinji
nji
nji
nji
nji
nji
nn Sffffffh
tff ,11,
1,
11,
2/1,1
2/1,
2/1,12
2/11 222
−+−++−∆=− +−
+++
+−
+++
++ α
or( ) ( ) ji
nyyn
nxxn S
htff ,2
2/1
211 ∆−+=− + αδρδρ
( ) ( ) jin
xxnn
yyn Sh
tff ,22/11
211 ∆−+=− ++ αδρδρ
∆=
2n
ntαρ
Computational Fluid Dynamics IPPPPIIIIWWWWAlternating-Direction Implicit - 3
Convergence of ADI
- Iteration parameter usually varies with iteration- For example, Wachspress
lower bound eigenvalue
2n
nt∆= αρ
( ) ( )nk
bab
ht
h
nkkk ,,1,
2
1/1
22 L=
=∆=
−−αρ
ab upper bound eigenvalue
- Comparison with SOR is difficult- ADI can be efficient if appropriate parameters are found.
Computational Fluid Dynamics IPPPPIIIIWWWW
1,if=
Boundary Conditions for Iterative Method
Dirichlet conditions are easily implemented.For Neumann condition, the simplest approach is
00 1,0, =−⇒=∂∂
ii ffnf
(1st order)
Update interior points and then setL,,, 3,2,1, iii fff 1,0, ii ff =
⇒ This generally does not converge.
Instead, incorporate BC directly into the equations
[ ]jiiiiii Shfffff ,2
0,2,1,11,11, 41 −+++= +−
[ ]jiiiii Shffff ,2
2,1,11,11, 31 −++= +−
Computational Fluid Dynamics IPPPPIIIIWWWWMultigrid Method - 1
Basic Idea
Suppose we want to solve a 1-D elliptic equation
02
2
=−∂∂ g
xf
An iterative process is analogous to solving an unsteadyproblem (recall ADI)
−
∂∂=
∂∂ g
xf
tf
2
2
α
And take it to the limit ∞→t
Computational Fluid Dynamics IPPPPIIIIWWWWMultigrid Method - 2
Analytic Solution Behavior
The equation can be solved by Fourier series
∑=k
ikxk etaf )(
and assume that can be expanded as
−
∂∂=
∂∂ g
xf
tf
2
2
α
Substituting into the equation,
g
∑=k
ikxk ekbg 2
( ) ikx
kkk
ikx
k
k ekbtaedt
da 2)(∑∑ −−= α
Computational Fluid Dynamics IPPPPIIIIWWWWMultigrid Method - 3
Solving for each
Therefore,
Rate of convergence
kk bta =∞→ )(
( )kkk btak
dtda −−= )(2α
k
( ) tkkkkk ebabta
2
)0()( α−−=−
= 2
2 1~k
k c ατα
High wave number modes damps out faster.
( )tkk e
2
0αεε −=
Computational Fluid Dynamics IPPPPIIIIWWWWMultigrid Method - 4
For iterative method to solve elliptic equations,there are high wave number and low wave number components of errors that need to be damped out.
Consider a domain discretized by points1=L N
1 N
kL=2π
kH=2πN
Computational Fluid Dynamics IPPPPIIIIWWWWMultigrid Method - 5
For explicit time step, stability condition yields
If the error at various wave number decays at
where
αα
2;
21 2
2
hth
t =∆=∆
tke2
0αεε −=
α2
2nhtnt =∆=
( )2/exp/ 220 nhk−=εε⇒
( ) NkhNn HH ππεε 2forexp 222
0
=−=
( ) ππεε 2forexp 22
0
=−= LL khn
Short-wave errorsdecay much faster
Computational Fluid Dynamics IPPPPIIIIWWWWMultigrid Method - 6
1 Nt = 0
t = 5∆t
ε
1 N
ε
1 N
ε
t = 500∆t
Computational Fluid Dynamics IPPPPIIIIWWWWMultigrid Method - 7
Multigrid Method – the Idea
- A low wave number component on a fine grid becomesa high wave number component on a coarse grid
- Use a coarse grid system to converge low wave numbercomponent of the solution rapidly
- Map it onto the fine grid system to converge high wavenumber component.
( ) ( )222220 /expexp/ NLnhnL ππεε −=−=
Computational Fluid Dynamics IPPPPIIIIWWWWMultigrid Method - 8
Suppose we are solving a Laplace equation:
jijijijijijiji
ji Ry
fffx
fffLf ,2
1,,1,2
,1,,1,
22=
∆+−
+∆
+−= −+−+
and iterate until residual becomes smaller than tolerance
ε<→ jiji RR ,, or0
Define ji
njiji fff ,,, ∆+=
ji
nji
ji
ff
f
,
,
,
∆
Converged solutionSolution after n-th iterationCorrection
Computational Fluid Dynamics IPPPPIIIIWWWWMultigrid Method - 9
Since
we have
0,,, =∆+= jinjiji fLLfLf
and ji
nji RLf ,, =
jiji RfL ,, −=∆ Coarse grid correction
Steps: Fine grid solution ⇒ Interpolation onto coarse grid (restriction)⇒ Coarse grid correction⇒ Interpolated onto fine grid (prolongation)⇒ Converge on fine grid
Computational Fluid Dynamics IPPPPIIIIWWWWMultigrid Method - 10
Two-Level Multigrid Method - (nx+1, ny+1); (nx/2+1, ny/2+1)
