5

Fundamental Solutions

We now consider the scalar partial differential equation (1.10),

(Lu)(x) = f (x) for x ∈ Ω ⊂ Rd ,

with an elliptic partial differential operator of second order,

d 
∂ ∂
(Lu)(x) = − aji (x) u(x) .
i,j=1
∂xj ∂xi

The associated conormal derivative (1.7) is

d
∂
γ1int u(x) = nj (x)aji (x) u(x) for x ∈ Γ
i,j=1
∂xi

and Green’s second formula (1.8) reads for the solution u of the partial dif-
ferential equation (1.10) and for an arbitrary test function v
  
(Lv)(y)u(y)dy = γ1 u(y)γ0 v(y)dsy − γ1int v(y)γ0int u(y)dsy
int int

Ω Γ Γ

+ f (y)v(y)dy.
Ω

If there exists for any x ∈ Ω a function v(y) := U ∗ (x, y) satisfying


(Ly U ∗ )(x, y)u(y)dy = u(x) (5.1)
Ω

then the solution of the partial differential equation (1.10) is given by the
representation formula for x ∈ Ω
90 5 Fundamental Solutions
 
u(x) = U ∗ (x, y)γ1int u(y)dsy − int U ∗ (x, y)γ int u(y)ds
γ1,y 0 y (5.2)
Γ Γ

+ U ∗ (x, y)f (y)dy.
Ω

Hence we can describe any solution of the partial differential equation (1.10)
just by knowing the Cauchy data [γ0int u(x), γ1int u(x)] for x ∈ Γ .
Due to 
u(x) = δ0 (y − x)u(y)dy for x ∈ Ω
Ω

we have to solve a partial differential equation in the distributional sense to

find the solution of (5.1),

(Ly U ∗ )(x, y) = δ0 (y − x) for x, y ∈ Rd . (5.3)

Any solution U ∗ (x, y) of (5.3) is called a fundamental solution.

The existence of a fundamental solution U ∗ (x, y) is essential to derive the
representation formula (5.2), and therefore to formulate appropriate boundary
integral equations to find the complete Cauchy data. For general results on
the existence of fundamental solutions for partial differential operators we
refer to [79, 90]. In particular for partial differential operators with piecewise
constant coefficients the existence of a fundamental solution is ensured. But
here we will only consider the explicit computation of fundamental solutions
for the Laplace operator, for the system of linear elastostatics, for the Stokes
system, and for the Helmholtz operator.

5.1 Laplace Operator

Let us first consider the Laplace operator

(Lu)(x) := −∆u(x) for x ∈ Rd , d = 2, 3.

The corresponding fundamental solution U ∗ (x, y) is the distributional solution

of the partial differential equation

−∆y U ∗ (x, y) = δ0 (y − x) for x, y ∈ Rd .

Since the Laplace operator is invariant with respect to translations and rota-
tions, we can find the fundamental solution as U ∗ (x, y) = v(z) with z := y −x.
Hence we have to solve

−∆v(z) = δ0 (z) for z ∈ Rd . (5.4)

Applying the Fourier transformation (2.14) this gives, when considering the
derivation rule (2.17),
 5.1 Laplace Operator 91

1
|ξ|2 v%(ξ) =
(2π)d/2
and therefore
1 1
v%(ξ) = ∈ S  (Rd ).
(2π)d/2 |ξ|2
For the Fourier transform v% of a tempered distribution v ∈ S  (Rd ) we have
by definition

%
v , ϕL2 (Rd ) = v, ϕ
% L2 (Rd ) for all ϕ ∈ S(Rd ).

Using 
ϕ(ξ) = (2π)−d/2 %
eiz,ξ ϕ(z)dz
Rd

it follows that
 
1 1
%
v , ϕL2 (Rd ) = %
eiz,ξ ϕ(z)dzdξ.
(2π)d |ξ|2
Rd Rd

Since the integral 

1
dξ
|ξ|2
Rd

does not exist we can not exchange the order of integration. However, using

∆z eiz,ξ = −|ξ|2 eiz,ξ

we can consider a splitting of the exterior integral, apply integration by parts

and exchange the order of integration, and repeat integration by parts to
obtain
 
1 1
%
v , ϕL2 (Rd ) = %
eiz,ξ ϕ(z)dzdξ
(2π)d |ξ|2
Rd Rd
     
1 1 1 1 eiz,ξ
= %
eiz,ξ ϕ(z)dzdξ + −∆z %
ϕ(z)dzdξ
(2π)d |ξ|2 (2π)d |ξ|2 |ξ|2
|ξ|≤1 Rd |ξ|>1 Rd
   
1 1 1 1
= %
eiz,ξ ϕ(z)dzdξ + %
eiz,ξ [−∆z ϕ(z)] dzdξ
(2π)d |ξ|2 (2π)d |ξ|4
|ξ|≤1 Rd |ξ|>1 Rd
  iz,ξ  
1 e 1 eiz,ξ
= %
ϕ(z) dξdz + %
[−∆z ϕ(z)] dξdz
(2π)d |ξ|2 (2π)d |ξ|4
Rd |ξ|≤1 Rd |ξ|>1
⎡ ⎤
  
