(9783110409475 - Mathematics For The Physical Sciences) 12 Non-Homogeneous Boundary Value Problems Greens Functions

12 Non-Homogeneous Boundary Value Problems:
Green’s Functions
12.1 Ordinary Differential Equations
12.1.1 Definition of a Green’s Function

Suppose that we wish to solve the non-homogeneous DE

d du
Lu(x) ≡ p(x) − q(x)u(x) = f (x) (12.1.1)
dx dx
in the interval a ≤ x ≤ b with f (x) a known function and with non-homogeneous
boundary conditions
du

α1 u(a) + α2 = α3
dx x=a

du

β1 u(b) + β2 = β3 (12.1.2)
dx x=b

for given values of the constants α1 , α2 , α3 , β1 , β2 , and β3 .

We recognize that L is the Sturm Liouville differential operator and so we know
that it satisfies the generalized green’s identity. This means that if u(x) and v(x) are
any twice- differentiable functions defined on a ≤ x ≤ b,
Zb x=b
du dv
[v(x)Lu(x) − u(x)Lv(x)]dx = p(x) v(x) − u(x) . (12.1.3)
dx dx x=a
a
Let us apply this to functions of our choosing. First we choose u(x) to be the solution
of our non-homogeneous boundary value problem. Then, we choose v(x) to be the
function G(x; x0 ) where
LG(x; x0 ) = δ(x − x0 ). (12.1.4)
With these choices, the identity (12.1.3) gives us

Zb x=b
du dG
[G(x; x0 )f (x) − u(x)δ(x − x0 )]dx = p(x) G(x; x0 ) − u(x) .
dx dx x=a
a
The second term on the left hand side is just −u(x0 ). Therefore, we are going to inter-
change x and x0 , x ↔ x0 , and then rearrange all the terms to yield an expression for
u(x):
Zb x0 =b
0 0 0 0 d0 0 0 d 0
u(x) = G(x ; x)f (x )dx − p(x ) G(x ; x) 0 u(x ) − u(x ) 0 G(x ; x) .
dx dx x0 =a
a
(12.1.5)
© 2014 Leslie Copley

This work is licensed under the Creative Commons Attribution-NonCommercial-NoDerivs 3.0 License.
Unauthenticated
Download Date | 5/24/18 7:11 AM
Ordinary Differential Equations | 385
This is a very auspicious result for it tells us that if we can determine G(x; x0 ), we
can simply write down the solution of any other non-homogeneous problem involving
L. The determination of G(x; x0 ) begins with the specification of boundary conditions
to complement our knowledge of its differential equation. The “surface term” in (12.1.5)
tells us that we must impose the homogeneous counterparts to the boundary condi-
tions imposed on u(x) since otherwise we will be unable to make a full evaluation of
that term. For example, if the boundary conditions on u(x) are u(a) = α and u(b) = β,
wehave to require
G(a; x0 ) = 0 and G(b; x0 ) = 0 to eliminate the unknown quantities
du
dx and dx . Thus, the Green’s function G(x; x0 ) is the unique solution of
du
x=a x=b
LG(x; x0 ) = δ(x − x0 ) (12.1.6)
subject to
dG

α1 G(a; x0 ) + α2 =0
dx x=a

dG

β1 G(b; x0 ) + β2 = 0. (12.1.7)
dx x=b

12.1.2 Direct Construction of the Sturm Liouville Green’ s Function
The Green’s function DE (12.1.7) is as close to a homogeneous equation as a non- homo-

geneous one can be. In fact, G(x; x0 ) satisfies Lu(x) = 0 everywhere except the point
x = x0 . Therefore, we should feel reasonably optimistic about our capacity to solve it.
We start by noting some basic attributes of the solution that flow from the properties
of the DE and the boundary conditions.
Applying the generalized Green’s identity (12.1.3) to u(x) = G(x; x0 ) and v(x) =
G(x; x00 ) we find
x=b
0 00 00 0 00 d 0 0 d 00
G(x ; x ) − G(x ; x ) = p(x) G(x; x ) G(x; x ) − G(x; x ) G(x; x ) =0
dx dx x=a
because G satisfies homogeneous boundary conditions. Thus, we conclude that

G(x; x0 ) is symmetric under x ↔ x0 :
G(x; x0 ) = G(x0 ; x). (12.1.8)
As we will see again in Section 12.3.4, the boundary conditions are replaced by an
initial condition 0
G t; t = 0 for t <t0
when the independent variable is time t, −∞ < t < ∞. This condition gives expres-
sion to the causality principle: G is the response of a system to an instantaneous
0
disturbance at t = t and that response cannot precede its cause. Used in an analo-
gous application
0
ofthe
0
Green’s
identity,
0
itchanges
0
the symmetry property (12.1.8)
from G x; x = G x ; x to G t; t = G −t ; −t .
Unauthenticated
386 | Non-Homogeneous Boundary Value Problems: Green’s Functions
Next, we note that integrating the DE from x0 − ε to x0 + ε gives us
x=x0 +ε Z x0 +ε
d 0
q(x)G(x; x0 )dx = 1.

p(x) G(x; x ) −
dx 0
x=x −ε
x0 −ε
Equation (12.1.8) tells us that G(x; x0 ) is symmetric about x = x0 ,
G(x0 + ε; x0 ) = G(x0 ; x0 + ε) = G(x0 − ε; x0 ),
and therefore that it is continuous there. So is q(x). This means that when we take the
limit as ε → 0, we will have
0
x +ε
Z
lim q(x)G(x; x0 )dx = 0.
ε →0
x0 −ε
Thus, while G(x; x0 ) is continuous at x = x0 , its first derivative must have a disconti-
nuity there:
x=x0 +ε
d 1
lim G(x; x0 ) = . (12.1.9)

ε→0 dx x=x0 −ε p(x0 )
We shall now use (12.1.8) and (12.1.9) to construct G(x; x0 ) from non-trivial solu-
tions of the homogeneous DE Lu(x) = 0.
Let u < (x) be such a solution that satisfies one additional constraint, the homoge-
neous boundary condition at x = a,

d u <
α1 u < (a) + α2 = 0.
dx x=a
Since we also have

dG
α1 G(a; x0 ) + α2 =0
dx x=a
and since these two algebraic equations have to admit a non-trivial solution for at least
one of α1 and α2 , the determinant of their coefficients must vanish. In other words, we
require
dg(x; x0 )

0 du < (x)

u < (a) −G(a; x ) = 0.
dx x=a
dx x=a
But this is just the Wronskian of two solutions of the same DE and since it vanishes
at a particular point (x = a), it must be identically zero on a ≤ x < x0 and the two
solutions must be linearly dependent there. Therefore, we can write
G(x; x0 ) = c < u < (x) for a ≤ x < x0 (12.1.10)
where c < is some constant.
Unauthenticated
Similarly, if u > (x) is a non-trivial solution of Lu(x) = 0 that satisfies the homoge-
neous boundary condition

d u >
β1 u > (b) + β2 = 0,
dx x=b
then
G(x; x0 ) = c > u > (x) for x0 < x ≤ b (12.1.11)
where c > is some other constant.

