70
The Application of Hypersingular Meshless Method for 3D Potential and Exterior Acoustics Problems Reporter Professor D. L. You ng 2008/01/03 Department of Civil Engineering and Hydrotech Research Institute National Taiwan University Scientific Computing & Visualization Lab

Reporter : Professor D. L. Young 2008/01/03

  • Upload
    rufin

  • View
    32

  • Download
    0

Embed Size (px)

DESCRIPTION

Department of Civil Engineering and Hydrotech Research Institute National Taiwan University. The Application of H ypersingular M eshless M ethod for 3D Potential and Exterior Acoustics Problems. Reporter : Professor D. L. Young 2008/01/03. - PowerPoint PPT Presentation

Citation preview

Page 1: Reporter : Professor D. L. Young                 2008/01/03

1

The Application of Hypersingular Meshless Method

for 3D Potential and Exterior Acoustics Problems

Reporter : Professor D. L. Young 2008/01/03

Department of Civil Engineering and Hydrotech Research Institute National Taiwan University

Scientific Computing & Visualization Lab

Page 2: Reporter : Professor D. L. Young                 2008/01/03

2

含超強奇異性無網格法於三維勢能及外域聲學問題之應用

楊德良 教授 2008/01/03

Department of Civil Engineering and Hydrotech Research Institute National Taiwan University

Scientific Computing & Visualization Lab

Page 3: Reporter : Professor D. L. Young                 2008/01/03

3

NNational ational TTaiwan aiwan UUniversityniversity

Outline: Introduction Potential problems

Formulation The diagonal coefficient of influence matrices Numerical results

Cube Cylinder Arbitrary shape

Exterior acoustics problems Formulation The diagonal coefficient of influence matrices Numerical results

Scattering by a soft sphere Scattering by a rigid sphere Scattering by a bean shape obstacle

Conclusions Further researches

Page 4: Reporter : Professor D. L. Young                 2008/01/03

4

NNational ational TTaiwan aiwan UUniversityniversity

Brief detail of MFS

FDM FEM BEM

M esh method

M Q M FS

M eshless method

Numerical method

Page 5: Reporter : Professor D. L. Young                 2008/01/03

5

NNational ational TTaiwan aiwan UUniversityniversity

Brief detail of MFS

Method of fundamental solutions ( MFS ) is involved through the combination of meshless and the concept of indirect boundary element method.

The MFS considers an artificial boundary outside the computational domain, to locate the source points and some field points locate on the boundary. Using these points and boundary conditions can solve the coefficients used in the fundamental solution.

Page 6: Reporter : Professor D. L. Young                 2008/01/03

6

NNational ational TTaiwan aiwan UUniversityniversity

Brief detail of MFS

V1

V2

0 0.5 10

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

Frame 001 29 Jul 2002 Frame 001 29 Jul 2002

X

Y

0 0.5 10

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

Frame 001 27 Oct 2003 MESH TEST

-40.00 0.00 40.00 80.00 120.00 160.00

-40.00

0.00

40.00

80.00

120.00

160.00Field point

Source point

Domain method

MFS

Page 7: Reporter : Professor D. L. Young                 2008/01/03

7

NNational ational TTaiwan aiwan UUniversityniversity

Brief detail of MFS From the principles of method of fundamental solutio

ns, for the given governing equation, the free space Green’s function has to be satisfied.

For example of the Laplace equation as follows the free space Green’s function can be written

where is the fundamental solutions is the Dirac delta function, is the position of t

he field point, and is the position of the source point.

2 ( ) ( )G x x

( ) ln ijG x r

)(

x

x

Page 8: Reporter : Professor D. L. Young                 2008/01/03

8

NNational ational TTaiwan aiwan UUniversityniversity

Brief detail of MFS

Method of Fundamental Solutions (MFS)

0 L

N

D

'G x x - L

1'N

j jjG x x

A B

Page 9: Reporter : Professor D. L. Young                 2008/01/03

9

NNational ational TTaiwan aiwan UUniversityniversity

Brief detail of MFS Using the above expression, the approximate solution

can be obtained as

1

,N

ji i ijj

x y G r

And the field points located on the boundary and com

bined with boundary condition that can solve coefficients and advance to solve any region in the solution domain.

