Fasshauer 2008 Lecture4

Meshfree Approximation with M ATLAB
Lecture IV: RBF Collocation and Polynomial Pseudospectral Methods
Greg Fasshauer
Department of Applied Mathematics

Illinois Institute of Technology
Dolomites Research Week on Approximation

September 8–11, 2008
[email protected] Lecture IV Dolomites 2008

Outline
1 RBF Collocation, Kansa’s method
2 PS Methods and Differentiation Matrices
3 PDEs with BCs via PS Methods
4 Symmetric RBF collocation
5 RBF Differentiation Matrices in M ATLAB
6 Solving PDEs via RBF-PS Methods

RBF Collocation, Kansa’s method PDEs of Interest
Linear elliptic PDE with boundary conditions
Lu = f in Ω
u = g on Γ = ∂Ω

RBF Collocation, Kansa’s method PDEs of Interest
Linear elliptic PDE with boundary conditions
Lu = f in Ω
u = g on Γ = ∂Ω
Time-dependent PDE with initial and boundary conditions
ut (x, t) + Lu(x, t) = f (x, t), x ∈ Ω ∪ Γ, t ≥ 0

u(x, 0) = u0 (x), x ∈Ω
u(x, t) = g(t), x ∈Γ

RBF Collocation, Kansa’s method Kansa’s (non-symmetric) Method
According to [Kansa (1986)] we consider an elliptic PDE and start with

N
X
u(x) = λj Φj (x) = ΦT (x)λ
j=1

According to [Kansa (1986)] we consider an elliptic PDE and start with

N
X
u(x) = λj Φj (x) = ΦT (x)λ
j=1
Coefficients λ determined by solving

" #
A
eL f
λ= ,
A
e g
with (rectangular) collocation matrices

A
e L,ij = LΦj (x i ) = Lϕ(kx − x j k)x=x ,
i
i = 1, . . . , N − NB , j = 1, . . . , N,
A
e ij = Φj (x i ) = ϕ(kx i − x j k),
i = N − NB + 1, . . . , N, j = 1, . . . , N.

" #
A
eL
If invertible (open problem since 1986), then approximate
A
e
solution (at any point x) found by inserting λ in basis expansion, i.e.,
u(x) = ΦT (x)λ

" #
A
eL
If invertible (open problem since 1986), then approximate
A
e
solution (at any point x) found by inserting λ in basis expansion, i.e.,
u(x) = ΦT (x)λ
Solution at collocation points

" #−1
A
eL f
u=A , Aij = Φj (x i )
Ae g
| {z }
=λ

Note " #−1 # "

A
eL f A
eL −1 f
u=A ⇐⇒ A u =
Ae g Ae g
| {z }
=LΓ

Note " #−1 # "

A
eL f A
eL −1 f
u=A ⇐⇒ A u =
Ae g Ae g
| {z }
=LΓ
with " #
e L A−1 Ae L A−1

A I B M P
LΓ = e −1 e −1 = ,
AA I AA B
0 I
−→ RBF-PS method (see later).

Example (2D Laplace Equation)
uxx + uyy = 0, x, y ∈ (−1, 1)2

Boundary conditions:
 4
sin (πx),
 y = 1 and − 1 < x < 0,
1
u(x, y ) = 5 sin(3πy ), x = 1,

0, otherwise.

See [Trefethen (2000)]

