ABS and PML

Martin J. Gander

Are Absorbing Boundary Conditions

and Perfectly Matched Layers
Really so Different ?

Domain
Truncation
History
Example

TBS and ABS

Construction
Application: ABCs

PML
Historical
Stretched-Coordinate
PML

Martin J. Gander
martin.gander@unige.ch
University of Geneva

RICAM, November 2011

Joint work with Achim Sch
adle

Pole Condition
Meaning
Approximations

Mathematical
Equivalences
With PML
With ABC
Numerical
Experiments
Optimization of s0

Conclusions

History of Absorbing Boundary Conditions

Engquist and Majda (1977): Wave Propagation: In
practical calculations, it is often essential to introduce artificial
boundaries to limit the area of computation.
Unfortunately [transparent boundary conditions] necessarily have
to be non-local in both space and time and thus are not useful for
practical calculations

Bayliss Turkel (1980): Wave-Like Equations: In the

numerical computation of hyperbolic equations it is not practical
to use infinite domains. Instead, one truncates the domain with an
artificial boundary

Halpern (Wave Propagation 1982, Diffusion 1987):

. . . one often introduces artificial boundaries with boundary
conditions chosen so that the problem one gets is well-posed and
the solution is as close as possible to that of the original
problem

ABS and PML

Martin J. Gander
Domain
Truncation
History
Example

TBS and ABS

Construction
Application: ABCs

PML
Historical
Stretched-Coordinate
PML

Pole Condition
Meaning
Approximations

Mathematical
Equivalences
With PML
With ABC
Numerical
Experiments
Optimization of s0

Conclusions

Truncation of an Infinite Computational Domain

Airbus A340 in approach over a city

ABS and PML

Martin J. Gander
Domain
Truncation
History
Example

TBS and ABS

Construction
Application: ABCs

PML
Historical
Stretched-Coordinate
PML

Pole Condition
Meaning
Approximations

Mathematical
Equivalences
With PML
With ABC
Numerical
Experiments
Optimization of s0

Conclusions

Truncation of the unbounded domain using

A transparent (or exact) boundary condition (TBC) or
an absorbing (or inexact) boundary condition (ABC)
A perfectly matched layer (PML)

Interest for Domain Decomposition Methods

Computation performed with an optimized Schwarz method
on 16 subdomains

ABS and PML

Martin J. Gander
Domain
Truncation
History
Example

TBS and ABS

Construction
Application: ABCs

PML
Historical
Stretched-Coordinate
PML

Pole Condition
Meaning
Approximations

Mathematical
Equivalences
With PML
With ABC
Numerical
Experiments
Optimization of s0

Conclusions

An optimized Schwarz method with two-sided Robin

transmission conditions for the Helmholtz equation (G,
Halpern and Magoules, 2006)

An Advice from the Past

Emile
Picard (1893): Sur lapplication des methodes
dapproximations successives `
a letude de certaines equations
differentielles ordinaires
Les methodes dapproximation
dont nous faisons usage sont
theoriquement susceptibles de
sappliquer `
a toute equation,
mais elles ne deviennent vraiment interessantes pour letude
des proprietes des fonctions
definies par les equations
differentielles que si lon ne
reste pas dans les generalites
et si lon envisage certaines
classes dequations.

ABS and PML

Martin J. Gander
Domain
Truncation
History
Example

TBS and ABS

Construction
Application: ABCs

PML
Historical
Stretched-Coordinate
PML

Pole Condition
Meaning
Approximations

Mathematical
Equivalences
With PML
With ABC
Numerical
Experiments
Optimization of s0

Conclusions

Construction of Transparent Boundary Conditions

We consider the model problem
( )u = f
u(x, 0) = 0
u(x, ) = 0

in = R (0, )

ABS and PML

Martin J. Gander
Domain
Truncation
History
Example

TBS and ABS

with f compactly supported in int = (0, 1) (0, ), and u

bounded at infinity.
y

Construction
Application: ABCs

PML
Historical
Stretched-Coordinate
PML

Pole Condition

Meaning
Approximations

0
0

support
of f

Mathematical
Equivalences

1
1

In order to solve this problem on a computer, the

computational domain needs to be truncated in x, and an
artificial boundary condition needs to be imposed at x = 0
and x = 1.