Page 10: Reporter : Professor D. L. Young                 2008/01/03

10

NNational ational TTaiwan aiwan UUniversityniversity

Singular Value Decomposition SVD is the technique for dealing with sets of

equations or matrices are either singular or else numerically very close to singular.

1

2

n

TA U V

Orthogonal matrix

Matrix of the singular values

Page 11: Reporter : Professor D. L. Young                 2008/01/03

11

NNational ational TTaiwan aiwan UUniversityniversity

Introduction"

B fB.C.

G.E.

G x s L

1

,n

j jj

x G x s

L 0Time-independent

Page 12: Reporter : Professor D. L. Young                 2008/01/03

12

NNational ational TTaiwan aiwan UUniversityniversity

Introduction

(n+1)dt

y

t

(n)dt

(n-)dt

Field pointSource pointTime-dependent

Page 13: Reporter : Professor D. L. Young                 2008/01/03

13

NNational ational TTaiwan aiwan UUniversityniversity

Numerical methodsfor Burgers’ eq.Mesh

method

Meshless method

FDM

FEM

MQ

Mesh-reduction method

BEM

MFS-DRM

MLPG

1980 Varoglu & Finn 1981 Caldwell & Wanless 1982 Nguyen & Reynen 2004 Dogan

1984 Evans & Abdullah

1990 Kakuda & Tosaka1998 Hon

2002 Li, Hon & Chen

Introduction (Burgers’ equation)

2000 Lin & Atluri

Modified Helmhotz fundamental solution

Domain-type method

Page 14: Reporter : Professor D. L. Young                 2008/01/03

14

NNational ational TTaiwan aiwan UUniversityniversity

Introduction (1/4)

The Method of Fundamental Solutions proposed by Kupradze and Aleksidze, 1964.

The MFS has been generally applied to solve some engineering problems. It is a kind of meshless methods, since only boundary nodes are distributed.

However because of the controversial artificial boundary (off-set boundary) outside the physical domain, the MFS has not become a popular numerical method.

MFS only works well in regular geometry with the Dirichlet and Neumann boundary conditions.

Page 15: Reporter : Professor D. L. Young                 2008/01/03

15

NNational ational TTaiwan aiwan UUniversityniversity

Introduction (2/4)

This research extends the Hypersingular Meshless Method to solve the 3D potential and exterior acoustics problems.

Young et al. 2005 J. Comput. Phys. Potential problems in 2D.

Chen et al. 2006 Eng. Anal. Bound. Elem.

Multiply-connected-domain Laplace problem in 2D.

Young et al. 2006 J. Acoust. Soc. Am. Exterior acoustics problems in 2D.

Page 16: Reporter : Professor D. L. Young                 2008/01/03

16

NNational ational TTaiwan aiwan UUniversityniversity

Introduction (3/4)Source point location

MFS HMM

Page 17: Reporter : Professor D. L. Young                 2008/01/03

17

NNational ational TTaiwan aiwan UUniversityniversityIntroduction (4/4) Comparison of HMM and MFS

HMM MFSMeshless features Yes Yes

Source point location Real boundary Fiction boundaryAccuracy

Acceptable Better

PotentialDouble layer Single layer

Kernel functions for 3D potential

problems

Kernel functions for 3D exterior acoustics

3,ij

kkiji

rnyxsB 5

2 3,

ij

lklkijkkiji

rnnyyrnn

xsB

ij

iji

rxsA 1, 3,

ij

kkiji

rnyxsA

3

1,

ij

kkikr

ijije

rnyeikr

xsAij

5

22 131,

ij

kkllikr

ijijklikr

ijkkllikr

ije

rnynyeikrrnneikrnynyek

xsBijijij

ij

ikrije

rexsA

ij

,

3

1,

ij

kkikr

ijije

rnyeikr

xsBij

Page 18: Reporter : Professor D. L. Young                 2008/01/03

1818

Potential ProblemsPotential Problems

Department of Civil Engineering and Hydrotech Research Institute Department of Civil Engineering and Hydrotech Research Institute National Taiwan UniversityNational Taiwan University

