On the far-field properties

of an acoustic horn

Eddie Wadbro

Department of Information Technology Technical report 2006-042

Uppsala University September 2006
Box 337, SE-751 05 Uppsala, Sweden ISSN 1404-3203
 1

Abstract

This report presents a derivation of an expression for time harmonic acoustic

wave propagation in the far field for two and three space dimensions, and includes
detailed descriptions of the numerical evaluation of the far-field pattern in some typ-
ical situations. The presentation covers all parts required for computing the far-field
properties of acoustical devices, and the report is designed to function as a single
reference for these computations.

1 Introduction
When designing a horn loudspeaker it is desired to have high transmission
efficiency as well as control of the far-field properties (the directivity) of the
horn. The efficiency of an acoustical device can in some cases, such as when
the device is connected to a waveguide, be computed in a straightforward way
using local measurements. The far field is computed either by solving the wave
propagation problem in an unbounded domain or by evaluating an integral of
the sound pressure field along a closed surface enclosing the horn.
Numerical design optimization of various acoustical devices, with respect
to efficiency and to some extent ’near-field’ directivity (using local measure-
ments for a region or point a finite distance from the device), has been
performed in the last years by among others Bängtson et al. [1], who use
boundary-shape optimization to improve the transmission efficiency of the
horn. Various results of applying topology optimization for the Helmholtz
equations are presented by Sigmund and Jensen [5, 7]. In the manuscript
“Topology optimization of an acoustic horn” [9] the author of this report
and Berggren present a comprehensive study of the transmission properties
of horns by topology optimization. A continuation of this study presents
acoustical horns optimized both for efficiency and far-field directivity [10].
Recently, Udawalpola and Berggren [8] present horns optimized with respect
to efficiency and directivity obtained using boundary-shape optimization.
The computation of the far-field relies on classical methods from scattering
theory. All parts required for the numerical evaluation of the far-field prop-
erties are well known, but scattered throughout the literature. To the best of
my knowledge, there is no single work in the literature that contains all parts.
This report is designed to function as a single reference for far-field compu-
tations of acoustical devices, and aims to describe the mathematical theory
behind the far-field properties and the numerical evaluation of the far-field
pattern in two and three space dimensions. The report is organized as fol-
lows. Section 2 introduces the problem. Section 3 contains the mathematical
derivation of the far-field pattern, and Section 4 describes how the far-field
pattern is computed numerically.
 3. DERIVATION OF AN EXPRESSION FOR THE FAR-FIELD
2 PATTERN

2 Problem Description
The wave propagation is assumed to be governed by the wave equation

@2P
= c2 ∆P;
@t2
where P is the acoustical pressure and c is the speed of sound. Seeking time
harmonic solutions, for a single frequency ! and the complex amplitude func-
tion p, making use of the ansatz P = < p(x)ei!t , results in the following
Helmholtz equation:
c2 ∆p + ! 2 p = 0: (1)
Further, it is required that p satisfies the Sommerfeld radiation condition,
stipulating that the waves are outgoing in the far field,
 
x
lim
jxj!+1
jxj (d 1)=2
jxj
rp + ikp = 0; (2)

uniformly for all directions, where d is the number of space dimensions con-
sidered.
In the far field, the complex amplitude function is essentially the product
of a function of the distance to the origin and a function of the direction.
More precisely, let  > 0, and let xb( ) be a point on the unit sphere, where
 represents the polar argument(s) of x b. Then, as we shall see below, for
x0 = xb,   
e ik
 ! +1;
1
p(x0 ) = (d 1)=2 p1 () + O
 
where k = !=c is the wave number. The function p1 () is denoted the far-field
pattern of the horn and is the subject of this report.

3 Derivation of an expression for the far-field

pattern
Assume that the horn is located inside a bounded domain Ω b with sufficiently
b
smooth boundary Γout , and let V be the domain exterior to Ω.b The exterior
b consists of one or more
domain V is assumed to be connected; in other words, Ω
simply connected components. The setup is illustrated in Figure 1. Then, for
any point x0 2 V , the complex amplitude function p can be evaluated at x0
using the following theorem.

Theorem 1. If p 2 C 2 (V ) satisfies (1) in V together with radiation condi-

tion (2), then for any point x0 2 V
Z  
@g 0 0 @p
p(x0 ) = p d gd dHd 1
(x); (3)
Γ
b out @n @n
 3

b
Ω

b
Γ o ut

b
Figure 1: The acoustical horn is located inside the domain Ω.

where n is the outward directed normal of Γ b out , d is the number of space

dimensions, and gd0 is the fundamental solution to the Helmholtz equation in
d dimensions with source located at x0 .

Proof. A proof of this theorem (d = 2; 3) is found in Appendix A.

