a localized method of particular solutions for solving near singular problems c.s. chen, guangming...

Post on 06-Jan-2018

214 Views

Category:

Documents

0 Downloads

Preview:

Click to see full reader

DESCRIPTION

2016/1/143 Radial Basis Functions Linear: Cubic: Multiquadrics: Polyharmonic Spines: Gaussian:

TRANSCRIPT

A Localized Method of Particular Solutions for Solving Near Singular Problems

C.S. Chen, Guangming Yao, D.L. Young

Department of MathematicsUniversity of Southern Mississippi

U.S.A.

112/05/03 2

OutlineOutline

• Radial Basis Functions

• The global approaches of the method of particular solutions

• Numerical examples of global method

• Local approach of the method of particular solutions

• Numerical examples of local method

• Near Singular Problems

112/05/03 3

Radial Basis Functions

Linear: rCubic: r 3

Multiquadrics: r c c2 2 where is a shape parameter.

Polyharmonic Spines:r r n

r n

n

n

2

2 1

11

log , ,, ,

in 2D, in 3D.

Gaussian: e cr 2

Let : be a continous function with (0) 0. If , letiR R x ,i i i x x x

where is the Euclidean norm. Then is called the RBF.i

112/05/03 4

Assume that )(ˆ)( xx ff To approximate f by

f we usually require fitting the given

data set xi

N

1of pairwise distinct centres with the imposed

conditions ˆ( ) ( ), 1 .i if f i N x x

The linear system 1

ˆ ( ) , 1 ,N

i i i ji

f a i N

x x x

is well-posed if the interpolation matrix is non-singular

1i j i NA

x x

Surface Reconstruction Scheme

112/05/03 5

The Splitting Method

Consider the following equation

,),( xxfLu( ), ,Bu g x x

Where ,3,2, dRd is a bounded open nonempty domain

with sufficiently regular boundary .Let puuv where pu satisfying )(xfLu p but does not necessary satisfy the boundary condition in (11).

(10)(11)

v satisfies , ,0 xLv. ),()( xxx pugBv

(12)

(13)(14)

112/05/03 6

Assume that )(ˆ)( xx ff and that we can obtain an analytical solution up

to

ˆˆ ( ).pLu f xThen

.ˆ pp uu

To approximate f by f we usually require fitting the given

data set xi

N

1 of pairwise distinct centres with the imposed

conditions

.1 ),(ˆ)( Niff xx

Particular Solutions

112/05/03 7

The linear system

1

ˆ ( ) , 1 ,N

i ji ii

af i N

x x x

is well-posed if the interpolation matrix is non-singular

1i j i NA

x x

ˆOnce in (*) has been established,f

1

ˆ i

N

p ii

u a

where

i iL and

, .i i i i x x x x

(*)

i iL

112/05/03 8

For 2 ,L 2 1( ) d dr dr drr r in 2D

2 2

2 4 41 116 32

1

( ) ln , =

ln ( ) lnd dr dr dr

r r r

r r r r r r r

2 2

2 2 2 2 2 2 231 1

9

( ) +c

( ) ln +c 4 +cc

r r

r c r c r c r

112/05/03 9

112/05/03 10

112/05/03 11

Where G(r) is the fundamental solution of L

Boundary Method is required.

112/05/03 12

The Method of Particular Solutions (MPS)

1

ˆn

p j jj

u u a

,),( xxfLu

( ), ,Bu g x x

j jL where

112/05/03 13

Impose boundary conditions

112/05/03 14

112/05/03 15

Once { } is known, the solution of PDEs

can be expressed as followsja

1

ˆ ( )n

j jj

u a r

112/05/03 16

Numerical Results

112/05/03 17

Example I

Analytical solution:

Computational Domain:

112/05/03 18

112/05/03 19c : shape parameter of MQ

112/05/03 20

Consider the Poisson’s equation( , ), ( , )

( , ), ( , )u f x y x y

u g x y x y

Given a large data set 1,

n

i i ix y

( ), ,( ), .

i ii

i i

f x xy

g x x

where

112/05/03 21

112/05/03 22

112/05/03 23

112/05/03 24

112/05/03 25

112/05/03 26

112/05/03 28

1x 2x 3x4x 5x

112/05/03 29

1x 2x 3x4x 5x

112/05/03 30

1x 2x 3x4x 5x

112/05/03 31

1x 2x 3x4x 5x

112/05/03 Non-Dirichlet boundary condition

112/05/03 33

112/05/03 34

112/05/03 35

112/05/03 36

112/05/03 37

112/05/03 38

112/05/03 39

The absolute errors of LMAPS with L=1, n=5, Sn=100, c=8.9

112/05/03

L=1, Sn = 100, N=225.

112/05/03 41

Local MPS verse Global MPS

112/05/03 42

n: number of neighbor points

112/05/03 43

112/05/03 44

112/05/03 45

LMPS verse LMQ

112/05/03

Near Singular Problem I

C.S. Chen, G. Kuhn, J. Li, G. Mishuris, Radial basis functions for solving near singular Poisson’s problems,Communication in Numerical Methods in Engineering, 2003, 19, 333-347.

112/05/03

1.5a

112/05/03 48

Profile of exact solution

112/05/03 49

CS-RBF

400 quasi-random points

112/05/03 50

Test 1 Test 2

112/05/03 51

Normalized Shape parameter

where

112/05/03 52

112/05/03 53

Sobel quasi-random nodes Von-Del Corput quasi-random nodes

Random nodes

112/05/03 54

Speed up

N=10,000 CPU = 0.5/3.42 sN=40,000 CPU = 3.31/14.06 sN=62,500 CPU = 7.01/25.28 s

112/05/03 55

RMSE error verse shape parameter for a=1.6 and various mesh sizes

LMPS

112/05/03 56

RMSE error verse shape parameter for h=1/200, and various value of a.

112/05/03 57

112/05/03 58

Near Singular Problem II

Exact solution

112/05/03

Profile of f(x,y)

f(1,1,) = -15,861, f(0,0)=237

112/05/03 60

112/05/03 61

Near Singular Problem III

112/05/03 62

112/05/03 63

Adaptive Method

First step Second step

112/05/03 64

3rd step 4th step

112/05/03 65

top related