1 ⎢ eiz,ξ eiz,ξ ⎥
= %
ϕ(z) ⎣ dξ − ∆z dξ ⎦ dz.
(2π)d |ξ|2 |ξ|4
Rd |ξ|≤1 |ξ|>1
92 5 Fundamental Solutions

In the three–dimensional case d = 3 we use spherical coordinates

ξ1 = r cos ϕ sin θ, ξ2 = r sin ϕ sin θ, ξ3 = r cos θ

for r ∈ (0, ∞), ϕ ∈ (0, 2π), θ ∈ (0, π) to obtain, by using Lemma 2.13,
⎡ ⎤
 iz,ξ  iz,ξ
1 ⎢ e e ⎥
v(z) = v(|z|) = ⎣ dξ − ∆z dξ ⎦
(2π)3 |ξ|2 |ξ|4
|ξ|≤1 |ξ|>1
⎡ 2π π 1
  
1 ⎣
= ei|z|r cos θ sin θ dr dθ dϕ
(2π)3
0 0 0
⎤
2ππ ∞
ei|z|r cos θ sin θ
−∆z dr dθ dϕ⎦
r2
0 0 1
⎡ π 1 ⎤
  π ∞ i|z|r cos θ
1 ⎣ e sin θ
= ei|z|r cos θ sin θ dr dθ − ∆z dr dθ ⎦ .
(2π)2 r2
0 0 0 1

The transformation u := cos θ gives

π 1
i|z|r cos θ 1 2 i|z|r 3 2
e sin θdθ = ei|z|ru du = e − e−i|z|r = sin |z|r
i|z|r |z|r
0 −1

and therefore
⎡ ⎤
1 ∞
1 ⎣ sin |z|r sin |z|r ⎦
v(z) = dr − ∆z dr .
2π 2 |z|r |z|r3
0 1

Using the transformation s := |z|r we obtain for the first integral

1 |z|
sin |z|r 1 sin s Si(|z|)
I1 := dr = ds = .
|z|r |z| s |z|
0 0

With
 
sin ax 1 sin ax a cos ax
dx = − + dx
x3 2 x2 2 x2

1 sin ax a cos ax a2 sin ax
=− − − dx
2 x2 2 x 2 x
the computation of the second integral gives
 5.1 Laplace Operator 93

∞  ∞ ∞
sin |z|r 1 sin |z|r 1 cos |z|r |z| sin |z|r
I2 := 3
dr = − 2
− − dr
|z|r 2 |z|r 2 r 1 2 r
1 1
⎡∞ ⎤
 1
1 sin |z| 1 |z| ⎣ sin |z|r sin |z|r ⎦
= + cos |z| − dr − dr
2 |z| 2 2 r r
0 0
1 sin |z| 1 |z| 2 π 3
= + cos |z| − − Si(|z|) .
2 |z| 2 2 2

Inserting this and applying the differentiation we obtain

  
1 Si(|z|) 1 sin |z| 1 π 1
v(z) = − ∆ z + cos |z| − |z| + |z|Si(|z|)
2π 2 |z| 2 |z| 2 4 2
  
1 1 Si(|z|) 1 sin |z| 1 1
= ∆z |z| + 2 − ∆z + cos |z| + |z|Si(|z|)
8π 2π |z| 2 |z| 2 2
4 56 7
1 1 =0
= .
4π |z|

Hence the fundamental solution of the Laplace operator in three space dimen-
sions is
1 1
U ∗ (x, y) = for x, y ∈ R3 .
4π |x − y|
For the two–dimensional case d = 2 the inverse Fourier transform of the
fundamental solution has to be regularized in some appropriate way [154]. By
 
1 ϕ(x) − ϕ(0) ϕ(x)
P 2 , ϕL2 (Rd ) = dx + dx
|x| |x|2 |x|2
x∈R2 :|x|≤1 x∈R2 :|x|>1

1
we first define the tempered distribution P ∈ S  (Rd ). Then,
|x|2
 
1 ϕ(ξ) − ϕ(0) ϕ(ξ)
2π v, ϕ
% L2 (R2 ) = P , ϕL2 (R2 ) = dξ + dξ
|ξ|2 |ξ|2 |ξ|2
ξ∈R2 :|ξ|≤1 ξ∈R2 :|ξ|>1

for all ϕ ∈ S(R2 ). With

 
1 1
ϕ(ξ) = e iz,ξ
%
ϕ(z)dz, ϕ(0) = %
ϕ(z)dz
2π 2π
R2 R2

we then obtain
   
1 1
(2π)2 v, ϕ
% L2 (R2 ) = [eiz,ξ − 1]ϕ(z)dzdξ
% + %
eiz,ξ ϕ(z)dzdξ.
|ξ|2 |ξ|2
ξ∈R2 :|ξ|≤1 R2 ξ∈R2 :|ξ|>1 R2
94 5 Fundamental Solutions

Again we can not exchange the order of integration in the second term. How-
ever, as in three–dimensional case we can write
⎡ ⎤
  