With PML
With ABC
Numerical
Experiments
Optimization of s0

Conclusions

Construction of Transparent Boundary Conditions

+

Based on the decomposition of = int

y

ABS and PML

Martin J. Gander
Domain
Truncation
History
Example

TBS and ABS

int

Construction
Application: ABCs

PML

Pole Condition
Meaning
Approximations

and the equivalent coupled problems

( )v
v
n v
( )v
n v
v+
( )v +

=
=
=
=
=
=
=

0
v
n v
f
n v +
v
0

Historical
Stretched-Coordinate
PML

in
on 0
on 0
in int
on 1
on 1
in +

with homogeneous conditions at y = 0 and y = .

Mathematical
Equivalences
With PML
With ABC
Numerical
Experiments
Optimization of s0

Conclusions

Solution of Exterior Problems

The exterior problems are independent of the data f and can
be readily solved using Fourier series in y : for example on
+ we have
( + k 2 xx )
v+ = 0

an ordinary differential equation, whose solution is

2
2
v+ (x, k) = A(k)e +k x + B(k)e +k x

Since v + needs to stay bounded for x +, we have

A(k) = 0 and using the Dirichlet data v + = v at x = 1 we
obtain

2
v + (x, k) = v (1, k)e +k (x1)
This implies
x v+ (x, k)|x=1 =

+ k 2 v(x, k)|x=1

and similarly on the left boundary 0 , we obtain

p
x v (x, k)|x=0 = + k 2 v (x, k)|x=0

ABS and PML

Martin J. Gander
Domain
Truncation
History
Example

TBS and ABS

Construction
Application: ABCs

PML
Historical
Stretched-Coordinate
PML

Pole Condition
Meaning
Approximations

Mathematical
Equivalences
With PML
With ABC
Numerical
Experiments
Optimization of s0

Conclusions

Closing the Interior Problem

Using these solutions for the interior problem, we get
( )v
n v
n v

= f
in int
p
1
2
= F (p + k v ) on 0
= F 1 ( + k 2 v ) on 1

where the inverse tranform in general is a convolution:

Z
p
1
2
F ( + k v ) =
f (y )v (, )d

with
p
F(f (y )) = + k 2 .

By construction, v coincides with u in int , and the

conditions obtained are called transparent boundary
conditions (TBCs). They require a non-local convolution
boundary condition.

ABS and PML

Martin J. Gander
Domain
Truncation
History
Example

TBS and ABS

Construction
Application: ABCs

PML
Historical
Stretched-Coordinate
PML

Pole Condition
Meaning
Approximations

Mathematical
Equivalences
With PML
With ABC
Numerical
Experiments
Optimization of s0

Conclusions

Using Transparent Boundary Conditions

Because of their non-local nature, one often approximates
TBCs and obtains Absorbing Boundary Conditions (ABCs):

by polynomial or rational approximations of the symbol

(Enquist, Majda, Halpern, Bruneau-Di Menza, Nataf,
Japhet, Szeftel, Shibata . . . )

by approximation of the convolution kernel (Hairer,

Lubich, Schlichte, Greengard, Strain, Sch
adle . . . )

using quadrature rules (Mayfield, Baskakov and

Popov,. . . )

Remark: To couple an incoming field uin from + :

p
x (v uin ) = F 1 ( + k 2 (
v uin ))
on 1

which by linearity is equivalent to

p
p
x v F 1 ( + k 2 v ) = x uin F 1 ( + k 2 uin ) on 1

ABS and PML

Martin J. Gander
Domain
Truncation
History
Example

TBS and ABS

Construction
Application: ABCs

PML
Historical
Stretched-Coordinate
PML

Pole Condition
Meaning
Approximations

Mathematical
Equivalences
With PML
With ABC
Numerical
Experiments
Optimization of s0

Conclusions

ABS and PML

Perfectly Matched Layers

J. Berenger (1994): A perfectly matched layer for the
absorption of electromagnetic waves
Split-Field PML: Original idea, split the electromagnetic
fields into two unphysical fields in the PML region
y