Page 19: Reporter : Professor D. L. Young                 2008/01/03

19

NNational ational TTaiwan aiwan UUniversityniversity

Formulation Governing equation: ,

the representation of the solution for interior problem can be approximated as:

Kernel functions:

023 x Dx

( )

1

( ) ( , )N

i i j i j

j

x A s x

( )

1

( ) ( , ) ,N

i i j i j

j

x B s x

3,ij

kkiji

rnyxsA

5

2 3,

ij

lklkijkkiji

rnnyyrnn

xsB

Page 20: Reporter : Professor D. L. Young                 2008/01/03

20

NNational ational TTaiwan aiwan UUniversityniversity

The Diagonal Coefficient of Influence Matrices

1, 1,1 1,2 1,1

2,1 2, 2,2 2,1

,1 ,2 , ,1

,

N

m Nm

N

m Ni jm

N

N N N m N Nm

a a a a

a a a a

a a a a

1, 1,1 1,2 1,1

2,1 2, 2,2 2,1

,1 ,2 , ,1

( )

( ).

( )

N

m Nm

N

m Ni jm

N

N N N m N Nm

b b b b

b b b b

b b b b

Page 21: Reporter : Professor D. L. Young                 2008/01/03

21

NNational ational TTaiwan aiwan UUniversityniversity

Analytical derivation of diagonal coefficients Analytical solution:

:kwhere

:Nwave number

number of nodes

radius of sphere:

}))!(()!(4

)!22()!2()()()124(

)!(4))!2(()()()14(2{

2

22)12()1(

)12(

1

1 0

')2(

042

2)1(

)2('

)2(

2

lmmlmmkhkjlm

mmkhkjm

Nika

mm

N

l mm

mmmmii

}))!(()!(4

)!22()!2()()()124(

)!(4))!2(()()()14(2{

2

22)12()1(

)12(

1

1 0

')12(

042

2)1(

)2('

)2(

3

lmmlmmkhkjlm

mmkhkjm

Nikb

mm

N

l mm

mmmmii

Page 22: Reporter : Professor D. L. Young                 2008/01/03

22

NNational ational TTaiwan aiwan UUniversityniversity

Case 1-1:Dirichlet boundary case:

1

0 0

22

22

1 1 sinh

sinh)1(1)1(1)sin()sin(4nm

znmnm

ynxmnm

m n

Page 23: Reporter : Professor D. L. Young                 2008/01/03

23

NNational ational TTaiwan aiwan UUniversityniversity

Point Distribution with Normal Vectors

Page 24: Reporter : Professor D. L. Young                 2008/01/03

24

NNational ational TTaiwan aiwan UUniversityniversityResultsCross-section at x=0.5

MFS HMM FEM RMSE: 6.26E-5 RMSE: 2.01E-3 RMSE: 4.10E-3 1350 NODES 1350 NODES 3375 NODES (13720 Elements)

0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 10

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 10

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 10

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

( analytical solution, numerical result)

Page 25: Reporter : Professor D. L. Young                 2008/01/03

25

NNational ational TTaiwan aiwan UUniversityniversity

Comparison of Three MethodsNumerical values at x=0.5, z=0.5

0 0.2 0.4 0.6 0.8 1

0

0.04

0.08

0.12

0.16

0.2

0 0.2 0.4 0.6 0.8 1

0

0.04

0.08

0.12

0.16

0.2

0 0.2 0.4 0.6 0.8 1

0

0.04

0.08

0.12

0.16

0.2

0 0.2 0.4 0.6 0.8 1Y

0

0.04

0.08

0.12

0.16

0.2

Ph a i

Exact solu tion

HM M (1350 nodes)

M FS (1350 nodes)

FE M (1000 nodes)