Now we shall impose the continuity of G(x; x0 ) and the discontinuity of d 0
dx G(x; x )
at x = x0 :
c < u < (x0 ) = c > u > (x0 )
and
d u < d u > 1
c< − c> =− .
dx x=x0 dx x=x0 p(x0 )
Solving for c < and c > , we find
u > (x0 ) u < (x0 )
c< = 0 0
and c > =
p(x )W(x ) p(x0 )W(x0 )
where W(x) = u < (x) ddxu > − u > (x) ddxu < is the Wronskian of u < (x) and u > (x). Thus, our final
expression for the Green’s function is
u (x) u (x0 )

 < 0 > 0 for a ≤ x < x0

G(x; x0 ) = p(x )W(x ) (12.1.12)
0
 u < (x ) u > (x) for x0 < x ≤ b

p(x0 )W(x0 )
This is called the direct construction formula for determining G(x; x0 ) and it results
in a closed form expression that contains no summation or integration to perform.
The function p(x0 )W(x0 ) in the denominator is in fact a constant as can be seen from
equation (9.2.6)  x 
 Z 
W(x) = W(x0 ) exp − a(ξ )dξ
 
x0
for the Wronskian of the DE

2
d u + a(x) du + b(x)u(x) = 0.
d x2 dx
Converting
d du
p(x) − q(x)u(x) = 0
dx dx
into this other canonical form, we see that
1 dp
a(ξ ) =
p(ξ ) dξ
Unauthenticated
and so,
p(x0 )
W(x) = W(x0 ) or p(x)W(x) = a constant. (12.1.13)
p(x)
Equation (12.1.12) provides a simple prescription for determining Sturm Liouville
Green’s functions and thence for determining the solutions of non-homogeneous
Sturm Liouville problems. The latter, obtained by inserting (12.1.12) into (12.1.5), have
a form reminiscent of what we found in Chapter 9 from application of the variation
of constants approach to solving non-homogeneous ODE’s. However, the present
approach represents a major advance on variation of constants because it
– incorporates boundary conditions, and
– can be generalized to apply to PDE’s in any number of dimensions.
There is a possible complication that can arise: the Wronskian that appears in the
denominator of (12.1.12) could be zero leaving our Green’s function undefined. How-
ever, the lack of mathematical definition does not prevent interpretation. If the Wron-
skian of u < (x) and u > (x) is zero, the two functions have to be proportional to each other
and so there must exist a single solution of the homogeneous DE that satisfies both
boundary conditions. Since the function q(x) can be written q(x) = r(x) − λρ(x) for
some fixed λ and ρ(x) ≥ 0, the homogeneous DE can be converted into the eigenvalue
equation
d du
p(x) − r(x)u(x) = −λρ(x)u(x)
dx dx
and we see that the vanishing of the Wronskian implies that λ is an eigenvalue of
d d

L ≡ dx p(x) dx − r(x). As we will see in the following example, this in turn implies a
physical interpretation in terms of resonant behaviour.
12.1.3 Application: The Bowed Stretched String
Suppose that we have a stretched string that is subjected to a transverse bowing force.
If the force per unit length at position x and time t is F(x, t), the transverse displace-
ment of the string is a solution of the non-homogeneous wave equation
2 2
∂ ψ − 1 ∂ ψ = − F(x, t)
∂x 2 c ∂ t2
2 T
where T is the tension in the string. Suppose further that the bowing force is harmonic
so that F(x, t) = −Tf (x) e−iωt . The forced response of the string will then be ψ(x, t) =
u(x) e−iωt where
2
d u + 2 u(x) = f (x), k = ω .
k
d x2 c
If, as we usually do, require the ends of the string to be fixed, this non-homogeneous
ODE will be accompanied by the boundary conditions u(0) = u(L) = 0.
Unauthenticated
The Green’s function that we need for this problem is the solution of
2
d G(x; x0 ) + 2 G(x; x0 ) = δ(x − x0 )
k
d x2
subject to G(0; x0 ) = G(L; x0 ) = 0. To apply the direct construction formula we require
a solution u < (x) of
2
d u + 2 u(x) = 0
k
d x2
( )
cos kx
that satisfies u < (0) = 0. The general solution of this equation is u(x) =
sin kx
and so a solution that meets our boundary condition is
u < (x) = sin kx, 0 ≤ x.
A solution u > (x) that clearly meets the condition at the other boundary, namely u > (L) =
0, is
u > (x) = sin k(L − x), x ≤ L.
Calculating their Wronskian, we have
d u> d u>
W(x) = u < (x) − u > (x) = − sin kx k cos k(L − x) − k cos kx sin k(L − x)
dx dx
= −k sin k(x + L − x) = −k sin kL.
Therefore, since p(x) ≡ 1 in this case, the desired Green’s function is

0
 − sin kx sin k(L − x ) for 0 ≤ x < x0


0
G(x; x ) = k sin kL (12.1.14)
0
 − sin kx sin k(L − x) for x0 < x ≤ L

k sin kL
Thus, the solution to our bowed string problem is ψ(x, t) = u(x) e−iωt where
ZL
u(x) = G(x; x0 )f (x0 )dx0
0
Zx ZL
sin k(L − x) 0 0 0 sin kx
=− sin kx f (x )dx − sin k(L − x0 )f (x0 )dx0 .
k sin kL k sin kL
0 x
We can use the same Green’s function to solve a different type of non-homogeneous
string problem. Suppose that there is no bowing force, f (x) ≡ 0, but one end of the
string is vibrated with frequency ω and amplitude A. This means that we will have
boundary conditions u(0) = 0 and u(L) = A. Equation (12.1.5) still applies as does the
Green’s function (12.1.14). In fact, inserting the latter along with the new boundary
conditions on u(x), (12.1.5) gives us the solution ψ(x, t) = u(x) e−iωt with