⎢ eiz,ξ − 1 eiz,ξ ⎥
(2π)2 v, ϕ
% L2 (R2 ) = ϕ(z)
% ⎣ 2
dξ + dξ ⎦ dz.
|ξ| |ξ|2
R2 ξ∈R2 :|ξ|≤1 ξ∈R2 :|ξ|>1

With Lemma 2.13 we further have

 
1 eiz,ξ − 1 eiz,ξ
v(z) = v(|z|) = dξ + dξ
(2π)2 |ξ|2 |ξ|2
ξ∈R2 :|ξ|≤1 ξ∈R2 :|ξ|>1

and using polar coordinates we obtain

1 2π 2 3 ∞ 2π
1 1 ir|z| cos ϕ 1 1 ir|z| cos ϕ
v(z) = e − 1 dϕdr + e dϕdr
(2π)2 r (2π)2 r
0 0 1 0
1 ∞
1 1 1 1
= [J0 (r|z|) − 1]dr + J0 (r|z|)dr
2π r 2π r
0 1

with the first order Bessel function [63, Subsection 8.411],

2π
1
J0 (s) = eis cos ϕ dϕ.
2π
0

Substituting r := s/ we compute

 ∞
1 J0 (s) − 1 1 J0 (s)
v(z) = ds + ds
2π s 2π s
0 
1 ∞ 1
1 J0 (s) − 1 1 J0 (s) 1 1
= ds + ds + ds
2π s 2π s 2π s
0 1 
1 C0
=− log |z| − (5.5)
2π 2π
with the constant
1 ∞
1 − J0 (s) J0 (s)
C0 := ds − ds.
s s
0 1

Since any constant satisfies the homogeneous Laplace equation we can ne-
glect constant terms in the definition of the fundamental solution. Hence the
fundamental solution of the Laplace operator in two space dimensions is
 5.1 Laplace Operator 95

1
U ∗ (x, y) = − log |x − y| for x, y ∈ R2 .
2π
In what follows we will describe an alternative approach to compute the fun-
damental solution for the Laplace operator in two space dimensions. From
Lemma 2.13 we know that the solution v(z) depends only on the absolute
value  := |z|. For z = 0 the partial differential equation (5.4) can be rewrit-
ten in polar coordinates as
 2 
∂ 1 ∂
+ v() = 0 for  > 0.
∂2  ∂

The general solution of this ordinary differential equation is given by

v() = a log  + b, a, b ∈ R.

In particular for b = 0 we have

U ∗ (x, y) = a log |x − y| .

For x ∈ Ω and for a sufficient small ε > 0 let Bε (x) ⊂ Ω be a ball with center
x and with radius ε. For y ∈ Ω\Bε (x) the fundamental solution U ∗ (x, y) is
a solution of the homogeneous Laplace equation −∆y U ∗ (x, y) = 0. Applying
Green’s second formula (1.8) with respect to the bounded domain Ω\Bε (x)
we obtain
  
∂ ∂
0 = U ∗ (x, y) u(y)dsy − U ∗ (x, y)u(y)dsy + U ∗ (x, y)f (y)dy
∂ny ∂ny
Γ Γ Ω\B ε (x)
 
∂ ∂
+ U ∗ (x, y) u(y)dsy − U ∗ (x, y)u(y)dsy .
∂ny ∂ny
∂Bε (x) ∂Bε (x)

Taking the limit ε → 0 we first bound

   
     
   
 ∗ ∂   ∂ 
 U (x, y) u(y)dsy  = |a| | log ε|  u(y)dsy 
 ∂ny   ∂ny 
 ∂Bε (x)   ∂Bε (x) 

≤ |a| 2π ε | log ε| uC 1 (Ω) .

1
Using ny = (x − y) for y ∈ ∂Bε (x) we have
ε
  
∂ ∗ (ny , y − x) a
U (x, y)u(y)dsy = a u(y)dsy = − u(y)dsy .
∂ny |x − y|2 ε
∂Bε (x) ∂Bε (x) ∂Bε (x)

The Taylor expansion

96 5 Fundamental Solutions

u(y) = u(x) + (y − x)∇u(ξ)

with a suitable ξ = x + t(y − x), t ∈ (0, 1) yields

 
∂ a
U ∗ (x, y)u(y)dsy = −a 2π u(x) − (y − x)∇u(ξ)dy
∂ny ε
∂Bε (x) ∂Bε (x)

where  
 
a  
 
 (y − x)∇u(ξ)dy  ≤ |a| 2π ε uC 1 (Ω) .
ε 
 ∂Bε (x) 
Taking the limit ε → 0 gives
 
∗ ∂ ∂
−a 2π u(x) = U (x, y) u(y)dsy − U ∗ (x, y)u(y)dsy
∂ny ∂ny
Γ Γ

+ U ∗ (x, y)f (y)dy
Ω

and therefore the representation formula when choosing a = −1/2π.