The fundamental solution of the Helmholtz equation in two dimensions

with source located at x0 is given by

g20 (x) = H0 ( k jx x0 j);

i (1)
(4)
4
(1)
where H0 is the Hankel function of the first kind of zero order. The Hankel
function of the first kind with order m has the following asymptotic behavior
(Appendix B):
r   
Hm (  ) = ( 1)
(1) m+1 2

e i( m=2 =4)
1+O
1

 ! +1: (5)

In three dimensions, the fundamental solution (with source located at x0 ) to

Helmholtz equation is given by

e ikjx x0 j
g30 (x) = :
4 jx x0 j
(6)
 3. DERIVATION OF AN EXPRESSION FOR THE FAR-FIELD
4 PATTERN

The far-field pattern

Assume that x0 2 V , define  = jx0 j and let xb be the unit vector such that
x0 = x b. By letting  ! 1 it is now possible to derive an expression for the
far-field pattern. For x 2 Γ
b out , let

r
r = x0 x and rb =
jr j : (7)

We have that
   
 x 2 x
x
jr j 2
= jx0 xj 2
=  xb2

b
x
= 2 x b
2

b
x+ 2
x

 
= 2
1
2
x
jxj2
b
x+ 2 ;
 

hence, by Taylor expansion,

s
jrj = jx0 xj =  1
2 jxj2
b
x+ 2
x
 
   2 !!
= 1+
1 2
x
jxj2
b
x+ 2 +O
2
x
jxj2
b
x+ 2 (8)
2    
    
b
x
x
b
x+O  ! 1:
1 1
= 1 +O = x
 2 

Moreover,
 
xb x x
b x=
 ! 1:
1
rb = =x
b+O
jxb xj jx
=
b x=j 
(9)

Three space dimensions

Rewrite (3), making use of the fundamental solution (6) and relations (7)
and (9), as
Z  
@g 0 @p
p(x
b) = p(x0 ) = p 3 g30 dH2 (x)
Γ @n @n
"  #
b out
Z 
e ikjrj e ikjrj @p
( rb
n) dH2 (x)
1
p ik
=
Γ
b out jr j 4 j r j 4 jrj @n
Z        
e ikjrj @p
=
4 j r j
p ik + O
1

x
b+O
1


n @n
dH2 (x):
Γ
b out
 5

Previous expression is expanded and simplified using (8) as follows:

Z
    
“ ”
b x+O 1 )
ik (  x
@p
b
n 1+O dH2 (x)
e 1 1
p(x
b) =   pik x
Γ
b out  b
x + O 1 4
x 
 @n

Z
   
“ ”
ik b x+O 1
ik x
@p
b
n+O dH2 (x)
e e 1 1
=   pik x
 Γ
b out 1 + O 1 4  @n

ik Z      
@p
e
+O b
n dH2 (x):
e 1 ik x
bx 1 1
= ipk x +O
 4 Γ
b out  @n 
Thus   
e ik 1
p(x
b) = p1 () + O ;
 
where  are the polar arguments of x
b and the far-field pattern is given by
Z  
@p
e
ikpx
b
n dH2 (x):
1 ik x
bx
p1 () =
4 Γ
b out @n

Two space dimensions

Inserting the fundamental solution (4) of the two dimensional Helmholtz equa-
tion and relation (7) into (3) yields
Z  
@g 0 @p
p(x
b) = p(x0 ) = p 2 g20 dH1 (x)
Γ @n @n
Z  
b out
(10)
ik (1) @p
p H1 ( k jrj)( rb
n) H0 ( k jrj) dH1 (x):
i (1)
=
Γ
b out 4 4 @n
Making use of the asymptotics for the Hankel functions given in (5), rela-
tion (9), and that jrj = O(), from (8), it is possible expand the integrand
in (10)
ik (1) @p
H ( k jrj)rb
n H ( k jr j)
i (1)
p
4 1" s 4 0 @n
  !
1
pk ( 1)2
2 jj
i(k r =2 =4)
1+O
1
k jrj k jr j
= e
4i
  
x
b+O
1


n (11)
s   ! #
2 i(k r jj =4) 1 @p
( 1)1 1+O
k jrj k jr j
+ e
@n
r       
ei=4 2 e ikjrj @p
b
n+O
1 1
= p 1+O ipk x :
4i k jrj   @n
6 4. NUMERICAL APPROACH

The first part of previous expression is rewritten using (8) as

r r

jj
“ ”
b x+O 1 )
ik (  x
ei=4 2 e ik r
e i=4 2 e
p = r
k jrj k  
b
x+O
4i 4
 x 1


p i=4 ik ik x

“ ”
b x+O 1 )   
(12)
1 i e ik
p ep r p p eikxb
x + O
2e e 1
=   = :
4 k 4 k  
1+O 
1

Inserting (12) into (11) yields

ik (1) @p
H ( k jrj)rb
n H ( k jr j)
i (1)
p
4 1

4 0

@n  
e ik 1 i ikxb
x @p
= p p
 4 k
e b
n
ipk x
@n
+O
1

:

Thus expression (10) can be written as

  
eik
b) = p
1
p(x p1 () + O ;
 

where  is the polar argument of x

b, and the far-field pattern is given by
Z  
@p
p1 () = p e
b
n dH1 (x):
1 i ik x
bx
ipk x
4 k Γ
b out @n
Hence, in both two and three space dimensions, the complex amplitude
function p has the asymptotic behavior
ik
  
 ! 1;
e 1
p(x0 ) = p(x
b( )) = p1 () + O
(d 1)=2 
where Z  
@p
p1 () = Cd eikxb
x ipk x
b
n dHd 1
(x); (13)
Γ
b out @n
and where Cd is a constant given by the number of space dimensions. For
d = 2 and d = 3, Cd is given by

C2 = p
1 i
and C3 =
1
4
:
4 k

4 Numerical approach
The numerical computation of the far-field pattern is first described in the case
where no symmetry properties are taken into account in the computations.
 7

Γo ut

Γh
Γs ym

Figure 2: The integration boundary Γout , the symmetry boundary Γsym , and
the boundary of the horn, Γh .