d 0
1 sin kx
u(x) = D 0 G(x; x ) =A sin kx k cos k(L − L) = A .
dx x0 =L
k sin kL sin kL
Unauthenticated
Notice that the Wronskian of u < (x)and u > (x) provides this particular Green’s func-
tion with simple poles located at the zeros of sin kL which means at the values k = ωc =
mπ
L , m = 1, 2, . . . . As we know and as the last sub-section suggested should happen,
these values correspond to the normal modes of vibration of the string. Thus, when
we stimulate the string at values of ω approaching one of its natural frequencies, the
Green’s function and hence the transverse displacement of the string increases with-
out limit. In other words, we produce a resonant response or resonance.
12.1.4 Eigenfunction Expansions : The Bilinear Formula
The origin of resonant behaviour and even of the Green’s function method itself be-
comes much more transparent when one attempts solution by means of an eigenfunc-
tion expansion.
We shall use the modification to the non-homogeneous Sturm-Liouville equation
that was introduced at the end of Section 12.1.2. Specifically, we shall write it in the
form

d d
Lu(x) + λρ(x)u(x) = f (x) where L ≡ p(x) − r(x) (12.1.15)
dx dx
so that the correspondence with (12.1.1) is brought about by the replacement of q(x)
in that equation by r(x) − λρ(x). Here λ is a constant and ρ(x) is the (positive-definite)
weight function that is defined by the eigenvalue equation
L u m (x) = − λ m ρ(x) u m (x) (12.1.16)
and the (homogeneous) boundary conditions

d u m d u m
α1 u m (a) + α2 = 0 and β 1 u m (b) + β 2 dx = 0. (12.1.17)
dx x=a x=b
We know that the eigenfunctions {u m (x)} form a complete orthogonal set for the
space of functions that are square integrable with respect to the weight function ρ(x)
over the interval a ≤ x ≤ b. Therefore, assuming the eigenfunctions to be normalized
f (x)
to 1 over a ≤ x ≤ b, we expand u(x) and ρ(x) in terms of them and write
∞ Zb
u*m (x0 )u(x0 )ρ(x0 )dx0 ,
X
u(x) = a m u m (x) where a m = (12.1.18)
m=1 a
and
∞ Zb
u*m (x0 )f (x0 )dx0 .
X
f (x) = ρ(x) b m u m (x) where b m = (12.1.19)
m=1 a
Unauthenticated
Notice that we are allowing for the possibility that the eigenfunctions are complex.
Substituting these expansions into the DE in (12.1.15), we obtain the algebraic equation
∞
X ∞
X
ρ(x) a m (λ − λ m ) u m (x) = ρ(x) b m u m (x).
m=1 m=1
Because of the orthogonality of the {u m (x)}, this implies

b m for each m = 1, 2, 3, . . . .
am =
λ − λm
Substituting back into the series in (12.1.18) and using the definition of b m in (12.1.19),
this yields a solution to the non-homogeneous DE, namely
∞ Zb
u m (x)
u*m (x0 )f (x0 )dx0 .
X
u(x) =
λ − λm
m=1 a
Both the integral and the series should be uniformly convergent and so we interchange
their order to obtain
Zb ∞
u m (x) u*m (x0 )
G(x; x0 )f (x0 )dx0 where G(x; x0 ) =
X
u(x) = . (12.1.20)
λ − λm
a m=1
This is called the bilinear formula for the Green’s function.

Note that if f (x0 ) = δ(x0 − x0 ), (12.1.20) yields the solution u(x) = G(x; x0 ). This
verifies that G(x; x0 ) as defined by the bilinear formula is a solution of the Green’s
function DE
LG(x; x0 ) + λρ(x)G(x; x0 ) = δ(x − x0 ).
Moreover, since each eigenfunction u m (x) satisfies the homogeneous boundary con-
ditions (12.1.17), so does G(x; x0 ).
We remarked at the beginning of this sub-section that the origin of resonant be-
haviour becomes particularly transparent when the bilinear formula is used to con-
struct a Green’s function. Indeed, it corresponds to λ = λ m for some m and when that
happens G(x; x0 ) clearly becomes undefined and there is no solution to the original
non-homogeneous problem unless, by chance,
Zb
u*m (x0 )f (x0 )dx0 = 0
a
for that particular value of m.

2
The normalized eigenfunctions of the bowed string differential operator, L ≡ d ,
d x2
corresponding to the homogeneous boundary conditions u(0) = u(L) = 0, are
r
2 mπx m2 π2
u m (x) = sin with λ m = m = 1, 2, . . . .
L L L2
Unauthenticated
Therefore, substituting into (12.1.20) and using λ = k2 , we find the bilinear form
∞ 0
2 X sin mπx
L sin L
mπx
G(x; x0 ) = 2 2 (12.1.21)
L m π
k2 − 2
m=1 L
which exhibits explicitly the poles at k = mπL . Evidently, this series is the Fourier sine
series expansion of (12.1.14)
0
 − sin kx sin k(L − x ) for 0 ≤ x < x0


0
G(x; x ) = k sin kL
0
 − sin kx sin k(L − x) for x0 < x ≤ L

k sin kL
12.1.5 Application: the Infinite Stretched String
In many problems that are amenable to use of the bilinear formula, the eigenvalue
spectrum is continuous. To illustrate what happens in such a situation, we shall con-
sider a one-dimensional analogue of acoustic and electromagnetic radiation.
Suppose that we have an infinitely long stretched string that is subjected to a trans-
verse harmonic force per unit length F(x, t) = −Tf (x) e−i ω0 t where T is, as usual, the
tension in the string. Here, “infinitely long” means long enough that the ends of the
string have a negligible effect on the behaviour of points in any neighbourhood of the
middle. The PDE satisfied by the transverse displacement is the same as for the finite
string,
2 2
∂ ψ − 1 ∂ ψ = − 1 F(x, t),
∂x 2 c ∂ t2
2 T
but the boundary conditions are, of course, quite different. We shall assume that the
displacement is everywhere bounded: |ψ(x, t)| < ∞ for all −∞ < x < ∞.
As with the finite string, we shall seek solutions of the form ψ(x, t) = u(x) e−i ω0 t
which reduces the problem to one of solving the non-homogeneous ODE
2
d u + 2 u(x) = f (x) where ω0
k0 k0 =
d x2 c
subject to |u(x)| < ∞ for all −∞ < x < ∞. Therefore, the Green’s function for this
problem must be the solution of
2
d G + 2 G(x; x0 ) = δ(x − x0 ) (12.1.22)
k0
d x2
that satisfies the same (homogeneous) boundary conditions. To find it by means of the
bilinear formula, we must first solve the eigenvalue problem
2
d u (x) = −λ u (x) with | u (x)| < ∞ for all − ∞ < x < ∞. (12.1.23)
d x2 λ λ λ
Unauthenticated
The boundedness condition can only be met if λ < 0. Therefore, we set λ = − k2 and ob-
tain the (normalized) eigenfunctions u k (x) = √12π e−ikx . This means that the bilinear
formula for our Green’s function is
Z∞ 0
1 e ikx e−ikx
G(x; x0 ) = dk. (12.1.24)
2π k20 − k2
−∞
Evidently, this is an inverse Fourier transform and so must be the solution of