To summarize, the fundamental solution of the Laplace operator is given
by ⎧
⎪ 1
⎪−
⎨ log |x − y| for d = 2,
2π
U ∗ (x, y) = (5.6)
⎪
⎪ 1 1
⎩ for d = 3.
4π |x − y|
Any solution of the partial differential equation

−∆u(x) = f (x) for x ∈ Ω ⊂ Rd

is therefore given by the representation formula for x ∈ Ω

 
∂ ∂
u(x) = U ∗ (x, y) u(y)dsy − U ∗ (x, y)u(y)dsy (5.7)
∂ny ∂ny
Γ Γ

+ U ∗ (x, y)f (y)dy.
Ω

5.2 Linear Elasticity

Let us now consider the system of linear elastostatics (1.25),

−µ∆u(x) − (λ + µ)grad div u(x) = f (x) for x ∈ Ω ⊂ Rd ,

 5.2 Linear Elasticity 97

and the associated second Betti formula (1.30),


d 
∂
σij (v, y)ui (y)dy = γ0int v(y) γ1int u(y)dsy (5.8)
i,j=1
∂yj
Ω
 Γ 
− γ0 u(y) γ1 v(y)dsy + f (y) v(y)dy.
int  int

Γ Ω

To derive a representation formula for the components uk (x), x ∈ Ω, we

therefore have to find solutions v k (x, y) satisfying

d
∂
σij (v k (x, y), y)ui (y)dy = uk (x) for x ∈ Ω, k = 1, . . . , d.
i,j=1
∂y j
Ω

Let ek ∈ Rd be the unit vector with ek = δk for k,  = 1, . . . , d. Using the

transformation z := y − x we have to solve the partial differential equations,
k = 1, . . . , d,

−µ∆z v k (z) − (λ + µ)gradz divz v k (z) = δ0 (z) ek for z ∈ Rd .

Using (1.32),

λ+µ
v k (z) := ∆[ψ(z)ek ] − grad div [ψ(z)ek ],
λ + 2µ
we have to find the Airy stress function ψ satisfying the Bi–Laplace equation

−µ∆2 ψ(z) = δ0 (z) for z ∈ Rd

or
−µ∆ϕ(z) = δ0 (z), ∆ψ(z) = ϕ(z) for z ∈ Rd .
From the fundamental solution of the Laplace operator we find
⎧
⎪ 1 1
⎪
⎨ − µ 2π log |z|,
⎪ for d = 2,
ϕ(z) =
⎪
⎪ 1 1 1
⎪
⎩ for d = 3.
µ 4π |z|

For d = 2 we have to solve the remaining Poisson equation when using polar
coordinates,
 2 
∂ 1 ∂  1 1
2
+ ψ() = − log  for  > 0,
∂  ∂ µ 2π

with the general solution

98 5 Fundamental Solutions

 1 1 0 2 1
ψ() = −  log  − 2 + a log  + b for  > 0, a, b ∈ R.
µ 8π
In particular for a = b = 0 we have

 1 1 0 2 1
ψ() = −  log  − 2 .
µ 8π

Due to ∆2 |z|2 = 0 we then obtain for  = |z|

1 1
ψ(z) = − |z|2 log |z| .
µ 8π

For k = 1 we find for v 1 (z)

λ + µ ∂2 λ + µ ∂2
v11 (z) = ∆ψ(z) − ψ(z), v21 (z) = − ψ(z).
λ + 2µ ∂z12 λ + 2µ ∂z1 ∂z2

∂ zi
Using |z| = we obtain
∂zi |z|

∂ 1 1
ψ(z) = − [2zi log |z| + zi ] (i = 1, 2)
∂zi µ 8π
 
∂2 1 1 z12
ψ(z) = − 2 log |z| + 2 + 1 (i = 1, 2)
∂zi2 µ 8π |z|2
∂2 1 1 z1 z 2
ψ(z) = −
∂z1 ∂z2 µ 4π |z|2

and therefore
 2 
1 λ + 3µ 1 λ+µ z1 3
v11 (z) = − log |z| + − ,
4π µ(λ + 2µ) 4π µ(λ + 2µ) |z|2 2
1 λ + µ z1 z 2
v21 (z) = .
4π µ(λ + 2µ) |z|2

For k = 2 the computation is almost the same. Since the constants are so-
lutions of the homogeneous system we can neglect them when defining the
fundamental solution. From v k for k = 1, 2 we then find the Kelvin solution
tensor U ∗ (x, y) = (v 1 , v 2 ) with the components
 
∗ 1 λ+µ λ + 3µ (yk − xk )(y − x )
Uk (x, y) = − log |x − y| δk +
4π µ(λ + 2µ) λ+µ |x − y|2

for k,  = 1, 2. Inserting the Lamé constants (1.24) this gives

 
∗ 1 1 1+ν (yk − xk )(y − x )
Uk (x, y) = (4ν − 3) log |x − y| δk + .
4π E 1 − ν |x − y|2
 5.2 Linear Elasticity 99

This is the fundamental solution of linear elastostatics in two space dimensions

which even exists for incompressible materials with ν = 1/2.
For d = 3 we have to solve the Poisson equation, by using spherical coor-
dinates we obtain
 
1 ∂ 2 ∂  1 1 1
 ψ() = for  > 0,
2 ∂ ∂ µ 4π 
with the general solution
 
 1 1 1 a
ψ() = + +b , for  > 0, a, b ∈ R.
µ 4π 2 
For a = b = 0 we have
1 1
ψ(z) = |z| .
µ 8π
For k = 1 we obtain v 1 (z)