1. Do n iterations on the fine grid (n = 3~4) using G-S
2. If then stop.3. Interpolate the residual onto the coarse grid
(restriction (injection if nx/2))4. Iterations on the coarse grid
5. Interpolate the correction onto the fine grid(prolongation)
6. Go to 1.
jinji RLf ,, =
jiR ,
jiji RIfL ,2
1,2 )( −=∆jif ,2 )(∆
ε<jiR ,
jinjiji fIff ,2
12,, )(∆+=
jiRI ,2
1
jifI ,212 )(∆
Computational Fluid Dynamics IPPPPIIIIWWWWMultigrid Method - 11
Details on ProlongationCoarse GridFine grid
jif ,∆ jif ,2+∆
2, −∆ jif
jif ,2−∆
2, +∆ jif ( )jijiji fff ,2,,1 21
++ ∆+∆=∆
( )2,,1, 21
++ ∆+∆=∆ jijiji fff
( )
( )2,22,,2,
1,22,11,,1
1,1
4141
++++
++++++
++
∆+∆+∆+∆=
∆+∆+∆+∆=
∆
jijijiji
jijijiji
ji
ffff
ffff
f
Computational Fluid Dynamics IPPPPIIIIWWWWMultigrid Method - 12
The process can be extended to multi-level grids(nx+1,ny+1), (nx/2+1,ny/2+1), (nx/4+1,ny/4+1), …, (3,3)
Various Strategies
V-Cycle W-Cycle
exact solution
Fine
Coarse
Computational Fluid Dynamics IPPPPIIIIWWWWMultigrid Method - 13
Example: 4-Level V-Cycle
1. Iterate on Grid 1 for n times2. Restriction by injection to Grid 23. Iterate on Grid 2
(Save )4. Restriction by injection to Grid 35. Iterate on Grid 3
(Save )6. Restriction by injection to Grid 47. Iterate on Grid 4
1,, 0 ji
nji RLf ⇒=
1,
21 jiRI
1,
21,2 )( jiji RIfL −=∆
1,
21,2 ,)( jiji RIf∆n
jijiji fLRIR ,21,
21
2, )(∆+=
2,
32 jiRI
2,
32,3 )( jiji RIfL −=∆
2,
32,3 ,)( jiji RIf∆
Fine ⇒⇒⇒⇒ Coarse
njijiji fLRIR ,3
2,
32
3, )(∆+=
3,
43,4 )( jiji RIfL −=∆
3,
43 jiRI
Computational Fluid Dynamics IPPPPIIIIWWWWMultigrid Method - 14
1. Prolongate from Grid 4 to Grid 3
2. Iterate on Grid 3 (saved)3. Prolongate from Grid 3 to Grid 2
4. Iterate on Grid 2 (saved)5. Prolongate from Grid 2 to Grid 1
6. Iterate on Grid 17. If then stop. Else, repeat the entire cycle.
njifI ,4
34 )(∆
2,
32,3 )( jiji RIfL −=∆
Coarse ⇒⇒⇒⇒ Fine
nji
nji
nji fIff ,4
34,3,3 )()()( ∆+∆=∆
njifI ,3
23 )(∆n
jin
jin
ji fIff ,323,2,2 )()()( ∆+∆=∆
1,
21,2 )( jiji RIfL −=∆
njifI ,2
12 )(∆
nji
nji
nji fIff ,2
12,, )(∆+=
0, =jiLfε<jiR ,
Computational Fluid Dynamics IPPPPIIIIWWWWMultigrid Method - 15
Test Problem (Tannehill, p. 170)- Laplace equation in a square domain - Dirichlet conditions on four boundaries- 5 Levels of resolution
9×9, 17×17, 33×33, 65×65, 129×129- 4 Methods
1. Conventional Gauss-Seidel (GS)2. Gauss-Seidel-SOR with optimal ω (GSopt )3. Multigrid with 2-level grids (MG2)4. Multigrid with maximum levels of grid (MGMAX)
Computational Fluid Dynamics IPPPPIIIIWWWWMultigrid Method - 16
217322826826129×12920258137228265×6520957571533×3319404021517×17171919629×9
MGMAXMG2GSωoptGSGrid size
Computational Fluid Dynamics IPPPPIIIIWWWWMultigrid Method - 17
The multigrid method is the fastest techniquecurrently available for general elliptic equations, and are said to compare favorably with the fastest direct solvers for special problems.
MUDPACKhttp://www.scd.ucar.edu/css/software/mudpack/FISHPACKhttp://www.scd.ucar.edu/css/software/fishpack/