(12.1.22) that we would have obtained had we used Fourier transforms. To confirm
this, we set F{G(x; x0 )} ≡ g(k; x0 ), and transform (12.1.22) to obtain
Z∞
0 0 1 1 0
2
− k g(k; x ) + 2
k0 g(k; x ) = √ e ikx δ(x − x0 )dx = √ e ikx ,
2π 2π
−∞
or 0
0 1 e ikx
g(k; x ) = √ 2− 2
.
2π k0 k
Thus, since
Z∞
0 1
G(x; x ) = √ e−ikx g(k; x0 )dk,
2π
−∞
we do indeed recover the bilinear formula (12.1.24).

We have some experience in evaluating Fourier integrals and so (12.1.24) should
lead to a closed form expression for G(x; x0 ) that can be compared to the expression
obtainable from the direct construction technique. There is a slight complication how-
ever: the integrand has (simple) poles on the real axis at k = ± k0 . This means that we
require an additional piece of information that instructs us how to deform the contour
of integration to avoid them. What that information may be becomes apparent as soon
as we investigate the residues at the poles. The residues at k = ± k0 are
0
1 e±ik(x −x)
∓
4π k0
respectively. The first of these would make a contribution to G(x; x0 )e−iω0 t that con-
0
tains the factor e−i[k0 (x−x )+ω0 t] while the contribution from the second would contain
0
e i[k0 (x−x )−ω0 t] . These are waves travelling to the left and to the right, respectively, from
the source point x = x0 . But the role of the Green’s function is to give us the the re-
sponse at a point x due to a disturbance at a point x0 . Therefore, if x is to the left of
x0 , x < x0 , then G(x; x0 ) should not include the wave travelling to the right. Conse-
quently, the contour should not enclose the pole k = − k0 . On the other hand, if x is
to the right of x0 , x > x0 , the wave travelling to the left must be excluded and so now
the contour should not enclose k = k0 . Thus, the physical identity of the Green’s
function provides a key piece of information. We shall now complement it with the re-
quirements of Jordan’s Lemma to come up with a unique prescription for the contour
Unauthenticated
of integration. If x < x0 , we are obliged to close the contour in the upper half plane
and so we avoid the poles by means of small semi-circle closing above k = − k0 and a
second semi-circle closing below k = k0 . If x > x0 , the contour is closed in the lower
half-plane and so the poles are avoided by exactly the same means.
The evaluation of the Fourier integral is now straightforward: one finds
" 0
#
0 1 e ik(x −x) i 0
G(x; x ) = 2πi − =− e i k0 (x −x) for x < x0 ,
2π k + k0 2 k0
k=k0
and " #
0
0 1 e ik(x −x) i 0
G(x; x ) = 2πi =− e−i k0 (x −x) for x > x0 ,
2π k − k0 2 k0
k=− k0
or
i 0
G(x; x0 ) = − e i k0 |x−x | for all − ∞ < x < ∞. (12.1.25)
2 k0
Therefore, the solution to the infinite string problem is ψ(x, t) = u(x) e−i ω0 t where
Z∞ Z∞
 x 
Z
i 0 0

u(x) = G(x; x0 )f (x0 )dx0 = − e i k0 (x−x ) f (x0 )dx0 + e i k0 (x −x) f (x0 )dx0 .
2 k0  
−∞ −∞ x
(12.1.26)
Notice that if we had chosen the time dependence to be e i ω0 t , we would have had
the reverse correspondence between the waves travelling to the left and right. In that
case, the Green’s function to be used is the complex( conjugate )of the one in (12.1.25)
cos ω0 t
If the applied force is real, F(x, t) = −Tf (x) , f * (x) = f (x), we can
sin ω0 t
express the transverse displacement in an explicitly real form by setting it equal to
the real or imaginary parts, respectively of u(x) e−i ω0 t (the real part yielding an even
function of t and the imaginary part an odd function of t to match the parity of cos ω0 t
and sin ω0 t). In the first case, this yields
Z∞ Z∞
n
0
o
0 10
ψ(x, t) = Re e −i ω0 t
G(x; x ) f (x )dx = sin(k0 |x − x0 | − ω0 t)f (x0 )dx0
2 k0
−∞ −∞
(12.1.27)
and in the second,

Z∞ Z∞
n
0
o
0 0 1
ψ(x, t) = Im e −i ω0 t
G(x; x ) f (x )dx = − cos(|x − x0 | − ω0 t)f (x0 )dx0 .
2 k0
−∞ −∞
(12.1.28)
Unauthenticated
Partial Differential Equations | 395
We shall complete our analysis of the infinite stretched string by constructing the
Green’s function directly. The relevant homogeneous DE is
2
d u + 2 u(x) = 0.
k0
d x2
We require a solution u < (x) which meets the boundary condition that ( as x → −∞ )
−i ω0 t e i k0 x
u < (x) e is a wave travelling to the left. Since the general solution is ,
e−i k0 x
this means that u < (x) = e−i k0 x . Next, we seek a solution u > (x) such that as x → +∞
u > (x) e−i ω0 t is a wave travelling to the right. The obvious choice is u > (x) = e i k0 x .
The Wronskian of u < (x) and u > (x) is
d u> d u>
W(x) = u < (x) − u < (x) = i2 k0 .
dx dx
Thus, since p(x) ≡ 1, equation (12.1.12) yields
0
(
0 i e i k0 (x−x ) for − ∞ < x < x0
G(x; x ) = − 0
2 k0 e i k0 (x −x) for x0 < x < ∞
or,
i 0
G(x; x0 ) = − e i k0 |x−x | for all − ∞ < x < ∞.
2 k0
in full agreement with (12.1.25).
12.2 Partial Differential Equations
12.2.1 Green’s Theorem and Its Consequences
In more than one dimension a non-homogeneous boundary value problem generally