λ + µ ∂2
v11 (z) = ∆ψ(z) − ψ(z),
λ + 2µ ∂z12
λ + µ ∂2
v21 (z) = − ψ(z),
λ + 2µ ∂z1 ∂z2
λ+µ ∂
v31 (z) = − ψ(z)
λ + 2µ ∂z1 ∂z3
and with the derivatives
 
∂ 1 1 zi ∂2 1 1 1 z2
ψ(z) = , 2 ψ(z) = − i3 ,
∂zi µ 8π |z| ∂zi µ 8π |z| |z|

∂2 1 1 zi zj
ψ(z) = − for i = j
∂zi ∂zj µ 8π |z|3

we then find for the components of the solution v 1

1 λ + 3µ 1 1 λ+µ z12
v11 (z) = + ,
8π µ(λ + 2µ) |z| 8π µ(λ + 2µ) |z|3
1 λ + µ z1 z2
v21 (z) = ,
8π µ(λ + 2µ) |z|3
1 λ + µ z1 z3
v31 (z) = .
8π µ(λ + 2µ) |z|3
For k = 2, 3 the computations are almost the same. Hence, the Kelvin solution
tensor is given by U ∗ (x, y) = (v 1 , v 2 , v 3 ) where
 
∗ 1 λ+µ λ + 3µ δk (yk − xk )(y − x )
Uk (x, y) = +
8π µ(λ + 2µ) λ + µ |x − y| |x − y|3
100 5 Fundamental Solutions

for k,  = 1, . . . , 3. Inserting the Lamé constants (1.24) this gives the funda-
mental solution of linear elastostatics in three space dimensions
 
∗ 1 1 1+ν δk (yk − xk )(y − x )
Uk (x, y) = (3 − 4ν) + .
8π E 1 − ν |x − y| |x − y|3
Hence we have the fundamental solution of linear elastostatics
 
∗ 1 1 1+ν (xk − yk )(x − y )
Uk (x, y) = (3 − 4ν)E(x, y)δk +
4(d − 1)π E 1 − ν |x − y|d
(5.9)
for k,  = 1, . . . , d with
⎧
⎨ − log |x − y|
⎪ for d = 2,
E(x, y) = 1
⎪
⎩ for d = 3.
|x − y|
Inserting the solution vectors v(y) = U ∗k (x, y) into the second Betti formula
(5.8) this gives the representation formula
 
uk (x) = U ∗k (x, y) γ1int u(y)dsy − u(y) γ1,yint U ∗ (x, y)ds
k y
Γ Γ

+ f (y) U ∗k (x, y)dy (5.10)
Ω

for x ∈ Ω and k = 1, . . . , d. Thereby, the boundary stress T ∗k (x, y) of the

fundamental solution U ∗k (x, y) is given for almost all y ∈ Γ by applying (1.27)
as

T ∗k (x, y) := γ1,y
int U ∗ (x, y)
k
∂
= λ divy U ∗k (x, y) n(y) + 2µ U ∗ (x, y) + µ n(y) × curly U ∗k (x, y).
∂ny k
Using
1 1 1+ν yk − xk
div U ∗k (x, y) = 2(2ν − 1)
4(d − 1)π E 1 − ν |x − y|d
we then obtain
1 ν yk − xk E ∂
T ∗k (x, y) = − n(y) + U ∗ (x, y)
2(d − 1)π 1 − ν |x − y|d 1 + ν ∂ny k
E
+ n(y) × curly U ∗k (x, y). (5.11)
2(1 + ν)
Obviously, both the fundamental solutions U ∗k (x, y) and the corresponding
boundary stress functions T ∗k (x, y) exist also for incompressible materials with
ν = 1/2.
 5.3 Stokes Problem 101

5.3 Stokes Problem

Next we consider the Stokes system (1.38),

−µ∆u(x) + ∇p(x) = f (x), div u(x) = 0 for x ∈ Ω ⊂ Rd .

For the solution u and for an arbitrary vector field v we obtain from Green’s
first formula (1.41) by using the symmetry a(u, v) = a(v, u) Green’s second
formula
d 

 
∂
−µ∆vi (y) + q(y) ui (y)dy + p(y)div v(y)dy (5.12)
∂yi
Ω i=1 Ω


d 

d 
= ti (u, p)vi (y)dsy − ti (v, q)ui (y)dsy + f (y) v(y)dy
Γ i=1 Γ i=1 Ω

with the conormal derivative t(u, p) as given in (1.43).

To obtain representation formulae for the components uk (x), x ∈ Ω, for
the velocity field u we have to find solutions v k (x, y) and q k (x, y) such that

d  
∂ k
−µ∆vi (x, y) +
k
q (x, y) ui (y)dy = uk (x), divy v k (x, y) = 0
i=1
∂yi
Ω

for x ∈ Ω, k = 1, . . . , d. With the transformation z := y − x we have to solve

for k = 1, . . . , d

−µ∆v k (z) + ∇q k (z) = δ0 (z)ek , div v k (z) = 0 for z ∈ Rd .