Figure: Gaussian-RBF (ε = 2.75), N = 24 × 24

Program (KansaLaplaceMixedBCTref_2D.m)
global rbf Lrbf; rbf_definition; epsilon = 3;
N = 289; M = 1681;
Lu = @(x,y) zeros(size(x));
[collpts, N] = CreatePoints(N, 2, ’u’); collpts = 2*collpts-1;
indx = find(abs(collpts(:,1))==1 | abs(collpts(:,2))==1);
bdypts = collpts(indx,:);
intpts = collpts(setdiff([1:N],indx),:);
ctrs = [intpts; bdypts];
evalpts = CreatePoints(M,2,’u’); evalpts = 2*evalpts-1;
DM_eval = DistanceMatrix(evalpts,ctrs);
EM = rbf(epsilon,DM_eval);
DM_int = DistanceMatrix(intpts,ctrs);
DM_bdy = DistanceMatrix(bdypts,ctrs);
LCM = Lrbf(epsilon,DM_int);
BCM = rbf(epsilon,DM_bdy);
CM = [LCM; BCM];
rhs = zeros(N,1); NI = size(intpts,1);
indx = find(bdypts(:,1)==1 | (bdypts(:,1)<0) & (bdypts(:,2)==1));
rhs(NI+indx) = (bdypts(indx,1)==1)*0.2.*sin(3*pi*bdypts(indx,2)) + ...
(bdypts(indx,1)<0).*(bdypts(indx,2)==1).*sin(pi*bdypts(indx,1)).^4;
Pf = EM * (CM\rhs);
disp(sprintf(’u(0,0) = %16.12f’,Pf(841)))

We just showed that we always have (even if the Kansa matrix is not
invertible) " #
A
eL −1 M P
LΓ = A =
A
e 0 I

invertible) " #
A
eL −1 M P
LΓ = A =
A
e 0 I
It is known that
LΓ is invertible for polynomial basis (1D)
In a certain limiting case RBF interpolant yields polynomial
interpolant

invertible) " #
A
eL −1 M P
LΓ = A =
A
e 0 I
It is known that
interpolant
=⇒ Kansa’s collocation matrix is invertible in the limiting case

invertible) " #
A
eL −1 M P
LΓ = A =
A
e 0 I
It is known that
interpolant
=⇒ Kansa’s collocation matrix is invertible in the limiting case
Other recent work on a well-defined approach to Kansa’s method, e.g.,

in [Schaback (2007)]

RBF Collocation, Kansa’s method Time-Dependent PDEs
Consider time-dependent PDE
ut + Lu = f
and discretize the time, e.g.,
u(x, tk ) − u(x, tk −1 ) uk (x) − uk −1 (x)

ut (x, tk ) ≈ =
∆t ∆t

RBF Collocation, Kansa’s method Time-Dependent PDEs
Consider time-dependent PDE
ut + Lu = f
and discretize the time, e.g.,
u(x, tk ) − u(x, tk −1 ) uk (x) − uk −1 (x)

ut (x, tk ) ≈ =
∆t ∆t
Then
uk (x) ≈ uk −1 (x) + ∆t [f (x) − Luk −1 (x)]

RBF Collocation, Kansa’s method Kansa for Time-Dependent PDEs
Start with
N
X
u(x, t) = λj (t)Φj (x) = ΦT (x)λ(t)
j=1

Start with
N
X
j=1
Then after time discretization

 
N N N
(k ) (k −1) (k −1)
X X X
λj Φj (x) = λj Φj (x) + ∆t f (x) − λj LΦj (x)
j=1 j=1 j=1

Start with
N
X
j=1
Then after time discretization

 
N N N
(k ) (k −1) (k −1)
X X X
λj Φj (x) = λj Φj (x) + ∆t f (x) − λj LΦj (x)
j=1 j=1 j=1
or h i
ΦT (x)λ(k ) = ΦT (x) − ∆tLΦT (x) λ(k −1) + ∆tf (x)

RBF Collocation, Kansa’s method Application to Time-Dependent PDEs
Collocation on X = {x 1 , . . . , x N } yields
Aλ(k ) = [A − ∆tAL ] λ(k −1) + ∆tf

Aλ(k ) = [A − ∆tAL ] λ(k −1) + ∆tf
Solve for λ(k ) , then (for any x)

N
(k )
X
u (k )
(x) = λj Φj (x) = ΦT (x)λ(k )
j=1

Aλ(k ) = [A − ∆tAL ] λ(k −1) + ∆tf

N
(k )
X
u (k )
(x) = λj Φj (x) = ΦT (x)λ(k )
j=1
u (k ) = Aλ(k ) , Aij = Φj (x i )

Aλ(k ) = [A − ∆tAL ] λ(k −1) + ∆tf

N
(k )
X
u (k )
(x) = λj Φj (x) = ΦT (x)λ(k )
j=1
u (k ) = Aλ(k ) , Aij = Φj (x i )
Much too complicated!