involves the solution of a PDE
Lψ(r) = f (r) with L ≡ ∇·[p(r)∇] + s(r) (12.2.1)
inside a volume V that is bounded by a surface S on which either ψ(r) or n·∇ψ is spec-
ified. The partial differential operator L in (12.2.1) is self-adjoint and satisfies Green’s
Theorem which states that if u(r) and v(r) are any two twice-differentiable functions,
Z Z
[u(r)Lv(r) − v(r)Lu(r)]dV = p(r)[u(r)∇v(r) − v(r)∇u(r)] · dS. (12.2.2)
V s
The proof of the theorem follows from a consideration of the integrals
Unauthenticated
Z Z
[u(r)p(r)∇v(r)] · dS= ∇·[u(r)p(r)∇v(r)]dV
S V
Z Z
= (∇u(r)) · (p(r)∇v(r))dV+ u(r)∇·(p(r)∇)v(r)dV
V V
and
Z Z
[v(r)p(r)∇u(r)] · dS= ∇·[v(r)p(r)∇u(r)]dV
S V
Z Z
= (∇v(r)) · (p(r)∇u(r))dV+ v(r)∇·(p(r)∇)u(r)dV
V V
where we have used the divergence theorem in the first line of both equations. Sub-
tracting the second from the first of these equations, there is a cancellation that gives
us
Z Z
p(r)[u(r)∇v(r) − v(r)∇u(r)] · dS= [u(r)∇·(p(r)∇)v(r) − v(r)∇·(p(r)∇)u(r)]dV .
S V
Adding u(r)s(r)v(r) − v(r)s(r)u(r) to the integrand on the left hand side completes the
derivation of (12.2.2).
Let us introduce a Green’s function G(r;r 0 ) by defining it to be a solution of
LG(r;r 0 ) =δ(r − r 0 ) (12.2.3)
in V subject to suitable boundary conditions on S. Applying Green’s Theorem with u(r)

replaced by ψ(r) (the solution of (12.2.1)) and v(r) by G(r;r 0 ) and using their respective
PDE’s, we find
Z Z
[ψ(r)LG(r;r 0 ) − G(r;r 0 )Lψ(r)]dV= [ψ(r)δ(r − r 0 ) − G(r;r 0 )f (r)]dV
V V
Z Z
=ψ(r 0 ) − G(r;r 0 )f (r)dV= p(r)[ψ(r)∇G(r;r 0 ) − G(r;r 0 )∇ψ(r)] · dS.
V S
Interchanging r and r 0 and rearranging terms, the last two lines of this equation be-
come
Z Z
ψ(r) = G(r 0 ;r)f (r 0 )dV 0 + p(r 0 )[ψ(r 0 )∇0 G(r 0 ;r) − G(r 0 ;r)∇0 ψ(r 0 )] · dS0 (12.2.4)
V S
Unauthenticated
which is the multi-dimensional analogue of equation (12.1.5).

We now choose boundary conditions for G(r;r 0 ) that will eliminate unknown
quantities from the surface integral on the right hand side of (10.2.4). Normally, there
are only two cases to consider.
Case 1 (Dirichlet Boundary Conditions): ψ(r) is given on S.

The obvious choice for the Green’s function under this circumstance is the homoge-
neous condition
G(r;r 0 ) = 0 for r on S. (12.2.5)
Equation (12.2.4) then becomes

Z Z
ψ(r) = G(r 0 ;r)f (r 0 )dV 0 + p(r 0 )ψ(r 0 )∇0 G(r 0 ;r) · dS0 . (12.2.6)
V S
Case 2 (Neumann Boundary Conditions): n·∇ψ(r) is given on S.

The choice here is not quite so obvious. If we apply the divergence theorem to
Z Z
LG(r;r 0 )dV = δ(r − r 0 )dV = 1,
V V
we find Z Z
p(r)∇G(r;r 0 ) · dS+ s(r)G(r;r 0 )dV = 1.
S V
This means that if s(r) ≡ 0, we cannot require n·∇G(r;r 0 ) = 0 for r on S since that
would produce a contradiction. Therefore, we are obliged to recognize two sub-cases:
if s(r) ≠ 0, we make the obvious choice and impose the homogeneous condition
n·∇G(r;r 0 ) = 0 for r on S (12.2.7)
and if s(r) ≡ 0, we require the next best thing,

Z
1
n·∇G(r;r 0 ) = , where A p = p(r)dS, for r on S. (12.2.8)
Ap
S
In the first instance, (12.2.4) becomes

Z Z
ψ(r) = G(r 0 ;r)f (r 0 )dV 0 − p(r 0 )G(r 0 ;r)∇0 ψ(r 0 ) · dS0 (12.2.9)
V S
Unauthenticated
and in the second,

Z Z
ψ(r) =hψ iS + G(r 0 ;r)f (r 0 )dV 0 − G(r 0 ;r)∇0 ψ(r0 ) · dS0 (12.2.10)
V S
1
p(r 0 )ψ(r 0 )dS0 is the weighted average of ψ(r) over the whole sur-
R
where hψ iS = Ap
S
face S.
Poisson’s equation is an important example of a PDE for which s(r) ≡ 0 and whose
Green’s function must therefore meet the non-homogeneous Neumann boundary con-
dition (12.2.8). In that case, hψ iS = A1 ψ(r 0 )dS0 where A is the area of the surface S.
R
S
This means that ψ(r) is determined only to within an additive constant by the bound-
ary condition. On the other hand, we know from electromagnetic theory that the def-
inition of zero potential is arbitrary and exercising that arbitrariness, we can set hψ iS
to zero.
Now that we know how to solve for ψ(r) in terms of G(r;r 0 ), it is time to turn
our attention to the construction of Green’s functions in more than one dimension.
We will do so by considering specific PDE’s and boundary conditions. But first, we
shall deduce a property common to all Green’s functions. What is involved is another
application of Green’s Theorem. Setting u(r) = G(r;r 0 ) and v(r) = G(r;r 00 ) in equation
(12.2.2), we note that the surface term vanishes and leaves us with the result G(r 0 ;r 00 ) −
G(r 00 ;r 0 ) = 0 or,
G(r 0 ;r 00 ) = G(r 00 ;r 0 ). (12.2.11)
In other words, G(r;r 0 ) is symmetric under r ↔ r 0 .
12.2.2 Poisson’s Equation in Two Dimensions and With Rectangular Symmetry
Suppose that we wish to find the static deflection u(x, y) of a rectangular membrane
due to an external force. Using f (x, y) to denote the external force per unit area divided
by the tension, this will require that we solve
2 2
∂ u + ∂ u = f (x, y) (12.2.12)
∂x 2 ∂ y2
subject to (homogeneous) Dirichlet conditions at the fixed edges which we shall locate
at x = 0, x = a, y = 0 and y = b.
The Green’s function for this problem is the solution of
2 2
∂ G + ∂ G = δ(x − x0 )δ(y − y0 ) (12.2.13)
∂x 2 ∂ y2
that satisfies
G(0, y; x0 , y0 ) = G(a, y; x0 , y0 ) = G(x, 0; x0 , y0 ) = G(x, b; x0 , y0 ) = 0. (12.2.14)
Unauthenticated
In general, it is either not possible or not useful to find a closed form expression
for a multi-dimensional Green’s function when the boundaries are finite closed sur-
faces. However, as in the one-dimensional case, there are two construction methods:
the eigenfunction expansion and the direct construction approach. The first of these
yields an expression, the bilinear formula, with as many summations or integrations
as there are dimensions (or separable differential operators in L) . The second elimi-
nates one of these summations by making use of our ability to directly construct closed
form expressions for one dimensional Green’s functions.
Eigenfunction Expansion Method:

Let u λ (x, y) denote the normalized eigenfunctions of ∇2 that satisfy homogeneous
Dirichlet conditions at x = 0, x = a, y = 0andy = b.
In other words, let
2
∇ u λ (x, y) = −λ u λ (x, y) with || u λ (x, y)|| = 1
and u λ (0, y) = u λ (a, y) = u λ (x, 0) = u λ (x, b) = 0. (12.2.15)
Since their closure relation must be
u λ (x, y) u*λ (x0 , y0 ) = δ(x − x0 )δ(y − y0 ),

X
(12.2.16)
λ
it is clear that the bilinear formula of Section 12.1.3 applies here too and yields
X u λ (x, y) u* (x0 , y0 )
G(x, y; x0 , y0 ) = λ
. (12.2.17)
−λ
λ
We know already (from Section 10.8) that the eigenvalues for the rectangular mem-
2 2 2 2
brane are λ m,n = ma2π + n b2π , m, n = 1, 2, 3, . . . corresponding to the (normalized)
nπy
eigenfunctions u m,n (x, y) = √2 sin mπx
a sin b . Therefore, the bilinear formula rep-
ab
resentation of our green’s function is
∞ ∞ mπx nπy mπx0 nπy0
0 0 4 X X sin a sin b sin a sin b
G(x, y; x , y ) = − m2 π2 n2 π2
. (12.2.18)
ab 2 + 2
m=1 n=1 a b
Direct Construction Method:

We start with a partial eigenfunction expansion. Choosing to do so in the y variable,
we write
r ∞
0 0 2X 0 0 nπy
G(x, y; x , y ) = G n (x; x , y ) sin (12.2.19)
b b
n=1
Unauthenticated
q
2 nπy
and invoke closure for the normalized eigenfunctions b sin b ,
∞
2X nπy nπy0
δ(y − y0 ) = sin sin .
b b b
n=1
Substituting these two expansions into the Green’s function PDE and equating on a
term by term basis, we conclude that G n (x; x0 , y0 ) factors according to
r
2 nπy0 2g
n2 π2
0 0
G n (x; x , y ) = sin g n (x; x0 ) where d 2n − g n (x; x0 ) = δ(x − x0 ).
b b dx b2
(12.2.20)
This is a one-dimensional Green’s function DE which we can solve by the direct con-
struction method.
Remembering that we have homogeneous boundary conditions, g n (0; x0 ) =
g n (a; x0 ) = 0, we proceed by seeking a solution u < (x) of the homogeneous DE
2 2 2
d u − n π u(x) = 0
2
dx b2
( )
cosh nπx
b
that satisfies u < (0) = 0. Since the general solution is u(x) = , the sim-
sinh nπx
b
plest choice is u < (x) = sinh nπx
b .
Next, we need a solution u > (x) of the homogeneous DE that satisfies u > (a) = 0.
Again, the simplest choice is pretty obvious: u > (x) = sinh nπ
b (x − a).
The Wronskian of u < (x)and u > (x) is
nπ nπa
W(x) = u < (x) u0> (x) − u > (x) u0< (x) = sinh .
b b
Therefore, using (12.1.12) we have
nπx nπ 0
b sinh b sinh b (a − x )

 − , 0 ≤ x < x0
sinh nπa

 nπ
0 b
g n (x; x ) = nπx0 (12.2.21)
sinh nπ
b (a − x) sinh b
 − b , x0 < x ≤ a


nπ sinh nπa
b
Substituting back into the expansion (12.2.19) for G(x, y; x0 , y0 ) we conclude that
∞ 2 sinh nπx sinh nπ (a − x 0 )

nπy0
b b
sin nπy
P
 − b sin b

nπa

nπ sinh
G(x, y; x0 , y0 ) = n=1 b
∞ 2 sinh nπ (a − x) sinh nπx
0 (12.2.22)

 −
P b b nπy nπy0
 sin sin
n=1 nπ sinh nπa
b
b b
(
0 ≤ x < x0
for , respectively. This is equivalent to the double Fourier representation
x0 < x ≤ a
(12.2.18) found by application of the bilinear formula but with the sum over m actually
performed.
Unauthenticated
Had our starting point been substitution of