Martin J. Gander
Domain
Truncation
History
Example

TBS and ABS

Construction
Application: ABCs

PML
Historical
Stretched-Coordinate
PML

Pole Condition

support
of f

Meaning
Approximations

Mathematical
Equivalences

S.D. Gedney (1996): An anisotropic perfectly matched

layer absorbing media for the truncation of FDTD latices
Uniaxial PML (UPML): The PML is described as an artificial
anisotropic absorbing material

With PML
With ABC
Numerical
Experiments
Optimization of s0

Conclusions

ABS and PML

Stretched-Coordinate PML
W.C. Chew and W. H. Weedon (1994): A 3d perfectly
matched medium from modified Maxwells equations with
stretched coordinates
F.L. Teixeira and W.C. Chew (1998): General
closed-form PML constitutive tensors to match arbitrary
bianisotropic and dispersive linear media
Idea: Analytic continuation of the solution and equation to
a complex x contour which changes oscillating waves into
exponentially decaying waves outside the region of interest:
e ikx e ik((x)+i (x)) = e ik(x) e k(x)
Then perform a coordinate transform to express the complex
x as a function of a real coordinate, which leads to complex
materials:

i
(x)
x
x
1+
k

Martin J. Gander
Domain
Truncation
History
Example

TBS and ABS

Construction
Application: ABCs

PML
Historical
Stretched-Coordinate
PML

Pole Condition
Meaning
Approximations

Mathematical
Equivalences
With PML
With ABC
Numerical
Experiments
Optimization of s0

Conclusions

ABS and PML

Example of a Coordinate Stretching

Martin J. Gander

0.8

Domain
Truncation

0.6

0.4

History
Example

0.5

0.2

TBS and ABS

Construction
Application: ABCs

0.2
0.5

0.4

PML
Historical
Stretched-Coordinate
PML

0.6
1

0.8

10

10

Meaning
Approximations

0.8

Mathematical
Equivalences

0.6

0.4

With PML
With ABC
Numerical
Experiments
Optimization of s0

0.5

0.2

Conclusions

0.2
0.5

0.4

0.6
1

0.8

Pole Condition

10

10

Alternative Formulation: the Pole Condition

Hohage, Schmidt and Zschiedrich (SIAM Math. Anal.
2003)
The pole condition is a general concept for the
theoretical analysis and the numerical solution of a
variety of wave propagation problems. It says that
the Laplace transform of the physical solution in
the radial direction has no poles in the lower
complex halfplane.

ABS and PML

Martin J. Gander
Domain
Truncation
History
Example

TBS and ABS

Construction
Application: ABCs

PML
Historical
Stretched-Coordinate
PML

Pole Condition
Meaning
Approximations

Mathematical
Equivalences

int

With PML
With ABC
Numerical
Experiments
Optimization of s0

Conclusions

The Pole Condition for our Model Problem

On + we had after a Fourier transform
2

History
Example

Applying a Laplace transform in the radial, i.e. x direction,

we obtain
2

( + k s )
v (s, k) n v (1, k) s v (1, k) = 0
Hence

v + (1, k)

s v + (1, k)

+
+ k2 s2
p
This solution has poles at s = + k 2 , and a partial
fraction expansion gives
v + (s, k) =

v + (1, k)
+

v (s, k) =

n v + (1,k)

2
+k
p
+
2(s + + k 2 )

Martin J. Gander
Domain
Truncation

( + k xx )
v =0

ABS and PML

v + (1, k)

n v + (1,k)

+ 2
+k
p
2(s + k 2 )

TBS and ABS

Construction
Application: ABCs

PML
Historical
Stretched-Coordinate
PML

Pole Condition
Meaning
Approximations

Mathematical
Equivalences
With PML
With ABC
Numerical
Experiments
Optimization of s0

Conclusions

ABS and PML

The Meaning of the Pole Condition

Martin J. Gander

Now the pole condition states that this solution

v + (1,k)

v + (1,k)

n
n
v + (1, k) +
v + (1, k)
+k 2
+k 2
p
p
+
v (s, k) =
2(s + + k 2 )
2(s + k 2 )

can not have any poles in the

p right half of the complex
plane, (s) 0, and since + k 2 > 0, the second term
can not be present, which means that
n v + (1, k)
v+ (1, k) + p
=0
+ k2

which is equivalent to

n v + (1, k) =

+ k 2 v+ (1, k)

as we found before chosing the bounded solution in the right

outer domain.