PS Methods and Differentiation Matrices Background
PS methods are known as highly accurate solvers for PDEs
Basic idea (in 1D)

N
X
u(x) = λj Φj (x), x ∈ [a, b]
j=1
with (smooth and global) basis functions Φj , j = 1, . . . , N
Here u is the unknown (approximate) solution of the PDE

PS Methods and Differentiation Matrices Differentiation Matrices
For PDEs we need to be able to represent values of derivatives of u.
For PS methods values at grid points suffice.
Key idea: find differentiation matrix D such that
u 0 = Du
where u = [u(x1 ), . . . , u(xN )]T

PS Methods and Differentiation Matrices Differentiation Matrices
For PDEs we need to be able to represent values of derivatives of u.
For PS methods values at grid points suffice.
Key idea: find differentiation matrix D such that
u 0 = Du
where u = [u(x1 ), . . . , u(xN )]T
Example
Chebyshev polynomials on Chebyshev points. In this case explicit
formulas for entries of D are known.

PS Methods and Differentiation Matrices General Differentiation Matrices
N N
X
0
X d
u(x) = λj Φj (x) =⇒ u (x) = λj Φj (x).
dx
j=1 j=1

N N
X
0
X d
u(x) = λj Φj (x) =⇒ u (x) = λj Φj (x).
dx
j=1 j=1
Evaluate at grid points:
u = Aλ with Aij = Φj (xi )
and
d
u 0 = Ax λ with Ax,ij = Φj (xi ).
dx

N N
X
0
X d
u(x) = λj Φj (x) =⇒ u (x) = λj Φj (x).
dx
j=1 j=1
Evaluate at grid points:
u = Aλ with Aij = Φj (xi )
and
d
u 0 = Ax λ with Ax,ij = Φj (xi ).
dx
Therefore
u 0 = Ax A−1 u =⇒ D = Ax A−1 .

PS Methods and Differentiation Matrices RBF Differentiation Matrices
Want to use radial basis functions
Φj (x) = ϕ(kx − x j k), x ∈ Rs
and a general linear differential operator L with constant coefficients.

Discretized differential operator (differentiation matrix):
L = AL A−1
with Aij = Φj (x i ) = ϕ(kx i − x j k)

and AL,ij = LΦj (x i ) = Lϕ(kx − x j k)x=x .
i

Discretized differential operator (differentiation matrix):
L = AL A−1
with Aij = Φj (x i ) = ϕ(kx i − x j k)

and AL,ij = LΦj (x i ) = Lϕ(kx − x j k)x=x .
i
A is (non-singular) RBF interpolation matrix

How to proceed for a time-independent PDE
Lu = f
Discretize space
Lu ≈ Lu
Then (at grid points)
u = L−1 f

How to proceed for a time-independent PDE
Lu = f
Discretize space
Lu ≈ Lu
Then (at grid points)
u = L−1 f
Challenge
Need to ensure invertibility of L

PS Methods and Differentiation Matrices A First (ill-posed) PDE
Want to solve Lu = f without BCs.
Discretize:
Lu = f =⇒ u = L−1 f = A (AL )−1 f
Invertibility of L (and therefore AL ) required.

Discretize:
Lu = f =⇒ u = L−1 f = A (AL )−1 f
Chebyshev PS: L is singular

Discretize:
Lu = f =⇒ u = L−1 f = A (AL )−1 f

RBF-PS: L is non-singular since AL invertible for positive definite
RBFs and elliptic L.

Discretize:
Lu = f =⇒ u = L−1 f = A (AL )−1 f

RBF-PS: L is non-singular since AL invertible for positive definite
RBFs and elliptic L.
Remark
RBFs “too good to be true”. Built-in regularization due to variational
framework.

PDEs with BCs via PS Methods
Consider linear elliptic PDE
Lu = f in Ω
with boundary condition
u=g on Γ = ∂Ω
and assume basis functions do not satisfy BCs.