00001

0.040190.040150.039430.040150.92857

0.077330.077340.077510.077340.85714

0.10910.109230.109420.109230.78571

0.134140.134440.134640.134440.71429

0.151970.152420.152620.152420.64286

0.162570.163120.163320.163120.57143

0.166080.166670.166860.166670.5

0.162570.163120.163320.163120.42857

0.151970.152420.152620.152420.35714

0.134140.134440.134640.134440.28571

0.10910.109230.109420.109230.21429

0.077330.077340.077510.077340.14286

0.040190.040150.039430.040150.07143

00000

FEMMFSHMMAnalytical solutionY

00001

0.040190.040150.039430.040150.92857

0.077330.077340.077510.077340.85714

0.10910.109230.109420.109230.78571

0.134140.134440.134640.134440.71429

0.151970.152420.152620.152420.64286

0.162570.163120.163320.163120.57143

0.166080.166670.166860.166670.5

0.162570.163120.163320.163120.42857

0.151970.152420.152620.152420.35714

0.134140.134440.134640.134440.28571

0.10910.109230.109420.109230.21429

0.077330.077340.077510.077340.14286

0.040190.040150.039430.040150.07143

00000

FEMMFSHMMAnalytical solutionY

Page 26: Reporter : Professor D. L. Young                 2008/01/03

26

NNational ational TTaiwan aiwan UUniversityniversity

Absolute Error Distribution Map at x=0.5MFS

Page 27: Reporter : Professor D. L. Young                 2008/01/03

27

NNational ational TTaiwan aiwan UUniversityniversity

Absolute Error Distribution Map at x=0.5HMM

Page 28: Reporter : Professor D. L. Young                 2008/01/03

28

NNational ational TTaiwan aiwan UUniversityniversity

Case 1-2:Dirichlet and Neumann mixed boundary case:

1

0 0

22

22

1 1 cosh

cosh)1(1)1(1)sin()sin(4nm

znmnm

ynxmnm

m n

Page 29: Reporter : Professor D. L. Young                 2008/01/03

29

NNational ational TTaiwan aiwan UUniversityniversityResultsCross-section at x=0.5

MFS HMM FEM RMSE: 8.23E-5 RMSE: 2.02E-3 RMSE: 4.10E-3 1350 NODES 1350 NODES 3375 NODES (13720 Elements)

0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 10

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 10

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 10

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

( analytical solution, numerical result)

Page 30: Reporter : Professor D. L. Young                 2008/01/03

30

NNational ational TTaiwan aiwan UUniversityniversity

Case 2:

Analytical solution:

: Bessel function : The root of

10

1

)(12n

nzz

nn

rJeeJee

nn

nn

n 0Jthn0J

Page 31: Reporter : Professor D. L. Young                 2008/01/03

31

NNational ational TTaiwan aiwan UUniversityniversity

ResultsCross-section at z=0.5

MFS HMM 1800 nodes 1800 nodes RMSE: 3.80E-3 RMSE: 2.41E-2

-0.8 -0 .6 -0.4 -0.2 0 0.2 0.4 0.6 0.8

-0.8

-0.6

-0.4

-0.2

0

0.2

0.4

0.6

0.8

-0 .8 -0 .6 -0 .4 -0.2 0 0.2 0.4 0.6 0.8

-0.8

-0.6

-0.4

-0.2

0

0.2

0.4

0.6

0.8

( analytical solution, numerical result)

Page 32: Reporter : Professor D. L. Young                 2008/01/03

32

NNational ational TTaiwan aiwan UUniversityniversity

Point Distribution Comparison

RMSE : 2.41E-2 RMSE : 0.1523

RMSE : 8.89E-2 RMSE : 0.1738

-1 .2 -0.8 -0.4 0 0.4 0.8 1.2

-1.2

-0.8

-0.4

0

0.4

0.8

1.2

-1 .2 -0 .8 -0.4 0 0.4 0.8 1.2

-1 .2

-0 .8

-0 .4

0

0 .4

0 .8

1 .2

-1 .2 -0.8 -0.4 0 0.4 0.8 1.2

-1.2

-0.8

-0.4

0

0.4

0.8

1.2

-1.2 -0.8 -0 .4 0 0.4 0.8 1.2

-1 .2

-0 .8

-0 .4

0

0.4

0.8

1.2

Page 33: Reporter : Professor D. L. Young                 2008/01/03

33

NNational ational TTaiwan aiwan UUniversityniversity

Sensitivity Test of Point Distribution

The distance of the nodes on the top surface is fixed at 0.0833

Number of nodes on Z axis

Distance of the nodes on Z axis

RMSE

6 0.1666 0.37647 0.1428 0.28718 0.1250 0.21319 0.1111 0.151010 0.1000 9.84E-211 0.0909 5.44E-212 0.0833 2.41E-213 0.0769 3.39E-214 0.0714 6.10E-215 0.0666 8.73E-2