The application of the Fourier transform (2.14) gives

1

d
µ |ξ|2 v%jk (ξ) + i ξj q%k (ξ) = δjk (j = 1, . . . , d), i ξj v%jk (ξ) = 0.
(2π)d/2 j=1

In particular for d = 2 and k = 1 we have to solve a linear system,

1
µ |ξ|2 v%11 (ξ) + i ξ1 q%1 (ξ) = ,
2π
µ |ξ|2 v%21 (ξ) + i ξ2 q%1 (ξ) = 0,

iξ1 v%11 (ξ) + iξ2 v%21 (ξ) = 0,

yielding the solution

 
1 1 1 ξ12 1 1 ξ1 ξ2 i ξ1
v%11 (ξ) = − , v%21 (ξ) = − , q%1 (ξ) = − .
µ 2π |ξ|2 |ξ|4 µ 2π |ξ|4 2π |ξ|2
102 5 Fundamental Solutions

As for the scalar Laplace equation we obtain

  
1 1 1 ξ12
v11 (z) = eiz,ξ
− dξ
µ (2π)2 |ξ|2 |ξ|4
R 2
⎡ ⎤
 2 
1 1 1 1 ∂ ⎣ 1 1
= eiz,ξ 2 dξ + eiz,ξ 4 dξ ⎦
µ (2π)2 |ξ| µ ∂z12 (2π)2 |ξ|
R2 R2

and with (5.5) we have


1 1 1 C0
eiz,ξ dξ = − log |z| − .
(2π)2 |ξ|2 2π 2π
R2

On the other hand,

⎛ ⎞
 
1 1 ⎠ = − 1 1 1 C0
∆⎝ 2
eiz,ξ
4
dξ 2
eiz,ξ 2 dξ = log |z| +
(2π) |ξ| (2π) |ξ| 2π 2π
R2 R2

implies

1 1 1 0 2 1 C0 2
eiz,ξ dξ = |z| log |z| − |z|2 + |z| + C1 + C2 log |z|
(2π)2 |ξ|4 8π 8π
R2

with some constants C1 , C2 ∈ R. In particular for C1 = C2 = 0 this gives

   
1 1 C0 1 ∂2 1  2  C0 2
v11 (z) = − log |z| − + |z| log |z| − |z|2
+ |z|
µ 2π 2π µ ∂z12 8π 8π
 2

1 1 z 2C0 + 1
= − log |z| + 12 − .
µ 4π |z| 2
Analogous computations yield
⎡ ⎤
 2 
1 1 ξ1 ξ2 1 ∂ ⎣ 1 1
v21 (z) = − eiz,ξ 4 dξ = eiz,ξ 4 dξ ⎦
µ (2π)2 |ξ| µ ∂z1 ∂z2 (2π)2 |ξ|
R2 R2
 
1 ∂2 1  2  C0 2 1 1 z1 z2
= |z| log |z| − |z|2 + |z| =
µ ∂z1 ∂z2 8π 8π µ 4π |z|2
and
⎡ ⎤
 
i ξ1 ∂ ⎣ 1 1
q 1 (z) = − eiz,ξ dξ = − eiz,ξ dξ ⎦
(2π)2 |ξ|2 ∂z1 (2π)2 |ξ|2
R2 R2
 
∂ 1 C0 1 ∂
=− − log |z| − = log |z|.
∂z1 2π 2π 2π ∂z1
 5.3 Stokes Problem 103

For d = 2 and k = 2 the computations are almost the same. Neglecting the
constants we finally have the fundamental solution for the Stokes system in
two space dimensions,
 
∗ 1 1 (yk − xk )(y − x )
Uk (x, y) = − log |x − y| δk + (5.13)
4π µ |x − y|2
1 yk − xk
Q∗k (x, y) = (5.14)
2π |x − y|2
and k,  = 1, 2.
For d = 3 we obtain in the same way the fundamental solution for the
Stokes system as
 
∗ 1 1 δk (yk − xk )(y − x )
Uk (x, y) = +
8π µ |x − y| |x − y|3
1 yk − xk
Q∗k (x, y) =
4π |x − y|3
and k,  = 1, . . . , 3.
Comparing the above results with the fundamental solution of the system
of linear elasticity we obtain the equality for
1 1+ν 1
= , (3 − 4ν) = 1
E 1−ν µ
and therefore for
1
ν =
, E = 3µ .
2
The fundamental solution of the linear elasticity system with incompressible
material therefore coincides with the fundamental solution of the Stokes sys-
tem.
Inserting the fundamental solutions v(y) = U ∗k (x, y) and q(y) = Q∗k (x, y)
into the second Greens formula (5.12) this gives the representation formulae
 
uk (x) = U k (x, y) t(u(y), p(y))dsy − u(y) t(U ∗k (x, y), Q∗k (x, y))dsy
∗ 

Γ Γ

+ f (y) U ∗k (x, y)dy (5.15)
Ω

for x ∈ Ω and k = 1, . . . , d. Hereby the conormal derivative T ∗k (x, y) is defined

via (1.43) for almost all y ∈ Γ by
T ∗k (x, y) = t(U ∗k (x, y), Q∗k (x, y))
∂
= −Q∗k (x, y)n(x) + 2µ U ∗ (x, y) + µ n(x) × curl U ∗k (x, y)
∂ny k
1 yk − xk ∂
= − n(x) + 2µ U ∗ (x, y) + µ n(x) × curl U ∗k (x, y).
2(d − 1)π |x − y|d ∂ny k
104 5 Fundamental Solutions