Domain
Truncation
History
Example

TBS and ABS

Construction
Application: ABCs

PML
Historical
Stretched-Coordinate
PML

Pole Condition
Meaning
Approximations

Mathematical
Equivalences
With PML
With ABC
Numerical
Experiments
Optimization of s0

Conclusions

ABS and PML

Approximations from the Pole Condition

Martin J. Gander

In order to satisfy the pole condition approximately, it is

natural to expand the desired analytic function in the right
half plane into a power series. To do so, we first map the
right half plane into the unit disk using the M
obius transform
s =

s s0
s + s0

s = s0

s + 1
s 1

History
Example

TBS and ABS

Construction
Application: ABCs

PML
Historical
Stretched-Coordinate
PML

Pole Condition

Domain
Truncation

Meaning
Approximations

Mathematical
Equivalences

s0
s

With PML
With ABC
Numerical
Experiments
Optimization of s0

Conclusions

and expand v + (s , k) into a power series about s = 0.

ABS and PML

An Educated Guess

Martin J. Gander

We use a particular Ansatz for the power series (Sch

adle
2008):
!

X
v + (1, k)
n
+
an s
v (s , k) = (s 1) (s 1)
2s0
n=0

v + (s , k)

because then
Laplace transforms

satisfies already the limit property of

lim s v + (s, k) = v + (1, k)

as one can se from (recall that s = s0 ss +1

1 )
s + 1 +
lim s0
v (s , k) = v + (1, k).

s 1
s 1

Domain
Truncation
History
Example

TBS and ABS

Construction
Application: ABCs

PML
Historical
Stretched-Coordinate
PML

Pole Condition
Meaning
Approximations

Mathematical
Equivalences
With PML
With ABC
Numerical
Experiments
Optimization of s0

Conclusions

Determination of the Coefficients

Inserting this power series Ansatz into the equation
( + k 2 s 2 )
v + (s, k) n v + (1, k) s v + (1, k) = 0
satisfied by v + (s , k) and comparing coefficients of powers of
s , we find
s 0 :

( + k 2 )a0 + ( + k 2 ) v

ABS and PML

Martin J. Gander
Domain
Truncation
History
Example

TBS and ABS

Construction
Application: ABCs

PML
Historical
Stretched-Coordinate
PML

(1,k)
s02 a0 + s20 v+ (1, k)=n v+ (1,k) Pole Condition
2s0
Meaning
(1,k)
(+k 2 )(a12a0)(+k 2 )v 2s
s02 (a1+2a0)+s20 v+(1,k)=0
Approximations
0
+

s 1 :
..
..
.
.
s i : ( + k 2 )(ai 2ai 1 + ai 2 ) s02 (ai + 2ai 1 + ai 2 )=0

for i = 2, . . .. Truncating this recurrence relation for ai at

some i = I , and setting ai = 0 for i I we obtain a linear
system for the unknowns ai , and an approximate relation
between v + (1, k) and n v + (1, k).

Mathematical
Equivalences

With PML
With ABC
Numerical
Experiments
Optimization of s0

Conclusions

ABS and PML

A Simple Example
For I = 0, all coefficients ai = 0, and only the relation from
the s 0 term remains:

s0
k2 +
n v + (1, k) + (
+ +
)
v (1, k) = 0
2s0
2
2s0
It remains to choose the expansion point s0 :
s
s0
p
2
+ kmin

Domain
Truncation
History
Example

TBS and ABS

Construction
Application: ABCs

PML
Historical
Stretched-Coordinate
PML

Pole Condition

s
p
2
+ kmax

The spectrum we try to approximate is at s = + k 2 ,

k {kmin , . . . , kmax }. If we choose s0 = , we get

k2

v + (1, k) = 0
n v + (1, k) + ( + )
2

a second order Taylor condition.