Page 34: Reporter : Professor D. L. Young                 2008/01/03

34

NNational ational TTaiwan aiwan UUniversityniversity

Sensitive Test of Number of nodesNodes RMSE

234 6.870E-2420 3.070E-2684 2.726E-2

1050 2.481E-21518 2.507E-21800 2.412E-22106 2.319E-22842 2.299E-23720 2.253E-24240 2.172E-2

0 1000 2000 3000 4000 5000N u m be r o f no de s

2.00E -002

3.00E -002

4.00E -002

5.00E -002

6.00E -002

7.00E -002

R M SE

Page 35: Reporter : Professor D. L. Young                 2008/01/03

35

NNational ational TTaiwan aiwan UUniversityniversity

Case 3-1:

Inside radius: 1Outside radius: 2

Height: 11

0

0

0

Page 36: Reporter : Professor D. L. Young                 2008/01/03

36

NNational ational TTaiwan aiwan UUniversityniversity

Point Distribution with Normal Vectors

1991 nodes

Page 37: Reporter : Professor D. L. Young                 2008/01/03

37

NNational ational TTaiwan aiwan UUniversityniversity

ResultsCross-section at z=0

MFS HMM FEM 1991 nodes 1991 nodes 1320 nodes (5000 elements)

- 2 - 1 . 5 - 1 - 0 . 5 0 0 . 5 1 1 . 5 2- 2

- 1 . 5

- 1

- 0 . 5

0

0 . 5

1

1 . 5

2

- 2 - 1 . 5 - 1 - 0 . 5 0 0 . 5 1 1 . 5 2- 2

- 1 . 5

- 1

- 0 . 5

0

0 . 5

1

1 . 5

2

- 2 - 1 . 5 - 1 - 0 . 5 0 0 . 5 1 1 . 5 2- 2

- 1 . 5

- 1

- 0 . 5

0

0 . 5

1

1 . 5

2

Page 38: Reporter : Professor D. L. Young                 2008/01/03

38

Case 3-2:

BC:Analytical solution:

zyex cos

02

zye x cos

Page 39: Reporter : Professor D. L. Young                 2008/01/03

39

Point Distribution with Normal Vectors

Page 40: Reporter : Professor D. L. Young                 2008/01/03

40

ResultsCross-section at x=0

MFS (d=0.5) MFS (d=1) HMM 2826 nodes 2826 nodes 2826 nodes RMSE: 9.63E22 RMSE: 1.26E-4 RMSE: 4.12E-2

( analytical solution, numerical result)

- 1 - 0 . 5 0 0 . 5 1- 1

- 0 . 5

0

0 . 5

1

1 . 5

2

2 . 5

Page 41: Reporter : Professor D. L. Young                 2008/01/03

41

Case 3-3:

02

1

0

Page 42: Reporter : Professor D. L. Young                 2008/01/03

42

ResultsCross-section at x=0

2826 nodes 2981 nodes 2826 nodes (a) MFS (d=1) (b) LDQ (c) HMM

-1 -0 .5 0 0.5 1-1

-0.5

0

0.5

1

1.5

2

2.5

-1 -0.5 0 0.5 1-1

-0.5

0

0.5

1

1.5

2

2.5

Page 43: Reporter : Professor D. L. Young                 2008/01/03

43

Case 3-4

02 1

0

Page 44: Reporter : Professor D. L. Young                 2008/01/03

44

Point Distribution with Normal Vectors

2261 nodes

Page 45: Reporter : Professor D. L. Young                 2008/01/03

45

ResultsCross-section at z=0

2826 nodes 2981 nodes 2826 nodes (a) MFS (d=2) (b) LDQ (c) HMM

-1 -0.5 0 0.5 10

0.5

1

1.5

2

2.5

3

-1 -0.5 0 0.5 10

0.5

1

1.5

2

2.5

3

Page 46: Reporter : Professor D. L. Young                 2008/01/03

4646

Exterior Acoustics Exterior Acoustics ProblemsProblems

Department of Civil Engineering and Hydrotech Research Institute Department of Civil Engineering and Hydrotech Research Institute National Taiwan UniversityNational Taiwan University