Hence the boundary stress (1.43) of the fundamental solution of the Stokes
system also coincides with the boundary stress (1.27) of the fundamental
solution of the linear elasticity system when choosing ν = 12 and E = 3µ.
It remains to find some appropriate representation formulae for the pres-
sure p. Let us first consider the case d = 2 and the second Green formula
(5.12) where we have to find solutions v 3 (z) and q 3 (z) with z := y − x such
that
−µ∆v 3 (z) + ∇q 3 (z) = 0, div v 3 (z) = δ0 (z) for z ∈ R2 .
By applying the Fourier transformation we obtain the linear system

µ |ξ|2 v%13 (ξ) + i ξ1 q%3 (ξ) = 0,

µ |ξ|2 v%23 (ξ) + i ξ2 q%3 (ξ) = 0,
1
iξ1 v%13 (ξ) + iξ2 v%23 (ξ) =
2π
with the solution
i ξ1 i ξ2 µ
v%13 (ξ) = − , v%23 (ξ) = − , q%3 (ξ) = .
2π |ξ|2 2π |ξ|2 2π

As before we obtain
1 ∂
vi3 (z) = log |z| (i = 1, 2), q 3 (z) = µδ0 (z).
2π ∂zi
Using z := y − x we conclude for x ∈ Ω a representation formula for the
pressure


2 

2
p(x) = ti (u, p)vi3 (x, y)dsy − ti (v 3 (x, y), q 3 (x, y))ui (y)dsy
Γ i=1 Γ i=1


2
+ v 3i (x, y)fi (y)dy
Ω i=1

where the conormal derivative (1.43) implies

ti (v 3 (x, y), q 3 (x, y)) = −[µδ0 (y − x) + div v 3 (x, y)]ni (x)

2


+2µ eij (v 3 (x, y), y)nj (y)
j=1

for i = 1, 2, x ∈ Ω and y ∈ Γ . Since v 3 is divergence–free,

2 2
∂ 3 1
∂2
div v 3 (x, y) = vi (x, y) = log |x − y| = 0,
i=1
∂yi 2π i=1 ∂yi2
 5.4 Helmholtz Equation 105

we obtain for Γ  y = x ∈ Ω
2


ti (v 3 (x, y), q 3 (x, y)) = 2µ eij (v 3 (x, y), y)nj (y).
j=1

Moreover,
 
3 1 ∂ 3 ∂ 3
eij (v (x, y), y) = v (x, y) + v (x, y)
2 ∂yi j ∂yj i
 
1 ∂ ∂ ∂ ∂
= log |x − y| + log |x − y|
4π ∂yi ∂yj ∂yj ∂yi
1 ∂ ∂
= log |x − y|
2π ∂yi ∂yj
1 ∂ ∂ ∂ ∗
=− log |x − y| = − Q (x, y).
2π ∂xj ∂yi ∂xj i

Finally we obtain the representation formula for the pressure p, x ∈ Ω,



2 

2
∂ ∗
p(x) = ti (u, p)Q∗i (x, y)dy + 2µ Q (x, y)nj (y)ui (y)dsy
i=1 i,j=1
∂xj i
Γ Γ

+ fi (y)Q∗i (x, y)dy . (5.16)
Ω

For d = 3 one may obtain a similar formula, we skip the details.

5.4 Helmholtz Equation

Finally we consider the Helmholtz

−∆u(x) − k 2 u(x) = 0 for x ∈ Rd , k ∈ R (5.17)

where the computation of the fundamental solution can be done as in the

alternative approach for the Laplace equation.
For d = 3, and by using spherical coordinates we have to solve the partial
differential equation to find v() = v(|x|) = u(x) such that
 
1 ∂ 2 ∂
− 2  v() − k 2 v() = 0 for  > 0.
 ∂ ∂

With the transformation

1
v() = V ()

this is equivalent to
106 5 Fundamental Solutions

V  () + k 2 V () = 0 for  > 0

where the general solution is given by

V () = A1 cos k + A2 sin k

and therefore we obtain

1 cos k sin 
v() = V () = A1 + A2 .
  