Then planar and cylindrical symmetries are addressed. First, recall that the
far-field pattern is given by
Z  
@p
p1 () = Cd e

ik x
bx
b
n
ipk x dHd 1
(x); (14)
Γ
b out @n

where Cd is a constant and Γ b out is the outer boundary of the domain Ω b con-
taining the horn.
Let Ω be a simply connected (computational) domain, with boundary Γ,
and assume that the solution to the wave propagation problem is known in Ω.
In the case where no symmetry assumptions are made and the horn is located
in Ω, its boundary Γ corresponds to the the evaluation boundary Γ b out in (14).
The numerical computation of the far-field pattern can be performed by first
computing the normal derivative @p=@n along the integration boundary (Γ);
see Section 4 for descriptions on how to compute this normal derivative nu-
merically. Once the normal derivative is computed, the far-field pattern can
be approximated by applying quadrature to numerically evaluate the integral.
Using planar symmetry, the boundary Γ, depicted in Figure 2, consists of
the parts Γsym , the symmetry boundary; Γout , the outer part of the boundary;
and—if the horn has been modeled explicitely—Γh , the boundary of the horn.
The evaluation boundary Γ b out consists of the union of Γout and its symmetric
1
twin, here denoted Γout . The normal derivative along Γout can be computed
according to the descriptions in Section 4. The expression for far-field pattern
using planar symmetry is
Z  
@p
p1 () = p eikxb
x ipk xb
n dH1 (x)
1 i
4 k Γbout @n
Z     
b
x @p b
x @p
= p b
n b
n dH1 (x);
1 i ik x ik x
e ipk x +e ipk x
4 k Γout @n @n

1
The over-line in Γout should here be viewed as an analogue to the complex conjugate
and not the closure of a set.
8 4. NUMERICAL APPROACH

where x and n are the mirror images of x and n on the other side of the
symmetry boundary. Note that, by symmetry
@p @p
(x) = (x):
@n @n
Having computed the normal derivative along Γout the remaining task is to
numerically evaluate the integral above.
For cylindrical symmetry it can, without loss of generality, be assumed that
the symmetry axis is aligned with the first coordinate axis. The boundary Γ
is illustrated in Figure 2 and partitioned as in the planar symmetric case
above. Here the integration surface Γ b out consists of all points x such that
there exist a point y = (y1 ; y2 ) 2 Γout and an angle ' 2 [0; 2 ] such that
x = (y1 ; y2 cos '; y2 sin '). The far-field behavior is symmetric with respect to
the symmetry axis of the horn. Thus it is sufficient to compute the far field
in the plane x3 = 0. Considering x b of the form x b = (xb1 ; x
b2 ; 0), the far-field
pattern is given by
Z  
@p
e
b
n(x) dH2 (x)
1 ik x
bx
p1 () = ipk x (x)
4 Γ
b out @n(x)
Z Z 2
1
= y2 eik(bx1 y1 +bx2 y2 cos ')
4 Γout 0
 
@p
ipk (x
b1 n1 (y ) + x
b2 n2 (y ) cos ') (y ) d' dH1 (y );
@n(y )
(15)
where n1 and n2 are components of the normal on Γ. Rearranging the terms
in (15) yields
Z " Z 2
1 ik x
b1 y1 @p
p1 () = y2 e ipk x
b1 n1 (y ) eikxb2 y2 cos ' d'
4 Γout @n 0
Z #
2
+ ipk x
b2 n2 (y ) e ik x
b2 y2 cos '
cos ' d' dH1 (y ):
0
(16)
The inner integrals in (16) can be expressed as Bessel functions. The Bessel
functions Jm of the first kind with order m are given by [6, p.63]
Z 
1
Jm (x) = eix sin ' e im'
d':
2 

Applying the substitution =2 t = ' and using that the integrand is 2 -
periodic result in
2 Z
1
Jm (x) = eix sin(=2 t) e im(=2 t) dt
2 0
Z 2 Z 2 (17)
1 1
= eix cos t e im=2 eimt dt = eix cos t eimt dt:
2 0 2 im 0
 9

Expression (17) can also be written as

Z 2 Z 2 
1 ix cos t ix cos t
Jm (x) = e cos mt dt + i e sin mt dt : (18)
2 im 0 0

When the argument x is real, the expression can be simplified by noticing

that, since the integrand is odd with respect to the center of the integration
interval, the last integral in (18) is zero; therefore
Z 2
m
2 i Jm (x) = eix cos t cos mt dt: (19)
0

Inserting expression (19) into the far-field expression for cylindrical symme-
try (16) results in
Z " 
1 ik x
b1 y1 @p
p1 () = y2 e ipk x
b1 n1 (y ) 2J0 (k x
b2 y2 )+
4 Γout @n
#
b2 y2 ) dH1 (y ):
b2 n2 (y )2 iJ1 (k x
ipk x