Consider linear elliptic PDE
Lu = f in Ω
with boundary condition
u=g on Γ = ∂Ω
and assume basis functions do not satisfy BCs.
Construct differentiation matrix L based on all grid points x i .

Then replace diagonal entries corresponding to boundary points
with ones and the remainder of those rows with zeros.

Reorder rows and columns to obtain

M P
LΓ = ,
0 I
Here
M is NI × NI (interior points)
I is NB × NB (boundary points)
NB number of grid points on boundary Γ
NI = N − NB number of grid points in the interior Ω

Reorder rows and columns to obtain

M P
LΓ = ,
0 I
Here
M is NI × NI (interior points)
I is NB × NB (boundary points)
NB number of grid points on boundary Γ
NI = N − NB number of grid points in the interior Ω
Then
f
u = L−1
Γ g

Using u = [u Ω , u Γ ]T and substituting u Γ = g back in we get
u Ω = M −1 (f − Pg),
or, for homogeneous boundary conditions,
u Ω = M −1 f .

Using u = [u Ω , u Γ ]T and substituting u Γ = g back in we get
u Ω = M −1 (f − Pg),
or, for homogeneous boundary conditions,
u Ω = M −1 f .
Remark
For standard PS methods the block M is invertible. Its spectrum is well
studied for many different L and BCs.

Symmetric RBF collocation
As in [F. (1997)] we start with

NI
X N
X
u(x) = λj L∗j Φ(x) + λj Φj (x).
j=1 j=NI +1
Since Φj (x) = ϕ(kx − x j k), L∗j denotes application of L to ϕ viewed as

a function of the second variable followed by evaluation at x j .

As in [F. (1997)] we start with

NI
X N
X
j=1 j=NI +1
Since Φj (x) = ϕ(kx − x j k), L∗j denotes application of L to ϕ viewed as

a function of the second variable followed by evaluation at x j .
Then λ = [λΩ , λΓ ]T obtained from

" #
ÂLL∗ ÂL λΩ f
= .
ÂL∗ Â λΓ g

The collocation matrix " #

ÂLL∗ ÂL
ÂL∗ Â
consists of
square blocks
h i
ÂLL∗ ,ij = L [L∗ ϕ(kx − ξk)]ξ=x j , i, j = 1, . . . , NI
x=x i
Âij = Φj (x i ) = ϕ(kx i − x j k), i, j = NI + 1, . . . , N
rectangular blocks

ÂL,ij = Lϕ(kx − x j k) x=x , i = 1, . . . , NI , j = NI + 1, . . . , N
i
ÂL∗ ,ij = [L∗ ϕ(kx i − xk)]x=x j , i = NI + 1, . . . , N, j = 1, . . . , NI

2D Laplace Equation from [Trefethen (2000)]
Figure: Gaussian-RBF (ε = 6), N = 17 × 17 (+ 64 boundary points)