Martin J. Gander

Meaning
Approximations

Mathematical
Equivalences
With PML
With ABC
Numerical
Experiments
Optimization of s0

Conclusions

Formulation for the More General Case

Discretization of v + in x using finite differences:
1
( Dyy )vij+ 2 (vi++1,j 2vij+ + vi+1,j ) = fn,j
h
y

ABS and PML

Martin J. Gander
Domain
Truncation
History
Example

TBS and ABS

Construction
Application: ABCs

+
v13

+
a03 a13 a23
vn3

PML

+
v11

+
a01 a11 a21
vn1

Pole Condition

Historical
Stretched-Coordinate
PML

Meaning
Approximations

x
0

Representation of the recurrence relation ai on the same grid:

D

( Dyy s02 )a0j + ( s20 + 2s0yy )vnj+ + Dn vnj+ = 0

D
(Dyy )(a1j 2a0j )s02 (a1j +2a0j )+( s20 2s0yy )vnj+ = 0
..
.
(Dyy )(ai +1,j 2ai ,j +ai 1,j )s02 (ai +1,j +2ai ,j +ai 1,j ) = 0

Mathematical
Equivalences
With PML
With ABC
Numerical
Experiments
Optimization of s0

Conclusions

Equivalence With Stretched-Coordinate PML

A general finite difference discretization of the PML layer
equation with the coordinate stretching
x (x)x u ( Dyy )u = 0

is given by
uj+1 uj
uj uj1
j
j1
hj
hj1
( Dyy )uj = 0
hj

Result (G, Schadle 2010)

The Pole condition gives a recurrence relation which can be
identified with a geometrically stretched-coordinate PML.

ABS and PML

Martin J. Gander
Domain
Truncation
History
Example

TBS and ABS

Construction
Application: ABCs

PML
Historical
Stretched-Coordinate
PML

Pole Condition
Meaning
Approximations

Mathematical
Equivalences
With PML
With ABC
Numerical
Experiments
Optimization of s0

Conclusions

ABS and PML

Operator in the General Case

Need to understand the general recurrence relation
2

( + k )(ai 2ai 1 + ai 2 )

s02 (ai

+ 2ai 1 + ai 2 ) = 0

Domain
Truncation
History
Example

TBS and ABS

which is equivalent to

Construction
Application: ABCs

( + k 2 s02 )ai 1 2( + k 2 + s02 )ai + ( + k 2 s02 )ai +1 = 0

or with

Martin J. Gander

+ k 2 + s02
b :=
+ k 2 s02

we have the simple recursion

ai 1 2bai + ai +1 = 0
which will be truncated for some i = I ,
aI 1 2baI = 0.

PML
Historical
Stretched-Coordinate
PML

Pole Condition
Meaning
Approximations

Mathematical
Equivalences
With PML
With ABC
Numerical
Experiments
Optimization of s0

Conclusions

Equivalence with Absorbing Boundary Conditions

Result (G, Schadle 2010)
The recurrence relation is equivalent to the continued
fraction expansion
!
1
n v + (1, k)
1
+
2b 2
v

(1,
k)
=

.
1
4s0
+ k 2 s02
2b 2b...

ABS and PML

Martin J. Gander
Domain
Truncation
History
Example

TBS and ABS

Construction
Application: ABCs

PML
Historical
Stretched-Coordinate
PML

In the limit, independantly of s0 , we recover the DtN

operator:
p
+ k 2 v + (1, k) = n v + (1, k).

Pole Condition

Result (G, Schadle 2010)

Conclusions

The Pole condition truncated at step I gives an absorbing

boundary condition which is the (2I+2,2I) Pade
approximation at s0 of the symbol of the DtN operator.

Meaning
Approximations

Mathematical
Equivalences
With PML
With ABC
Numerical
Experiments
Optimization of s0

ABS and PML

Pole Condition as ABC

Martin J. Gander

= (0.5, 1) (0,
1) with truncation at x = 0.5,
2 2
uex = sin(ky )e k +(x1) , = 0, s0 = 1.5, I = 0

Domain
Truncation
History
Example

0.6

0.6

TBS and ABS

Construction
Application: ABCs

PML

0.8

0.8

0.4

0.4

0.2

0.2

0
1

Historical
Stretched-Coordinate
PML

0
1
0.8

0.8

0.6

0.6
0.8

0.4
0.6

0.2
0

0.4

0.6

0.2
0

0.4

x 10

Pole Condition
Meaning
Approximations

Mathematical
Equivalences
With PML
With ABC
Numerical
Experiments
Optimization of s0