Again, the normal derivative of p along Γout can be computed using the strate-
gies in Section 4, and the integral can be evaluated numerically using quadra-
ture.
Remark 1. If the wave propagation problem has been solved using the finite
element method, then the integral
Z
f (x) dHd 1
(x);
Γ

can be
P
approximated by expanding f on Γ in the finite element basis, f 
fh = n fn 'n . The integral above is then computed as the sum
Z Z X Z
f (x) dHd 1
(x)  fh (x) dHd 1
(x) = fn 'n (x) dHd 1
(x):
Γ Γ n Γ

This method is usually fast, since the element mass matrix for Γ has been com-
puted during the initial assembly when solving the wave propagation problem.

The normal derivative @p=@n

The numerical evaluation of the far-field pattern according to (14) requires the
normal derivative @p=@n along the integration boundary Γ b out . In this section
the computation of this normal derivative is described in two cases.
 10 4. NUMERICAL APPROACH

Case 1. Assume that there is a direct known relationship between p and @p=@n
on Γb out . This is the case when, for example, the outgoing wave prop-
erty is imposed by a Robin boundary condition, such as the first-order
Engquist–Majda boundary condition [3]. Thus, after solving the wave
propagation problem for p in Ω,b the normal derivative @p=@n is acquired
by employing the relationship between p and @p=@n on Γ b out .

Now assume that the wave propagation problem is solved in the planar or
b is a cylinder (planar
cylindrical symmetric case, and that the domain Ω
symmetric case) or a ball (cylindrical symmetric case) with radius R,
and that the first-order Engquist–Majda boundary condition is used to
impose the outgoing wave property. That is,
 c  @p
i! + p+c b out ;
= 0 on Γ (20)
2R @n
for the planar symmetric case, and
 c @p
i! + p+c b out ;
= 0 on Γ (21)
R @n
for cylindrical symmetry. Then the far-field pattern in the planar sym-
metric case is given by combining (13) and (20),
Z  
b
x
ik x
= p e
p(x) dH1 (x):
1 i ik x
bx 1
p1 + ik + (22)
4 k Γ
b out R 2R

The corresponding expression in the cylindrical symmetric case is given

by combining (13) and (21),
Z  
b
x
ik x
e
p(x) dH2 (x):
1 ik x
bx 1
p1 = + ik + (23)
4 Γ
b out R R

Case 2. There is no direct known relationship between p and @p=@n on Γb out , so

we need to compute a numerical approximation to @p=@n on Γe out , given
e
an approximation to p in Ω.
Assume that p 2 H 2 (Ω) satisfies (1) with radiation condition (2). Fur-
ther, assume that @p=@n = g on a part Γp of Γ, note that Γp might be
empty (Γp = ;). The normal derivative @p=@n on Γ Γp satisfies
Z Z Z Z
@p
Ψ
@n
= rΨ
rp k 2
Ψp Ψg; 8 Ψ 2 H 1(Ω): (24)
Γ Γp Ω Ω Γp

Let ph 2 Vh (Ωh )  H 1 (Ωh ), where Ωh is an approximation of Ω, be a

finite element solution to the discretized version of (1) with radiation
 11

condition (2). Let Γh and Γph be the approximations of the boundaries

Γ and Γp respectively. Define the spaces
n  o
Vh0 = vh 2 Vh ; vhΓh Γph = 0 ;
n o
Vh = fh 2 L2 (Γh Γph ) ; 9 vh 2 Vh : fh = vh on Γh Γph ;

and let Mh be the space such that

Vh = Vh0  Mh;
that is, the space Mh  Vh is defined such that each element in Vh can
be written uniquely as a sum of one element from Vh0 and one element
from Mh .
By construction, the spaces Mh and Vh have the same dimension, in
fact, the restriction on Γh Γph of functions in Mh spans Vh . Let gh
correspond to the discretized version of g on Γph . Choosing Ψ 2 Mh
in (24), we may define h 2 Vh , the approximation of the normal
derivative @p=@n at Γh Γph , as the solution to the variational problem

Find h
Z
2 Vh such
Z
that:

p
Ψh h = rΨh
rph
Γh Γh Ωh (25)
Z Z
k2 Ψh ph Ψh gh ; 8 Φ h 2 Mh :
Ωh Γph

The variational problem (25) is equivalent to a linear system whose ma-

trix is square, symmetric, and sparse. Computing the approximation
of the derivative using this variational approach produces a much more
accurate approximation than a direct approximation of element deriva-
tives [4, p. 398].

Acknowledgments
I thank my scientific advisor Martin Berggren for the many helpful discussions
and for carefully reading through the manuscript and giving constructive com-
ments.