Program (HermiteLaplaceMixedBCTref_2D.m)
global rbf Lrbf L2rbf; rbf_definition; epsilon = 3;
N = 289; M = 1681;
Lu = @(x,y) zeros(size(x));
[datasites, N] = CreatePoints(N, 2, ’u’); intdata = 2*datasites-1;
sg = sqrt(N); bdylin = linspace(-1,1,sg)’; bdy1 = ones(sg-1,1);
bdydata = [bdylin(1:end-1) -bdy1; bdy1 bdylin(1:end-1);...
flipud(bdylin(2:end)) bdy1; -bdy1 flipud(bdylin(2:end))];
h = 2/(sg-1); bdylin = (-1+h:h:1-h)’; bdy0 = repmat(-1-h,sg-2,1); bdy1 = repmat(1+h,sg-2,1);
bdycenters = [-1-h -1-h; bdylin bdy0; 1+h -1-h; bdy1 bdylin;...
1+h 1+h; flipud(bdylin) bdy1; -1-h 1+h; bdy0 flipud(bdylin)];
centers = [intdata; bdycenters];
evalpoints = CreatePoints(M, 2, ’u’); evalpoints = 2*evalpoints-1;
DM_inteval = DistanceMatrix(evalpoints,intdata);
DM_bdyeval = DistanceMatrix(evalpoints,bdycenters);
LEM = Lrbf(epsilon,DM_inteval);
BEM = rbf(epsilon,DM_bdyeval);
EM = [LEM BEM];
DM_IIdata = DistanceMatrix(intdata,intdata);
DM_IBdata = DistanceMatrix(intdata,bdycenters);
DM_BIdata = DistanceMatrix(bdydata,intdata);
DM_BBdata = DistanceMatrix(bdydata,bdycenters);
LLCM = L2rbf(epsilon,DM_IIdata);
LBCM = Lrbf(epsilon,DM_IBdata);
BLCM = Lrbf(epsilon,DM_BIdata);
BBCM = rbf(epsilon,DM_BBdata);
CM = [LLCM LBCM; BLCM BBCM];
rhs = [Lu(intdata(:,1),intdata(:,2)); zeros(sg-1,1); 0.2*sin(3*pi*bdydata(sg:2*sg-2,2)); ...
zeros((sg-1)/2,1); sin(pi*bdydata((5*sg-3)/2:3*sg-3,1)).^4; zeros(sg-1,1)];
Pf = EM * (CM\rhs);
disp(sprintf(’u(0,0) = %16.12f’,Pf(841)))

It is well known that the symmetric collocation matrix

" #
ÂLL∗ ÂL
ÂL∗ Â
is invertible.

It is well known that the symmetric collocation matrix

" #
ÂLL∗ ÂL
ÂL∗ Â
is invertible.
Therefore, the solution at any point is obtained by inserting λ into basis

expansion, or at grid points
" #−1
h
eT
i ÂLL∗ ÂL f
u= AL∗ A
| {z } ÂL∗ Â g
=K T
| {z }
=λ
with evaluation matrix K T .

Entries of K T given by
AL∗ ,ij = [L∗ ϕ(kx i − xk)]x=x j , i = 1, . . . , N, j = 1, . . . , NI

eT
A = Φj (x i ) = ϕ(kx i − x j k), i = 1, . . . , N, j = NI + 1, . . . , N
ij

Entries of K T given by
AL∗ ,ij = [L∗ ϕ(kx i − xk)]x=x j , i = 1, . . . , N, j = 1, . . . , NI

eT
A = Φj (x i ) = ϕ(kx i − x j k), i = 1, . . . , N, j = NI + 1, . . . , N
ij
e T , and therefore
Symmetry of RBFs implies AL∗ = A L
" #T
T
h
eT
i A
eL
K = AL∗ A =
Ae
→ transpose of Kansa’s matrix

Symmetric RBF collocation Symmetric RBF-PS Method
Start with “symmetric basis” expansion

NI
X N
X
j=1 j=NI +1
At the grid points in matrix notation we have

h i λ
T Ω
u = AL∗ A e
λΓ
or h i−1
λ= AL∗ eT
A u.

Apply L to basis expansion and restrict to grid

Lu = ALL∗ AL λ
with
h i
ALL∗ ,ij = L [L∗ ϕ(kx − ξk)]ξ=x j , i = 1, . . . , N, j = 1, . . . , NI
x=x i

AL,ij = Lϕ(kx − x j k) x=x , i = 1, . . . , N, j = NI + 1, . . . , N
i

Apply L to basis expansion and restrict to grid

Lu = ALL∗ AL λ
with
h i
ALL∗ ,ij = L [L∗ ϕ(kx − ξk)]ξ=x j , i = 1, . . . , N, j = 1, . . . , NI
x=x i

AL,ij = Lϕ(kx − x j k) x=x , i = 1, . . . , N, j = NI + 1, . . . , N
i
Replace λ and get

h i−1
L̂u = ALL∗ AL A L∗ eT
A u.
Remark
Note that L̂ differs from
" #
ÂLL∗ ÂL h i−1
L̂Γ = AL∗ eT
A
ÂL∗ Â
since the BCs are not yet enforced.