∞
2X mπx mπx0
δ(x − x0 ) = sin sin ,
a a a
m=1
and r ∞
0 0 2X 0 0 mπx
G(x, y; x , y ) = G m (y; x , y ) sin
a a
m=1
mπx0
into the PDE (12.2.13), we would have found G m (y; x0 , y0 ) = g m (y; y0 ) sin a where
mπy 0
a sinh a sinh mπ

a (b − y )
− , 0 ≤ y < y0


 mπ sinh mπb

0 a
g m (y; y ) = 0 (12.2.23)
a sinh mπ (b − y) sinh mπy
a a
, y0 < y ≤ b

 −

mπ sinh mπb

a
This, of course, yields an expression for G(x, y; x0 , y0 ) that is equivalent to summing

over n in the double Fourier series (12.2.18). It could have been obtained directly from
(12.2.22) by invoking the problem’s symmetry under x ↔ y, (x0 ↔ y0 ), and a ↔ b.
12.2.3 Potential Problems in Three Dimensions and the Method of Images
Coulomb’s Law is an implicit expression of the solution of the Green’s function PDE
2 0 0
∇ G(r;r ) =δ(r − r ) (12.2.24)
plus the (Dirichlet) boundary condition lim G(r;r 0 ) = 0. It tells us that the potential
r→∞
due to unit charge located at the point r=r 0 is
1 1
ψ(r) = −
4π ε0 |r − r 0 |
and, since the charge density associated with the charge is ρ(r) =δ(r − r 0 ), this means
that
ρ(r) 1
2
∇ ψ(r) = − = − δ(r − r 0 ).
ε0 ε0
Comparing this with (12.2.24), we deduce that
1 1
G(r;r 0 ) = − . (12.2.25)
4π |r − r 0 |
This is confirmed by direct integration of (12.2.24). Integrating over a spherical

volume centred at r=r 0 and applying the divergence theorem, we find
Z
∇G(r;r 0 ) · dS= 1
S
Unauthenticated
where S denotes the boundary surface of the sphere. But the normal gradient on a
spherical surface is just the partial derivative with respect to the radial coordinate.
Therefore, we can rewrite this last equation as
Z Z2π Zπ
∂G ∂G 2
dS = r sin θdθdφ = 1
∂r ∂r
S 0 0
where r = |r − r 0 |. With the origin of coordinates at r=r 0 , the delta function in the PDE
and the boundary condition to be imposed on its solution both depend only on the
radial variable and so the same must be true of the solution itself. Thus, the integration
can be performed to give us
dG 1 1 1
= = .
dr 4π r2 4π |r − r 0 |2
Integrating once more and using the boundary condition to dispose of the integration
constant, we obtain, as expected,
1 1
G(r;r 0 ) = − .
4π |r − r 0 |
This result can also be obtained by use of Fourier transforms as will be demonstrated
in Section 12.3.3.
If the Dirichlet condition is imposed on a finite surface, we can still solve for the
Green’s function in closed form by using a trick that is called, in electrostatic theory,
the method of images. Any Green’s function can be set equal to a superposition of
solutions of the non-homogeneous and homogeneous PDE’s. In the present case, this
means that we can set
1 1
G(r;r 0 ) = G s (r;r 0 )+ G o (r;r 0 ) where G s (r;r 0 ) = − (12.2.26)
4π |r − r 0 |
and
2 0 0 0

∇ G o (r;r ) = 0 with [G s (r;r )+ G o (r;r )]r on surface
= 0. (12.2.27)
For specificity, let us take the boundary surface to be a sphere of radius R. If r is con-
strained to vary inside the spherical volume, r < R, and r 00 is a point outside, r00 > R,
the delta function δ(r − r 00 ) is zero and |r−r1 00 | is a solution of the homogeneous PDE
there. This means we can set
0 1 k
G o (r;r ) = − , r ≤ R and r00 > R,
4π |r − r 00 |
where k is a constant and r 00 is chosen to lie along the same radius vector as r 0 . The
values of k and r00 are to be determined by imposing the boundary condition (12.2.27).
00
Since r 00 = rr0 r 0 , our Green’s function is
" #
0 0 0 1 1 k
G(r; r ) = G s (r; r ) + G0 (r; r ) = − + .
4π |r − r 0 | r − r000 r 0
r
Unauthenticated
This vanishes at r = R if
R
2
r00
r00 = 0
and k = −
r R
which gives us the unique solution
 
1  1 R 1
G(r; r 0 ) = − +  , r and r0 ≤ R. (12.2.28)
4π |r − r 0 | r0 r − rR2 r 0
r02
Visual inspection of this result reveals why its construction is called the method
of images. We know that G s (r;r 0 ) has the physical significance of the potential due to
an isolated point charge located at r=r 0 . Similarly, the Green’s function we are trying
to find is the potential due to that same charge when it is enclosed within a conducting
sphere of radius R. What we have found is that the effect of the sphere is the same as
that of adding an “image” point charge located at the inverse of r=r 0 with respect to
the spherical surface.
This approach works well for any simple boundary surface, a plane and a cylin-
der of infinite length being two other examples. However, finding image points is a
challenge with more complicated surfaces and is generally not worth the effort. Even
(12.2.28) is difficult to work with in the context of finding the potential due to a contin-
uous charge distribution via (12.2.6). A more fruitful approach is to proceed as we did
in Section 12.2.2 and expand the Green’s function in series (or integrals).
12.2.4 Expansion of the Dirichlet Green’s Function for Poisson’s Equation When
There Is Spherical Symmetry
Suppose that we have Dirichlet conditions imposed on a surface consisting of two con-
centric spheres of radii r = a and r = b, b > a. This means that we wish to solve

∇2 G(r; r 0 ) = δ(r − r 0 ) subject to G(r; r 0 ) = G(r; r 0 ) = 0. (12.2.29)

r=a r=b
In spherical coordinates, the delta function in the PDE can be expanded according
to
1
δ(r − r 0 ) = δ(r − r0 )δ( cos θ − cos θ0 )δ(φ − φ0 )
r2
∞ X l
1
= 2 δ(r − r0 ) (Y lm (θ0 , φ) )* Y lm (θ, φ)
X
(12.2.30)
r
l=0 m=−l
where we have used the closure relation for spherical harmonics. Similarly, we can
write
∞ X
l
G(r;r 0 ) = 0
X m
G lm (r;r ) Y l (θ, φ) (12.2.31)
l=0 m=−l
Unauthenticated
and substitute both expansions into the PDE of (12.2.29) to obtain

∞ X
l
1 d2

X 0 l(l + 1) 0 m
(r G lm (r; r ) − G lm (r;r ) Y l (θ, φ)
r d r2 r2
l=0 m=−l
∞ X
l
1 0
( Y lm (θ0 , φ0 ) )* Y lm (θ, φ).
X
= δ(r − r ) (12.2.32)
r2
l=0 m=−l
Invoking the orthogonality of the spherical harmonics to set up equations on a term

by term basis, we deduce that G lm (r; r 0 ) must factor according to
0 0 m 0 0
G lm (r; r ) = g l (r; r ) Y l (θ , φ ) (12.2.33)
where
2
d g (r; r0 ) + 2r d g (r; r0 ) − l(l + 1) g (r; r0 ) = δ(r − r 0 ).
r2 (12.2.34)
d r2 l dr l l
The homogeneous counterpart of (12.2.34) is Euler’s equation