A Proof of Theorem 1
The proof of Theorem 1 given here covers the two and three dimensional cases
in detail. A similar proof in the three dimensional case is given by Colton and
Kress [2]. The differences in signs between the proofs are due to the choice
12 APPENDIX A. PROOF OF THEOREM 1

V Rε SR
Sε

b
Ω
b
Γ o ut

b out [ SR [ S" ; and Ω,

Figure 3: The regions VR" , with boundary Γ b with boundary
b
Γout .



of sign  in the time harmonic ansatz P = < p(x)ei!t . The proof of the
theorem is divided into three lemmas.
b that is V = Rd Ω,
Recall that V is the domain exterior to Ω, b where the
overbar denotes closure, and that Theorem 1 is stated as
Theorem 1. If p 2 C 2 (V ) satisfies (1) in V together with radiation condi-
tion (2), then for any point x0 2 V
Z  
@g 0 0 @p
p(x0 ) = p d gd dHd 1
(x);
Γ
b out @n @n

where n is the outward directed normal of Γ b out , d is the number of space

dimensions, and gd is the fundamental solution to the Helmholtz equation in
0

d dimensions with source located at x0 .

For R > jx0 j, R > maxx2Ωb jxj, 0 < " < R jx0 j, and " < minx2Ωb jx x0 j,
let BR = B (0; R) be the open ball of radius R centered at 0 with boundary
SR . Moreover, let B" = B (x0 ; ") be the open ball of radius " centered at x0
with boundary S" . Define VR" as the region (Figure 3)

VR" = (V \ BR ) B";

where the overbar denotes closure.

Under the assumption that p and gd0 are sufficiently smooth in VR" , the
following version of Green’s formula holds:
Z Z  
0 @p @g 0
(gd ∆p
0
p∆gd ) dx =
0
gd p d dHd 1
(x); (26)
VR" @VR" @n
e @n
e
 13

where n e is the outward directed normal of VR" . Recall that n is the outward
directed normal of Ω,b hence ne = n on Γ b out . Making use of the fact that p
and gd satisfy the Helmholtz equation (1), the left hand side of the previous
0

relation reads
Z Z
(gd ∆p
0
p∆gd ) dx =
0
(gd0 ( k 2 p) p( k 2 gd0 )) dx = 0;
VR" VR"

hence (26) simplifies to

Z   Z  
0 @p @g 0 0 @p @g 0
gd p d dH d 1
(x) + gd p d dHd 1
(x)
Γ
b out @n
e @n
e SR @n
e @n
e
Z  
@p @gd0
+ gd0 p dHd 1
(x) = 0:
S" @n
e @n
e
(27)

Lemma 1. Z
jpj2 = O(1) R ! 1:
SR

Proof. A direct calculation shows that

Z 
Z 2 Z  
 @p   @p 2
jk j 2
jpj dH
2 d 1
(x) = 
 e + ikp  d
 H d 1
(x)  
 e d Hd 1
(x)
SR SR @ n SR @ n
Z 
@p
2= kp dHd 1
(x) :
SR @ n
e

The Sommerfeld condition implies that the first term tends to zero as R tends
to infinity, and the second term is negative. Thus given " > 0, there exists
R" 2 R such that the magnitude of the first term is less than " whenever
R > R" . Hence, by using integration by parts, equation (1), the fact that
n
e = n, and stipulating that R > R1 , the following holds:
Z Z 
@p
jk j 2
jpj dH
2 d 1
(x)  1 0 2= kp
@ n
e
dHd 1
(x)
SR SR
( Z Z )
@p

=1 2= kp dHd 1
(x) k jkpj2 k jrpj 2
dx
Γout
b @ n
e VR 0
(Z Z )
@p

=1 2= kp dHd 1
(x) k jkpj 2
k jrpj 2
dx
Γout
b @n VR0
Z  Z
@p

=1 2= kp dHd 1
(x) + 2=fk g jkpj2 + jrpj2 dx:
Γ
b out @n VR0
14 APPENDIX A. PROOF OF THEOREM 1

Recall that the wave number k , given by !=c, is real,2 hence

Z Z 
@p
0  jk j 2
jpj dH
2 d 1
(x)  1 2= kp dHd 1
(x) = const:
SR Γ
b out @n

The expression above holds for any R > R1 , thus

Z
jpj2 dHd 1
(x) = O(1) R ! 1:
SR

Lemma 2. For any point x0 2V

Z  
0 @p @g 0
lim gd p d dHd 1
(x) = 0:
R !1 SR @n
e @n
e

Proof.

Z  
@p @gd0
gd0 p dHd 1
(x)
SR @n
e @n
e
Z  
@p @g 0
= ikgd0 p p d + ikpgd0 dHd 1 (x)
gd0
SR @n
e @n
e
Z  0  Z  
@gd @p
= p + ikgd0 dHd 1 (x) + gd0 + ikp dHd 1
(x)
SR @ne SR @n
e
= I1 (R) + I2 (R):

@p
The fact that p satisfies the Sommerfeld radiation condition (2), that is @n
+
ikp = o(R 2 ), and gd0 = O(R 2 ) directly implies that I2 (R) ! 0 as R ! 1.
d d
e
1 1