Page 47: Reporter : Professor D. L. Young                 2008/01/03

47

Formulation Governing equation:

Sommerfeld radiation condition:

the representation of the solution for exterior problem can be approximated as:

Kernel functions:

,0223 xkx eDx

3

1,

ij

kkikr

ijije

rnyeikr

xsAij

5

22 131,

ij

kkllikr

ijijklikr

ijkkllikr

ije

rnynyeikrrnneikrnynyek

xsBijijij

1

( ) ( , ) , N

i e j i j e

j

x A s x x D

1

( ) ( , ) , N

i e j i j e

j

x B s x x D

1 ( 1)2lim ( ) 0,

d

rr ik r

r

Page 48: Reporter : Professor D. L. Young                 2008/01/03

48

The Diagonal Coefficient of Influence Matrices

3,,limij

kkijeije

sx rnyxsAxsA

ji

ikr

nynyrnnxsBxsB

ij

llkkijkkijeije

sx ji4

3,,

2

5

2

lim

The kernel function will be approximated by:

The diagonal coefficients for the exterior problem can be extracted out as:

NNNN

N

N

mm

N

N

mm

i

aaa

aaaa

aaaa

,2,1,

,21

2,2,21,2

,12,11

1,1,1

NNNN

N

N

mm

N

N

mm

i

bbb

bbbb

bbbb

,2,1,

,21

2,2,21,2

,12,11

1,1,1

Page 49: Reporter : Professor D. L. Young                 2008/01/03

49

Analytical derivation of diagonal coefficients Analytical solution:

:kwhere

:Nwave number

number of nodes

radius of sphere:

}))!(()!(4

)!22()!2()()()124(

)!(4))!2(()()()14(2{

2

22)12(

)1()12(

'1

1 0)12(

042

2)1(

)2('

)2(

2

lmmlmmkhkjlm

mmkhkjm

Nika

mm

N

l mm

mmmmii

}))!(()!(4

)!22()!2()()()124(

)!(4))!2(()()()14(2{

2

22)12(

)1()12(

'1

1 0

')12(

042

2)1(

)2(''

)2(

3

lmmlmmkhkjlm

mmkhkjm

Nikb

mm

N

l mm

mmmmii

Page 50: Reporter : Professor D. L. Young                 2008/01/03

50

Scattering of a Plane Wave by a Soft Sphere

0),,(22 rk

1,1 ka

Governing equation:

Plane wave incidence:y

z

a

x

cossinsincossin

rzryrx

ikzi eAnalytical solution of total field:

0

1 cos12,,n

nnnnnt Pkrhakrjnir

J. J. Bowman,T. B. A. Senior, P. L. E.Uslenghi, Electromagnetic and Acoustic Scattering by Simple Shapes, Hemisphere publishing Corp., 1987.