Symmetric RBF collocation RBF-PS Methods: Summary
Non-symmetric (Kansa):

" #
A
eL
Can formulate discrete differential operator LΓ = A−1
Ae

" #
A
eL
Ae
Cannot ensure general invertibility of LΓ

" #
A
eL
Ae
=⇒ OK for time-dependent PDEs

" #
A
eL
Ae
Symmetric:

" #
A
eL
Ae
Symmetric:
Can ensure general solution of Lu = f

" #
A
eL
Ae
Symmetric:
Cannot" in general formulate
# discrete differential operator
L̂Γ = AL∗ A eT
ÂL∗ Â

" #
A
eL
Ae
Symmetric:
Cannot" in general formulate
# discrete differential operator
L̂Γ = AL∗ A eT
ÂL∗ Â
=⇒ OK for time-independent PDEs

RBF Differentiation Matrices in M ATLAB
First-order Derivatives
The chain rule says
∂ xj d
ϕ(kxk) = ϕ(r ),
∂xj r dr
q
where xj is a component of x, and r = kxk = x12 + . . . + xs2 .

The chain rule says
∂ xj d
ϕ(kxk) = ϕ(r ),
∂xj r dr
q
This implies we need to
provide code for derivatives of ϕ, e.g., for the Gaussian
dxrbf = @(ep,r,dx) -2*dx*ep^2.*exp(-(ep*r).^2);

The chain rule says
∂ xj d
ϕ(kxk) = ϕ(r ),
∂xj r dr
q
compute distances r (with DistanceMatrix.m),

The chain rule says
∂ xj d
ϕ(kxk) = ϕ(r ),
∂xj r dr
q
compute distances r (with DistanceMatrix.m),
and compute differences in x.
Program (DifferenceMatrix.m)
1 function DM = DifferenceMatrix(dcoord,ccoord)
2 [dr,cc] = ndgrid(dcoord(:),ccoord(:));
3 DM = dr-cc;
Program (DRBF.m)
1 function [D,x] = DRBF(N,rbf,dxrbf,ep)

2 if N==0, D=0; x=1; return, end
3 x = cos(pi*(0:N)/N)’; % Chebyshev points
4 r = DistanceMatrix(x,x);
5 dx = DifferenceMatrix(x,x);
6 A = rbf(ep,r);
7 Ax = dxrbf(ep,r,dx);
8 D = Ax/A;
Remark
The differentiation matrix is given by D = Ax A−1 . In M ATLAB we
implement this as solution of DA = Ax using mrdivide (/).
Could add a version of LOOCV to find “optimal” ε.

Solving PDEs via RBF-PS Methods
Example (1D Transport Equation)

Consider
ut (x, t) + cux (x, t) = 0, x > −1, t > 0

u(−1, t) = 0
u(x, 0) = f (x)
with solution
u(x, t) = f (x − ct)
Use implicit Euler for time discretization

Solving PDEs via RBF-PS Methods
Figure: Gaussian RBFs with “optimal” ε = 1.874528, ∆t = 0.01, collocation

on 21 Chebyshev points
Solving PDEs via RBF-PS Methods Example: Allen-Cahn Equation
Example (Allen-Cahn)
ut = µuxx + u − u 3 , x ∈ (−1, 1), t ≥ 0,

µ coupling parameter (governs transition between stable solutions),
here µ = 0.01
Initial condition:

3
u(x, 0) = 0.53x + 0.47 sin − πx , x ∈ [−1, 1],
2
u(−1, t) = −1 and u(1, t) = sin2 (t/5)

Figure: Matérn-RBF with “optimal” ε = 0.351011, collocated on 21

Chebyshev points
Figure: Chebyshev pseudospectral with 21 points

Solving PDEs via RBF-PS Methods Example: 2D Laplace Equation
Example (2D Laplace Equation)
uxx + uyy = 0, x, y ∈ (−1, 1)2

 4
sin (πx),
 y = 1 and − 1 < x < 0,
1
u(x, y ) = 5 sin(3πy ), x = 1,

0, otherwise.