When considering the behavior as  → 0 we find a fundamental solution of
the Helmholtz equation given by

1 cos k|x − y|
Uk∗ (x, y) = for x, y ∈ R3 .
4π |x − y|

However, it is more common to use a complex combination of the above fun-

damental system to define the fundamental solution by

1 eik|x−y|
Uk∗ (x, y) = for x, y ∈ R3 . (5.18)
4π |x − y|

For d = 2, and by using polar coordinates the Helmholtz equation (5.17) reads

∂2 1 ∂
− v() − v() − k 2 v() = 0 for  > 0,
∂2  ∂
or
∂2 ∂
2 v() +  v() + k 2 2 v() = 0 for  > 0.
∂2 ∂
With the substitution
s 1 ∂
s = k, v() = v( ) = V (s), V  (s) = v()
k k ∂
we then obtain a Bessel differential equation of order zero,

s2 V  (s) + sV  (s) + s2 V (s) = 0 for s > 0. (5.19)

To find a fundamental system of the Bessel differential equation (5.19) we first

consider the ansatz
∞

∞

∞


V1 (s) = vk sk , V1 (s) = vk ksk−1 , V1 (s) = vk k(k − 1)sk−2 .
k=0 k=1 k=2

By inserting this into the differential equation (5.19) we obtain

 5.4 Helmholtz Equation 107

0 = s2 V1 (s) + sV1 (s) + s2 V1 (s)

∞
∞ ∞


= vk k(k − 1)sk + vk ksk + vk sk+2
k=2 k=1 k=0

∞
0 1
= vk−2 + k 2 vk sk + v1 s for s > 0
k=2

and thus
1
v1 = 0, vk = − vk−2 for k ≥ 2.
k2
Hence we obtain
1
v2 −1 = 0, v2 = − v2( −1) for  = 1, 2, . . .
42
and therefore
(−1)
v2 = v0 for  = 1, 2, . . . .
4 (!)2
In particular for v0 = 1 we have
∞

(−1) 2
V1 (s) = 1 + s =: J0 (s)
4 (!)2
=1

which is the first kind Bessel function of order zero. Note that

lim J0 (s) = 1.
s→0

To find a second solution of the fundamental system including a logarithmic

singularity we consider the ansatz

V2 (s) = J0 (s) ln s + W (s) for s > 0.

By using
1
V2 (s) = J0 (s) ln s + J0 (s) + W  (s),
s
2 1
V2 (s) = J0 (s) ln s + J0 (s) − 2 J0 (s) + W  (s)
 
s s
we obtain

0 = s2 V2 (s) + sV2 (s) + s2 V2 (s)

0 1
= s2 J0 (s) + sJ0 (s) + s2 J0 (s) ln s
+2sJ0 (s) + s2 W  (s) + sW  (s) + s2 W (s)
= 2sJ0 (s) + s2 W  (s) + sW  (s) + s2 W (s)

since J0 (s) is a solution of the Bessel differential equation (5.19). Hence we

have to solve the differential equation
108 5 Fundamental Solutions

s2 W  (s) + sW  (s) + s2 W (s) + 2sI0 (s) for s > 0.

With
∞

∞


W (s) = wk sk , J0 (s) = vk ksk−1
k=0 k=1

we have to solve
∞

∞


0 2 1
0= k wk + wk−2 sk + w1 s + 2 vk ksk
k=2 k=1

∞
0 2 1
= k wk + wk−2 + 2kvk sk + [w1 + 2v1 ] s for s > 0.
k=2

Hence we find
w1 = −2v1 = 0,
and
k 2 wk + wk−2 + 2kvk = 0 for k ≥ 2,
i.e.
1
wk = − [wk−2 + 2kvk ] for k ≥ 2.
k2
By using v2 −1 = 0 for  ∈ N we then obtain w2 −1 = 0 for  ∈ N and

1 0 1 1 1 (−1)
w2 = − w2( −1) + 4v2 = − w2( −1) − .
42 42  4 (!)2

When choosing w0 = 0 we find by induction

(−1) +1
1

w2 = for  ∈ N.
4 (!)2 j=1 j

Hence we have

V2 (s) = J0 (s) ln s + W (s)

# $ ⎛ ⎞

∞ ∞



(−1) 2 ⎝ 1
⎠ (−1) s2 .
= 1+ s ln s −
4 (!)2 j=1
j 4 (!)2
=1 =1

Instead of V2 (s) we will use a linear combination of V1 (s) and V2 (s) to define
a second solution of the fundamental system, in particular we introduce the
second kind Bessel function of order zero,

Y0 (s) = [ln 2 − γ] J0 (s) − V2 (s)

where
 5.5 Exercises 109
⎡ ⎤

n
1
γ = lim ⎣ − ln n⎦ ≈ 0.57721566490 . . .
n→∞
j=1
j

is the Euler–Mascheroni constant. Note that Y0 (s) behaves like − ln s as s → 0.

The fundamental solution of the Helmholtz equation is then given by
1
Uk∗ (x, y) = Y0 (k |x − y|) for x, y ∈ R2 . (5.20)
2π

5.5 Exercises
5.1 Consider the recursion
1 1 (−1)
w0 = 0, w2 = − w2( −1) − for  ∈ N.
42  4 (!)2

Prove by induction that

(−1) +1
1

w2 = for  ∈ N.
4 (!)2 j=1 j

5.2 Compute the Green function G(x, y) such that

1
u(x) = G(x, y)f (y)dy for x ∈ (0, 1)
0

is the unique solution of the Dirichlet boundary value problem

−u (x) = f (x) for x ∈ (0, 1), u(0) = u(1) = 0.