Analytical solution of the scattered field:it

Page 51: Reporter : Professor D. L. Young                 2008/01/03

51

Point Distribution and Normal Vectors

1866 nodes

Page 52: Reporter : Professor D. L. Young                 2008/01/03

52

ResultsValues on the y=0, z=0 line

0.8 1.2 1.6 2 2.4 2.8 3.2

-1

-0.8

-0.6

-0.4

-0.2

0

0.2

0.8 1.2 1.6 2 2.4 2.8 3.2

-1

-0.8

-0.6

-0.4

-0.2

0

0.2

0.8 1.2 1.6 2 2.4 2.8 3.2r

-1

-0.8

-0.6

-0.4

-0.2

0

0.2

A na lytical so lution

M FS

HM M

Real part Imaginary part

0.8 1.2 1.6 2 2.4 2.8 3.2

-0.4

-0.3

-0.2

-0.1

0

0.8 1.2 1.6 2 2.4 2.8 3.2

-0.4

-0.3

-0.2

-0.1

0

0.8 1.2 1.6 2 2.4 2.8 3.2r

-0.4

-0.3

-0.2

-0.1

0

Ana lytica l solu tion

M FS

H M M

Page 53: Reporter : Professor D. L. Young                 2008/01/03

53

ResultsCross-section of y=0 plan for real part

RMSE: 8.73E-5 RMSE: 4.73E-3 1866 nodes 1866 nodes MFS HMM

Exact solution

Numerical solution

- 2 - 1 . 5 - 1 - 0 . 5 0 0 . 5 1 1 . 5 2- 2

- 1 . 5

- 1

- 0 . 5

0

0 . 5

1

1 . 5

2

- 2 - 1 . 5 - 1 - 0 . 5 0 0 . 5 1 1 . 5 2- 2

- 1 . 5

- 1

- 0 . 5

0

0 . 5

1

1 . 5

2

Page 54: Reporter : Professor D. L. Young                 2008/01/03

54

ResultsCross-section of y=0 plan for imaginary part

RMSE: 1.26E-5 RMSE: 9.69E-3 1866 nodes 1866 nodes MFS HMM

Exact solution

Numerical solution

- 2 - 1 . 5 - 1 - 0 . 5 0 0 . 5 1 1 . 5 2- 2

- 1 . 5

- 1

- 0 . 5

0

0 . 5

1

1 . 5

2

- 2 - 1 . 5 - 1 - 0 . 5 0 0 . 5 1 1 . 5 2- 2

- 1 . 5

- 1

- 0 . 5

0

0 . 5

1

1 . 5

2

Page 55: Reporter : Professor D. L. Young                 2008/01/03

55

Scattering of a Plane Wave by a Rigid Sphere

0),,(22 rk

11

ka

Governing equation:

Sommerfeld radiation condition:

Neumann boundary condition:

4

04 sincoscos,

n

n nnP

y

z

a

x

:4nP Associated Lengendre polynomial

cossinsincossin

rzryrx

Page 56: Reporter : Professor D. L. Young                 2008/01/03

56

Analytical Solution

nBnAPkrhr nnm

n

sincoscos,, 444

4

04

,1

1

44

r

n

rkrh

A 1

44

1

r

n

rkrh

B

where

K. Gerdes, L. Demkowicz, Solution of 3D-Laplace and Helmholtz equations in exterior domains using hp-infinite elements. Comput. Meth. Appl. Mech. Eng. 137 1996 239–273.

:4h Spherical Hankel function of the first kind

Page 57: Reporter : Professor D. L. Young                 2008/01/03

57

Results

Real part Imaginary part

,

2,2

,

2,2

0 100 200 300 400

-2

-1

0

1

2

0 100 200 300 400

-2

-1

0

1

2

0 100 200 300 400

-2

-1

0

1

2Exact solution

M FS

HM M

Page 58: Reporter : Professor D. L. Young                 2008/01/03

58

ResultsCross-section of

RMSE: 1.34E-4 RMSE: 1.47E-2 1866 nodes 1866 nodes MFS HMM

),

2,(Re r

- 2 - 1 . 5 - 1 - 0 . 5 0 0 . 5 1 1 . 5 2- 2

- 1 . 5

- 1

- 0 . 5

0

0 . 5

1

1 . 5

2

- 2 - 1 . 5 - 1 - 0 . 5 0 0 . 5 1 1 . 5 2- 2

- 1 . 5

- 1

- 0 . 5

0

0 . 5

1

1 . 5

2

Exact solution

Numerical solution

Page 59: Reporter : Professor D. L. Young                 2008/01/03

59

ResultsCross-section of

RMSE: 5.98E-3 RMSE: 5.76E-2 1866 nodes 1866 nodes MFS HMM

),

2,(Im r

- 2 - 1 . 5 - 1 - 0 . 5 0 0 . 5 1 1 . 5 2- 2

- 1 . 5

- 1

- 0 . 5

0

0 . 5

1

1 . 5

2

- 2 - 1 . 5 - 1 - 0 . 5 0 0 . 5 1 1 . 5 2- 2

- 1 . 5

- 1

- 0 . 5

0

0 . 5

1

1 . 5

2

Exact solution

Numerical solution

Page 60: Reporter : Professor D. L. Young                 2008/01/03

60

Scattering of a Plane Wave by a Soft Bean Shape Obstacle

,0),,(22 rk

Governing equation:

Sommerfeld radiation condition:

Plane wave incidence:

ikxi e

M. Ganesh and I.G. Graham, A high-order algorithm for obstacle scattering in three dimensions, J. Comput. Phys. 198 2004 211–242.

Radius parameter R:

RzRzyRzR

Rzx

2

22

cos4.0164.0cos3.0

cos1.0164.0

1k

Page 61: Reporter : Professor D. L. Young                 2008/01/03

61

Mesh of the Bean Shape Obstacle

2600 elements

Page 62: Reporter : Professor D. L. Young                 2008/01/03

62

Point Distribution and Normal Vectors

2500 nodes

Page 63: Reporter : Professor D. L. Young                 2008/01/03

63

Results

real part imaginary part

Page 64: Reporter : Professor D. L. Young                 2008/01/03

64

Results of Real Part by MFS

d=-0.1 d=-0.2

d=-0.15 d=-0.25

Page 65: Reporter : Professor D. L. Young                 2008/01/03

65

Results of Imaginary Part by MFS

d=-0.1 d=-0.2

d=-0.15 d=-0.25

Page 66: Reporter : Professor D. L. Young                 2008/01/03

66

NNational ational TTaiwan aiwan UUniversityniversity

Conclusions The controversy of the artificial (off-set) boundary

outside the physical domain by using the MFS no longer exists.

From the series cases of the complex irregular shape, MFS required a lot of time to adjust the distance of source points, HMM has figured out acceptable answers immediately.

From the sensitivity test of point distribution, we can know that to obtain the high accuracy of HMM, improving the point seeding is necessary. HMM required the uniform point distribution to obtain the good results.

Page 67: Reporter : Professor D. L. Young                 2008/01/03

67

NNational ational TTaiwan aiwan UUniversityniversity

Further Researches

For the next step, to solve the Helmholtz problem in vector field which relate to the electromagnetic problem in three dimensions is we are going to do.

The combination of other numerical methods such as method of particular solutions (MPS) or domain decomposition method (DDM) and HMM to solve Poisson, Helmholtz, modified Helmholtz equation would be interesting topics to research.

Page 68: Reporter : Professor D. L. Young                 2008/01/03

6868

Department of Civil Engineering and Hydrotech Research Institute Department of Civil Engineering and Hydrotech Research Institute National Taiwan UniversityNational Taiwan University

Thank YouThank YouScientific Computing & Visualization Lab

Page 69: Reporter : Professor D. L. Young                 2008/01/03

69

The Diagonal Coefficient of Influence Matrices for BEM

ji 21

ji )()()1(4

1 2

1 12

N

pq

N

qp

ikr

j

ij

WWikr

er

nrJH

1 2

0

1 1

-ikr

( ) ( ) i j4

-1 (e -1) i j2

N N ikr

j p qp q

ij

eJ W WrG

ik

Page 70: Reporter : Professor D. L. Young                 2008/01/03

70

The Diagonal Coefficient of Influence Matrices for HMM

3,,limij

kkijeije

sx rnyxsAxsA

ji

ikr

nynyrnnxsBxsB

ij

llkkijkkijeije

sx ji4

3,,

2

5

2

lim

The kernel function will be approximated by:

The diagonal coefficients for the exterior problem can be extracted out as:

NNNN

N

N

mm

N

N

mm

i

aaa

aaaa

aaaa

,2,1,

,21

2,2,21,2

,12,11

1,1,1

NNNN

N

N

mm

N

N

mm

i

bbb

bbbb

bbbb

,2,1,

,21

2,2,21,2

,12,11

1,1,1