Figure: Matérn-RBF (ε = 2.4), N = 24 × 24

Figure: Chebyshev pseudospectral, N = 24 × 24

Solving PDEs via RBF-PS Methods M ATLAB Code
Program (p36_2D.m)
1 rbf=@(e,r) exp(-e*r).*(15+15*e*r+6*(e*r).^2+(e*r).^3);
2 Lrbf=@(e,r)e^2*exp(-e*r).*((e*r).^3-(e*r).^2-6*e*r-6);
3 N = 24; ep = 2.4;
4 [L,x,y] = LRBF(N,rbf,Lrbf,ep);
5 [xx,yy] = meshgrid(x,y); xx = xx(:); yy = yy(:);
6 b = find(abs(xx)==1 | abs(yy)==1); % boundary pts
7 L(b,:) = zeros(4*N,(N+1)^2); L(b,b) = eye(4*N);
8 rhs = zeros((N+1)^2,1);
9a rhs(b) = (yy(b)==1).*(xx(b)<0).*sin(pi*xx(b)).^4 + ...
9b .2*(xx(b)==1).*sin(3*pi*yy(b));
10 u = L\rhs;
11 uu = reshape(u,N+1,N+1); [xx,yy] = meshgrid(x,y);
12 [xxx,yyy] = meshgrid(-1:.04:1,-1:.04:1);
13 uuu = interp2(xx,yy,uu,xxx,yyy,’cubic’);
14 surf(xxx,yyy,uuu), colormap(’default’);
15 axis([-1 1 -1 1 -.2 1]), view(-20,45)
16a text(0,.8,.4,sprintf(’u(0,0) = %12.10f’,...
16b uu(N/2+1,N/2+1)))
Remark
Note that for this type of elliptic problem we require inversion of the
differentiation matrix.
As pointed out above, we use the non-symmetric RBF-PS method
even though this may not be warranted theoretically.

Program (LRBF.m)
1 function [L,x,y] = LRBF(N,rbf,Lrbf,ep)
2 if N==0, L=0; x=1; return, end
3 x = cos(pi*(0:N)/N)’; % Chebyshev points
4 y = x; [xx,yy] = meshgrid(x,y);
5 points = [xx(:) yy(:)];
6 r = DistanceMatrix(points,points);
7 A = rbf(ep,r);
8 AL = Lrbf(ep,r);
9 L = AL/A;

Summary
Coupling RBF collocation and PS methods yields additional

insights about RBF methods
Provides “standard" procedure for solving time-dependent PDEs
with RBFs
Can apply many standard PS procedures to RBF solvers, but now
can take advantage of scattered (multivariate) grids
RBF-PS method for ε = 0 identical to Chebyshev-PS method and
more accurate for small ε
RBF-PS method has been applied successfully to a number of
engineering problems (see, e.g.,
[Ferreira & F. (2006), Ferreira & F. (2007)])

Future work:
Need to think about stable way to compute larger problems with
RBFs (preconditioning) – especially for eigenvalue problems
Need efficient computation of differentiation matrix analogous to
FFT
Can think about adaptive PS methods with moving grids

Appendix References
References I
Buhmann, M. D. (2003).
Radial Basis Functions: Theory and Implementations.
Cambridge University Press.
Fasshauer, G. E. (2007).
Meshfree Approximation Methods with M ATLAB.
World Scientific Publishers.
Fornberg, B. (1998).
A Practical Guide to Pseudospectral Methods.
Cambridge Univ. Press.
Higham, D. J. and Higham, N. J. (2005).
M ATLAB Guide.
SIAM (2nd ed.), Philadelphia.
Trefethen, L. N. (2000).
Spectral Methods in M ATLAB.
SIAM (Philadelphia, PA).