Notice that the plus sign in the Sommerfeld condition is due to the choice
ei!t in the time harmonic ansatz.
Further, for x 2 SR the following holds:

r
2x0
x
e jx0j2
jx0 xj = R 1
R
+ 2 = O(R) R ! 1;
R

2
For the general Helmholtz equation ∆p + k2 p = 0, where k2 2 C, choose k with
=fkg  0.
 15

where x
e = x=R and hence

r
@ @ 2x0
x
e jx0j2
@R
jx0 xj =
@R
R 1
R
+ 2
R
r 2x0
x j j2
2x0
x jx0j2 2x
e
e 0

+ 2 +R q R R3
2
= 1
R R
2 1 2xR0
xe + jx0 j2
R2

2
+ 2jx02j + 2x0
xe

4 x0 x
e 2 j j
2 x0 2
1

x0 x
e
R R
= q R R2
=q R
2x0
x
+ jx02j
+ jx02j2
2
2 x0 x
2 1 e
R R
1 R
e
R
   
x0
x
e x0
x
e 1
= 1 1+ +O
R R R2
 
R ! 1:
1
=1+O
R2

In the two dimensional case,

 
@g20 @
H0 ( k jx x0 j) + ik H0 ( k jx x0 j)
i (1) i (1)
+ ikg20 =
@n
e @R 4 4
 
@
H ( k jx x0 j) f kjx x0jg + ik 4i H0(1)( kjx x0j)
i (1)
=
4 1 @R
    
ik (1) 1 (1)
= H1 ( kd0 (x)) 1 + O + iH0 ( kd0 (x)) ;
4 R2

where d0 (x) = jx x0 j. Recall the asymptotics for the Hankel function given
in (5),

r   
Hm (  ) = ( 1)
(1) m+1 2

e i( m=2 =4)
1+O
1

 ! +1:
16 APPENDIX A. PROOF OF THEOREM 1

Hence
    
@g20 ik (1) 1 (1)
+ ikg20 = H1 ( kd0 (x)) 1 + O + iH0 ( kd0 (x))
@n
e 4 R2
"s      
2 i(kd0 (x) =2 =4) 1 1
= e 1+O 1+O
kd0 (x) R2 kd0 (x)
s    #
2 i(kd0 (x) =4) 1
+ i( 1) e 1+O
kd0 (x) kd0 (x)
"      
ik i=2 1 1
= e 1+O 1+O
4 R2 R
   # s
1 i(kd0 (x) =4) 2
i 1+O e
R kd0 (x)
"       #  
ik
= i 1+O
1
R
i 1+O
1
R
O p1
4 R
     
= O(1)O
1
R
O p1 =O
1
R3=2
:
R
(28)
In the three dimensional case,
!
@g30 @ e ikjx x0 j e ikjx x0 j
+ ikg30 = + ik
@n
e @R 4 jx x0 j 4 jx x0 j
  
e ikjx x0 j @ e ikjx x0 j
=
4 jx x0 j
ik
jx
1
x0 j @R
jx x0 j + ik
4 jx x0 j
j j    
e i k x x0 1 1 e ikjx x0 j (29)
ik 1+O + ik
4 j x x0 j jx x0 j 4 jx x0 j
=
R2
  
e ikjx x0 j 1 1
1+O
4 jx x0 j jx x0 j
=
R2
     
1 1 1
=O O O(1) = O :
R R R2
Using Lemma 1 together with expressions (28) and (29) results in
Z  Z   
 @gd0   @g 0 
j j
I1 (R) =  p
@n
e
+ ikgd  
0 
j pj  d + ikgd0 
@n
e
SR SR
Z  0
Z 2 ! 12  Z   1
 @g 
 jj 1 2
p 2  d 
+ ikgd 
0
= O(1) O
 @n
e Rd+1
SR SR SR
   1  
1 1
!0 as R ! 1:
2
= O(1)O =O
R2 R
 17

Lemma 3.
Z  
0 @p @g 0
lim gd p d dHd 1
(x) = p(x0 ):
"!0 S" @n
e @n
e

Proof. On S" the following holds:

i (1) @g20 i (1)

g20 = H0 ( k"); = kH1 ( k") ;
4 @n
e 4
   
e ik" @g30 1 e ik" 1 e ik"
g30 = ; = ik = ik + :
4" @n
e " 4" " 4"
Note the minus sign due to the fact that n
e is pointing inward towards x0 . Since
x0 is fixed in the interior of V , we can pick  > 0, such that B (x0 ;  )  V .
Let  be an arbitrary unit vector, then, since p 2 C 2 (V ), for all x 2 B (x0 ;  )
 
 @p 
 (x) = 
 @  j
rp(x)j = jrp(x)j  max jrpj = C < 1;
B (x0 ; )

where C is a constant (depending on x0 and  ). In particular this implies that

j@p(x)=@ nej  C for x 2 S" for "   . Hence
Z   Z 
 @p  i (1) @p 

 g20
@n
e
d H 1 (x) 
 H0 ( k") @ ne
d d 1 (x) H
S" S" 4
Z    
C C
 4 S"
 (1) 
H0 ( k") d
d 1
H
 (1)
(x) = 2" H0 ( k")

4
C  (1) 
"H0 ( k") ! 0 as " ! 0;

=
2
in two space dimensions, and
Z  Z  ik" 
 @p  e @p 

 g30 dH 2
(x)  
 4" @ n d Hd 1
(x)
S" @n
e S" e
Z
C 1
4"
dHd 1
(x) = C" ! 0 as " ! 0;
S"

in three space dimensions.