20

15

10
error

0.8

0.4

Conclusions

5
1
0.8

1
0.6
0.8

0.4
0.6

0.2

0.4

ABS and PML

Error Level of the ABC

1

Martin J. Gander

10

10
error
order 2

error
order 2

10

error

error

Domain
Truncation

10

10

History
Example
3

10

TBS and ABS

Construction
Application: ABCs

10

10

PML
5

10

10

10

10

10

10

10

I = 0, s0 = 1.5

10

Meaning
Approximations

10
error
order 2

error
order 2

Mathematical
Equivalences

10

error

10

10

With PML
With ABC
Numerical
Experiments
Optimization of s0

10

Conclusions

10

10

10

10

10

I = 0, s0 = 2

10

10

Historical
Stretched-Coordinate
PML

Pole Condition

10

error

10

I = 2, s0 = 1.5

10

10

10

10

10

I = 2, s0 = 3

10

Varying s0 : s0 =

2
+ kmin

ABS and PML

Martin J. Gander
Domain
Truncation

History
Example

0.2

TBS and ABS

Construction
Application: ABCs

PML
Historical
Stretched-Coordinate
PML

error

4
3
2

Pole Condition
Meaning
Approximations

Mathematical
Equivalences

1
0
1

1
1
0.95

0.5

1
0.95

0.5

0.9

0.9

0.85
y

0.8

0.85
x

0.8

With PML
With ABC
Numerical
Experiments
Optimization of s0

Conclusions

Varying s0 : s0 =

+ k22

ABS and PML

Martin J. Gander
Domain
Truncation
History
Example

0.08

TBS and ABS

Construction
Application: ABCs

0.06

PML
Historical
Stretched-Coordinate
PML

4
error

0.04
3

Pole Condition
0.02

Meaning
Approximations

Mathematical
Equivalences

2
1
0

With PML
With ABC
Numerical
Experiments
Optimization of s0

0.02

1
1
0.95

0.5

1
0.95

0.5

0.9

0.9

0.85
y

0.8

0.85
x

0.8

Conclusions

Varying s0 : s0 =

+ k32

ABS and PML

Martin J. Gander
Domain
Truncation
History
Example

0.03

TBS and ABS

Construction
Application: ABCs

0.025

PML

0.02

Historical
Stretched-Coordinate
PML

0.015
error

4
3

0.01

Pole Condition
Meaning
Approximations

0.005

Mathematical
Equivalences

0
1

0.005

With PML
With ABC
Numerical
Experiments
Optimization of s0

0.01

1
1
0.95

0.5

1
0.95

0.5

0.9

0.9

0.85
y

0.8

0.85
x

0.8

Conclusions

Varying s0 : s0 =

+ k42

ABS and PML

Martin J. Gander
Domain
Truncation
History
Example

x 10
6

14

TBS and ABS

Construction
Application: ABCs

12
5
10
4

PML
Historical
Stretched-Coordinate
PML

8
error

6
3
2

Pole Condition

Meaning
Approximations

Mathematical
Equivalences

0
1

With PML
With ABC
Numerical
Experiments
Optimization of s0

2
0

1
1
0.95

0.5

1
0.95

0.5

0.9

0.9

0.85
y

0.8

0.85
x

0.8

Conclusions

Varying s0 : s0 =

+ k52

ABS and PML

Martin J. Gander
Domain
Truncation
History
Example

x 10
6

20

TBS and ABS

Construction
Application: ABCs

15

PML
Historical
Stretched-Coordinate
PML

4
error

10
3

Pole Condition
5

Meaning
Approximations

Mathematical
Equivalences

2
1
0

With PML
With ABC
Numerical
Experiments
Optimization of s0

1
1
0.95

0.5

1
0.95

0.5

0.9

0.9

0.85
y

0.8

0.85
x

0.8

Conclusions

Varying s0 : s0 =

+ k62

ABS and PML

Martin J. Gander
Domain
Truncation
History
Example

0.04

TBS and ABS

Construction
Application: ABCs

0.035
5
0.03
4

PML
Historical
Stretched-Coordinate
PML

0.025
error

0.02
3
2

Pole Condition

0.015

Meaning
Approximations

0.01

Mathematical
Equivalences

0.005
1

With PML
With ABC
Numerical
Experiments
Optimization of s0

0
0

0.005

1
1
0.95

0.5

1
0.95

0.5

0.9

0.9

0.85
y

0.8

0.85
x

0.8

Conclusions

Varying s0 : s0 =

+ k72

ABS and PML

Martin J. Gander
Domain
Truncation
History
Example

0.06

0.05

TBS and ABS

Construction
Application: ABCs

PML

0.04

Historical
Stretched-Coordinate
PML

4
error

0.03
3
2

Pole Condition
0.02

Meaning
Approximations

0.01

0.01

Mathematical
Equivalences
With PML
With ABC
Numerical
Experiments
Optimization of s0