Appendix References
References II
Wendland, H. (2005).
Scattered Data Approximation.
Cambridge University Press.
Fasshauer, G. E. (1997).
Solving partial differential equations by collocation with radial basis functions.
in Surface Fitting and Multiresolution Methods, A. Le Méhauté, C. Rabut, and L.
L. Schumaker (eds.), Vanderbilt University Press (Nashville, TN), pp. 131–138.
Ferreira, A. J. M. and Fasshauer, G. E. (2006).
Computation of natural frequencies of shear deformable beams and plates by an
RBF-pseudospectral method.
Comput. Methods Appl. Mech. Engrg. 196, pp. 134–146.
Ferreira, A. J. M. and Fasshauer, G. E. (2007).
Analysis of natural frequencies of composite plates by an RBF-pseudospectral
method.
Composite Structures, 79, pp. 202–210.

Appendix References
References III
Kansa, E. J. (1986).
Application of Hardy’s multiquadric interpolation to hydrodynamics.
Proc. 1986 Simul. Conf. 4, pp. 111–117.
Schaback, R. (2007).
Convergence of unsymmetric kernel-based meshless collocation methods,
SIAM J. Numer. Anal. 45(1), pp. 333–351.

Fasshauer 2008 Lecture4

Uploaded by

Copyright:

Available Formats

Fasshauer 2008 Lecture4

Uploaded by

Document Information

Original Description:

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Fasshauer 2008 Lecture4

Uploaded by

Copyright:

Available Formats

Meshfree Approximation with M ATLAB

Lecture IV: RBF Collocation and Polynomial Pseudospectral Methods

Department of Applied Mathematics

Dolomites Research Week on Approximation

[email protected] Lecture IV Dolomites 2008

2 PS Methods and Differentiation Matrices

3 PDEs with BCs via PS Methods

4 Symmetric RBF collocation

5 RBF Differentiation Matrices in M ATLAB

6 Solving PDEs via RBF-PS Methods

[email protected] Lecture IV Dolomites 2008

Linear elliptic PDE with boundary conditions

[email protected] Lecture IV Dolomites 2008

Linear elliptic PDE with boundary conditions

Time-dependent PDE with initial and boundary conditions

ut (x, t) + Lu(x, t) = f (x, t), x ∈ Ω ∪ Γ, t ≥ 0

[email protected] Lecture IV Dolomites 2008

According to [Kansa (1986)] we consider an elliptic PDE and start with

[email protected] Lecture IV Dolomites 2008

According to [Kansa (1986)] we consider an elliptic PDE and start with

Coefficients λ determined by solving

with (rectangular) collocation matrices

[email protected] Lecture IV Dolomites 2008

[email protected] Lecture IV Dolomites 2008

Solution at collocation points

[email protected] Lecture IV Dolomites 2008

Note " #−1  # "

[email protected] Lecture IV Dolomites 2008

Note " #−1  # "

−→ RBF-PS method (see later).

[email protected] Lecture IV Dolomites 2008

Example (2D Laplace Equation)

uxx + uyy = 0, x, y ∈ (−1, 1)2

See [Trefethen (2000)]

[email protected] Lecture IV Dolomites 2008

Figure: Gaussian-RBF (ε = 2.75), N = 24 × 24

[email protected] Lecture IV Dolomites 2008

[email protected] Lecture IV Dolomites 2008

[email protected] Lecture IV Dolomites 2008

[email protected] Lecture IV Dolomites 2008

=⇒ Kansa’s collocation matrix is invertible in the limiting case

[email protected] Lecture IV Dolomites 2008

=⇒ Kansa’s collocation matrix is invertible in the limiting case

Other recent work on a well-defined approach to Kansa’s method, e.g.,

[email protected] Lecture IV Dolomites 2008

Consider time-dependent PDE

and discretize the time, e.g.,

u(x, tk ) − u(x, tk −1 ) uk (x) − uk −1 (x)

[email protected] Lecture IV Dolomites 2008

Consider time-dependent PDE

and discretize the time, e.g.,

u(x, tk ) − u(x, tk −1 ) uk (x) − uk −1 (x)

[email protected] Lecture IV Dolomites 2008

[email protected] Lecture IV Dolomites 2008

Then after time discretization

[email protected] Lecture IV Dolomites 2008

Then after time discretization

[email protected] Lecture IV Dolomites 2008

Note " #−1 # "

Note " #−1 # "