The second part of the integral over S" is rewritten, using the asymptotics
near the origin, as
Z Z
@g ik (1)
p dHd 1
(x) = (p(x0 ) + O(")) H ( k") dHd 1
(x)
S" @n
e S" 4 1
k (1)
= 2"(p(x0 ) + O(")) H1 ( k")
4i
k (1)
= (p(x0 ) + O(")) "H1 ( k")
2i
k 2i
! (p(x0) + 0) 2i k = p(x0) as " ! 0;
18 APPENDIX B. ASYMPTOTICS FOR THE BESSEL FUNCTIONS

in the two dimensional case, and

Z Z  
@g e ik"
dHd
1
p 1
(x) = (p(x0 ) + O(")) ik +
S" @ n
e S" " 4"
 
1 e ik"
= 4" (p(x0 ) + O(")) ik +
2
" 4"
= (p(x0 ) + O("))(ik" + 1)e ik"
! p(x0) as " ! 0;
in the three dimensional case. Thus in both two and three space dimensions
Z  
@g 0
0 @p
lim gd p d dHd 1 (x)
"!0 S @n
e @n
e
"
Z Z
0 @p @g 0
= lim gd dH (x) lim
d 1
p d dHd 1
(x)
"!0 S @n "!0 S
" e " @n
e
= 0 p(x0 ) = p(x0 ):
Proof of Theorem 1. Taking the limits R ! 1 and " ! 0 using the results
from Lemma 2 and 3 yields
Z  
@p @g
g p dHd 1
(x)
Γ @n
e @ne
Z    
b out
Z
@p @gd0 @p @gd0
= gd0 p dHd 1
(x) gd0 p dHd 1
(x)
SR @ n
e @n
e S" @n
e @n
e
! 0 ( p(x0)) = p(x0);
as R ! 1 and " ! 0. Hence
Z   Z  
0 @p @g 0 @g 0 0 @p
p(x0 ) = gd p d dH d 1
(x) = p d gd dHd 1
(x);
Γ
b out @n
e @n
e Γ
b out @n @n
b
where n is the outward directed normal of Ω.

B Asymptotics for the Bessel functions

The Hankel functions of the first kind are defined as
Hm
(1)
(z ) = Jm (z ) + iYm (z );
where m is the order Jm and Ym are the Bessel functions of the first and
second kind respectively. The asymptotic behavior of the Hankel functions
for positive real arguments can be found in books on special functions, see
for example the book by Schäfke [6]. The asymptotic behavior for the Hankel
functions of the first kind with positive real arguments is given by
r   
Hm ( ) =
(1) 2 i(

e m=2 =4)
1+O
1

 ! +1:
 19

For  > 0 the Bessel functions are given by the following expressions [6]:
1 ( 1)k (=2)p+2k
X
Jp ( ) = ;
k !Γ(p + k + 1)
k=0
1 ( 1)k (=2)2k
X p
J p ( ) = ;
k !Γ(k + 1 p)
k=0
Jp ( ) cos p J p ( )
Ym ( ) = lim ;
p !m sin p

where p  0 and m = 0; 1; : : :. For Jp we have the following:

1 ( 1)k ( =2)p+2k
X
Jp (  ) =
k !Γ(p + k + 1)
k=0
1
X ( 1)k (=2)p+2k
= ( 1)p ( 1)2k ( 1)p Jp ( );
k !Γ(p + k + 1)
k=0
1 ( 1)k ( =2)2k
X p
J p( ) =
k !Γ(k + 1 p)
k=0
1
X 1 ( 1)k (=2)2k
X p
p
= ( 1) ( 1)2k = ( 1) p
J p ( ):
k !Γ(k + 1 p)
k=0 k=0

For Ym we have the following:

Jp (  ) cos p J p (  )
Ym (  ) = lim
p !m sin p
( 1) Jp ( ) cos p ( 1) p J p ( )
p
= lim
p !m sin p
 
( 1)p Jp ( ) cos p ( 1)p J p ( ) (( 1)p ( 1) p ) J p ( )
= lim +
p !m sin p sin p
 
Jp ( ) cos p J p ( )
= lim ( 1)p + 2iJ p ( )
p !m sin p
= ( 1)m Ym ( ) + 2iJ m ( ) = ( 1)m [Ym ( ) + 2iJm ( )] :

Further, since  > 0 it is known that the values of the Bessel functions are
real, hence

Hm
(1)
(  ) = Jm (  ) + iYm (  )
= ( 1)m Jm ( ) + i( 1)m [Ym ( ) + 2iJm ( )]
= ( 1)m Jm ( ) + i( 1)m Ym ( ) 2( 1)m Jm ( )

= ( 1)m+1 (Jm ( ) iYm ( )) = ( 1)m+1 Hm ( );

(1)
20 BIBLIOGRAPHY

where the overline denotes the complex conjugate. Hence, the asymptotic
(1)
behavior of Hm for large real negative arguments is given by

(  ) = ( 1)m+1 Hm ( )
(1)
Hm
(1)
r   
m+1 2 i( 1
= ( 1) e m=2 =4) 1+O
 
r   
= ( 1) m+1 2

e i( m=2 =4)
1+O
1

 ! +1:

Bibliography
[1] E. Bängtsson, D. Noreland, and M. Berggren. Shape optimization of an
acoustic horn. Computer Methods in Applied Mechanics and Engineering,
192:1533–1571, 2003. doi:10.1016/S0045-7825(02)00656-4.