1
1
0.95

0.5

1
0.95

0.5

0.9

0.9

0.85
y

0.8

0.85
x

0.8

Conclusions

Varying s0 : s0 =

+ k82

ABS and PML

Martin J. Gander
Domain
Truncation
History
Example

0.08

TBS and ABS

Construction
Application: ABCs

0.07
5
0.06
4

PML
Historical
Stretched-Coordinate
PML

0.05
error

0.04
3
2

Pole Condition

0.03

Meaning
Approximations

0.02

Mathematical
Equivalences

0.01
1

With PML
With ABC
Numerical
Experiments
Optimization of s0

0
0

0.01

1
1
0.95

0.5

1
0.95

0.5

0.9

0.9

0.85
y

0.8

0.85
x

0.8

Conclusions

Varying s0 : s0 =

+ k92

ABS and PML

Martin J. Gander
Domain
Truncation
History
Example

0.12

0.1

TBS and ABS

Construction
Application: ABCs

PML

0.08

Historical
Stretched-Coordinate
PML

4
error

0.06
3
2

Pole Condition
0.04

Meaning
Approximations

0.02

0.02

Mathematical
Equivalences
With PML
With ABC
Numerical
Experiments
Optimization of s0

1
1
0.95

0.5

1
0.95

0.5

0.9

0.9

0.85
y

0.8

0.85
x

0.8

Conclusions

ABS and PML

Varying s0 : Error Levels Attained

Martin J. Gander

10

Domain
Truncation
History
Example

TBS and ABS

Construction
Application: ABCs

PML

error

Historical
Stretched-Coordinate
PML

Pole Condition
1

10

Meaning
Approximations

Mathematical
Equivalences
With PML
With ABC
Numerical
Experiments
Optimization of s0

Conclusions

10

km

10

12

14

16

18

ABS and PML

Optimization of s0

Martin J. Gander
Domain
Truncation

Result (G, Schadle 2010)

P (k,s )

0
be the (2j + 2, 2j) Pade approximation at s0 of
Let Qjj (k,s0)
the symbol of the DtN operator.
Then the reflection coefficient of the jth pole condition
approximation,
p
P (k,s )
+ k 2 Qjj (k,s00 )
,
(k, s0 ) = p
P (k,s )
+ k 2 + Qjj (k,s00 )

is minimized for k [0, kmax ], kmax large, if the expansion

point is chosen to be

History
Example

TBS and ABS

Construction
Application: ABCs

PML
Historical
Stretched-Coordinate
PML

Pole Condition
Meaning
Approximations

Mathematical
Equivalences
With PML
With ABC
Numerical
Experiments
Optimization of s0

Conclusions

s02

1/2

1/2
2
kmax

ABS and PML

Conclusions

Martin J. Gander

The Pole Condition leads to Pade approximations of

the symbol of the DtN operator, and hence gives an
Absorbing Boundary Condition
The Pole Condition also leads to a geometrically
stretched-coordinate Perfectly Matched Layer
Complete error estimates for the Helmholtz case,
including optimal choice of s0 for any order.
An improved Pole Condition would use collocation in
order to obtain best approximation. For example:

2
2
s0 =( + kmin
)1/4 ( + kmax
)1/4
1
2
2
s1 = 2 ( + kmin
)5/12 ( + kmax
)1/12
2
2
s2 =2( + kmin
)1/12 ( + kmax
)5/12

History
Example

TBS and ABS

Construction
Application: ABCs

PML
Historical
Stretched-Coordinate
PML

Pole Condition
Meaning
Approximations

Current efforts:

Domain
Truncation

R =14

2
( + kmin
)1/12
2
( + kmax )1/12

Use of the Pole Condition in Domain Decomposition

Mathematical
Equivalences
With PML
With ABC
Numerical
Experiments
Optimization of s0

Conclusions