[2] D. Colton and R. Kress. Integral Equation Methods in Scattering Theory.

John Wiley & Sons Inc., 1983.

[3] B. Engquist and A. Majda. Absorbing boundary conditions for numeri-

cal simulation of waves. Mathematics of Computation, 31(139):629–651,
1977. doi:10.2307/2005997.

[4] R. Glowinski. Numerical methods for nonlinear variational problems.

Springer-Verlag, New York, Berlin, Heidelberg, Tokyo, 1984.

[5] J. S. Jensen and O. Sigmund. Systematic design of acoustic devices by

topology optimization. In Twelfth international congress on sound and
vibration, ICVS12 2005, Lisbon, 2005.

[6] F. W. Schäfke. Einfürung in die theorie der speziellen funktionen der

mathematischen physik. Springer, 1963.

[7] O. Sigmund and J. S. Jensen. Design of acoustic devices by topology op-

timization. In In C. Cinquini, M. Rovati, P. Venini, and R. Nascimbene,
editors, Short papers of the 5th World Congress on Structural and Mul-
tidisciplinary Optimization WCSMO5. Lido de Jesolo, May, 2003, pp.
267–268, 2003.

[8] R. Udawalpola and M. Berggren. Optimization of an acoustic horn with

respect to efficiency and directivity. In review., 2006.

[9] E. Wadbro and M. Berggren. Topology optimization of an acoustic horn.

Computer Methods in Applied Mechanics and Engineering, To Appear,
2006. doi:10.1016/j.cma.2006.05.005.
 21

[10] E. Wadbro and M. Berggren. Topology optimization of wave transducers.

In M. P. Bendsøe, N. Olhoff, and O. Sigmund, editors, IUTAM Sympo-
sium on Topological Design Optimization of Structures, Machines and
Materials, pages 301–310. Springer, 2006.
Recent technical reports from the Department of Information Technology
2006-041 Markus Nordén: Performance Modelling for Parallel PDE Solvers on NUMA-Systems
2006-040 Mei Hong, Torsten Söderström, and Wei Xing Zheng: A Simplified Form of the Bias-
Eliminating Least Squares Method for Errors-In-Variables Identification
2006-039 Andreas Hellander and Per Lötstedt: Hybrid Method for the Chemical Master Equation
2006-038 Markus Nordén, Henrik Löf, Jarmo Rantakokko, and Sverker Holmgren: Geographical
Locality and Dynamic Data Migration for OpenMP Implementations of Adaptive PDE
Solvers
2006-037 Elisabeth Larsson, Krister Åhlander, and Andreas Hall: Multi-Dimensional Option Pric-
ing using Radial Basis Functions and the Generalized Fourier Transform
2006-036 Stefan Engblom: A Discrete Spectral Method for the Chemical Master Equation
2006-035 Owe Axelsson and Janos Karátson: Mesh Independent Superlinear PCG Rates via
Compact-Equivalent Operators
2006-034 Agnes Rensfelt and Torsten Söderström: Optimal Excitation for Nonparametric Identi-
fication of Viscoelastic Materials
2006-033 Parosh Aziz Abdulla, Noomene Ben Henda, Richard Mayr, and Sven Sandberg: Lim-
iting Behavior of Markov Chains with Eager Attractors
2006-032 Torsten Söderström: Extending the Frisch Scheme for Errors-in-Variables Identifica-
tion to Correlated Output Noise
2006-031 Håkan Zeffer and Erik Hagersten: A Case for Low-Complexity Multi-CMP Architec-
tures
2006-030 Lars-Henrik Eriksson: The GTO Toolset and Method
2006-029 Lars-Henrik Eriksson: Use of Domain Theories in Applied Formal Methods
2006-028 Ulrika Pettersson, Elisabeth Larsson, Gunnar Marcusson, and Jonas Persson: Im-
proved Radial Basis Function Methods for Multi-Dimensional Option Pricing
2006-027 Jan Nordström: Error Bounded Schemes for Time-Dependent Hyperbolic Problems
2006-026 Jan Nordström, Ken Mattsson, and Charles Swanson: Boundary Conditions for a
Divergence Free Velocity-Pressure Formulation of the Incompressible Navier-Stokes
Equations
2006-025 Johan Wikström, Arvid Kauppi, Arne W. Andersson, and Bengt Sandblad: Designing
a Graphical User Interface for Train Traffic Control
2006-024 Arvid Kauppi, Johan Wikström, Bengt Sandblad, and Arne W. Andersson: Control
Strategies for Managing Train Traffic, Difficulties Today and Solutions for the Future
2006-023 Gergana Bencheva, Svetozar Margenov, and Jiřı́ Starý: MPI Implementation of a PCG
Solver for Nonconforming FEM Problems: Overlapping of Communications and Com-
putations

September 2006
ISSN 1404-3203
http://www.it.uu.se/