2
d u + 2r du − l(l + 1)u(r) = 0
r2
d r2 dr
( )
rl
which has the general solution . Therefore, a solution u < (r) that satisfies
r−l−1
the boundary condition u < (a) = 0 is
2l+1

l a
u < (r) = r − l+1 , a ≤ r
r
while one that satisfies u > (b) = 0 is
rl

1
u > (r) = − , r ≤ b.
r l+1 b2l+1
The Wronskian of u < (r) and u > (r) is

2l + 1 a 2l+1
W(r) = u < (r) u0> (r) − u > (r) u0< (r) = −1
r2 b
and the Sturm-Liouville function p(r) is r2 . Thus, from the direct construction formula
(12.1.12) for one-dimensional Green’s functions, we have
r 0l

2l+1

l a 1
 r − l+1 − , a ≤ r < r0


0 (−1)
 r r0l+1 b2l+1
g l (r; r ) = h 2l+1 i 
rl a2l+1

(2l + 1) 1 − ba 1
− r − 0l+1 , r0 < r ≤ b
0l


r l+1 b2l+1 r

or,
a2l+1 r l>

(−1) 1
g l (r; r0 ) = r l
> − − (12.2.35)
b2l+1
h i
r l+1 r l+1
2l+1
(2l + 1) 1 − ba < >
Unauthenticated
where r < ≡ the smaller of r and r0 and r > ≡ the larger of r and r0 . Substituting this into
(12.2.33) and the latter into (12.2.31) gives us a final expression for the Poisson equation
Green’s function for a spherical shell bounded by r = a and r = b:
∞ X l m m 0 0 *
Y l (θ, φ)( Y l (θ , φ ) ) a2l+1

1
G(r;r 0 ) = −
X l
h i r < − . (12.2.36)
a 2l+1 r l+1 r l+1

l=0 m=−l (2l + 1) 1 − b < >
There are three special cases:

1. if a = 0 and b → ∞, we have
∞ X
l
( − 1) r l< m 0 * 1 1
G(r;r 0 ) = 0
X m
Y l (θ, φ)( Y l (θ , φ ) ) = − (12.2.37)
2l + 1 r l+1
> 4π | r − r0 |
l=0 m=−l
where the last equality was derived earlier as an application of the addition theo-
rem of spherical harmonics;
2. if a remains finite and b → ∞, we have an exterior problem and the Green’s
function is
∞ X l
a2l+1

0
X ( − 1) 1 l m m 0 0 *
G(r;r ) = r < − l+1 Y l (θ, φ)( Y l (θ , φ ) ) ; (12.2.38)
2l + 1 r2l+1
> r <
l=0 m=−l
3. if a = 0 and b remains finite, we have an interior problem and the Green’s func-
tion is
∞ Xl
0
X ( − 1) l 1 m m 0 0 *
G(r;r ) = r Y l (θ, φ)( Y l (θ , φ ) ) . (12.2.39)
2l + 1 < r > l+1
l=0 m=−l
12.2.5 Applications
Solution of Laplace’s Equation

We know already that the potential inside a sphere of radius b with no charges present
but subject to ψ(b, θ, φ) = V(θ, φ) is
∞ X
X l
ψ(r, θ, φ) = c lm r l Y lm (θ, φ)
l=0 m=−l
with
Z2π Z1
1
c lm = l
(Y lm (θ0 , φ0 )* V(θ0 , φ0 )d(cos θ0 )dφ0 .
b
0 −1
What we wish to verify now is that the Green’s function we have just derived yields
exactly the same solution. To do so, we require the normal gradient

∂

0 0

n · ∇G(r; r ) = G(r; r ) .
r=b ∂r r=b
Unauthenticated
From (12.2.39) we have

∞ l
∂ ∂ 1 X X r l m 0 0 * m
G(r; r 0 ) = G(r; r 0 )

= 2 ( Y l (θ , φ ) ) Y l (θ, φ).
∂r r=b
∂ r> r > =b b
b
l=0 m=−l
Thus, since ρ(r) ≡ 0 and dS0 = b2 d(cos θ0 )dφ0 , we obtain from equation (12.2.9)
Z Z
1
ψ(r) = − G(r;r 0 )ρ(r 0 )dV 0 + ψ(r 0 )n·∇0 G(r;r 0 )dS0
ε0
V S
Z2π Z1
 
∞ l r l
V(θ0 , φ0 )( Y lm (θ0 , φ0 ) )* d( cos θ0 )dφ0 
X X m
=  Y l (θ, φ)
b
l=0 m=−l 0 −1
which is identical to our earlier result as required.

Note that if V(θ, φ) is independent of φ (that is, if we have azimuthal symmetry),
only the m = 0 terms are retained in G(r;r 0 ). Thus, since
m m 0 0 * 2l + 1 0
Y l (θ, φ)(Y l (θ , φ ) ) → P l (cos θ) P l (cos θ ) for m = 0,
4π
the electrostatic potential becomes
Z1
 
∞ r l
 2l + 1 V(θ0 ) P l (cos θ0 )d(cos θ0 )
X
ψ(r, θ) = P l (cos θ).
2 b
l=0 −1
Solution of Poisson’s Equation

Consider a hollow, grounded sphere of radius b with a concentric ring of charge, of
radius c < b and total charge Q, inside it. Taking the ring to lie in the xy-plane, we can
assert that the charge density inside the sphere will be independent of φ and have a
delta function dependence on r and θ. Thus, we can write
ρ(r) = Aδ(r − c)δ( cos θ) where A is a constant of proportionality.
The constant A can be determined from the normalizing condition
Zb Zπ Z2π
Q= ρ(r) r2 dr sin θdθdφ.
0 0 0
Q
Thus, A = 2π c2
. This means that the electrostatic potential inside the sphere is the
solution of
2 Q
∇ ψ(r) = − δ(r − c)δ( cos θ)
2π c2 ε0
that satisfies ψ(b, θ, φ) = 0.
Unauthenticated
The Non-Homogeneous Wave and Diffusion Equations | 407
Using the interior problem Green’s function (12.2.39) modified for azimuthal sym-
metry, we have
Z2π Z1 Zb
Q 0 0 2
ψ(r) = − δ(r − c)δ( cos θ ) G(r;r 0 )r 0 dr0 d( cos θ0 )dφ0
2π c2 ε0
0 −1 0
and so,
∞ Z2πZ1Zb
P l ( cos θ) Q 0 0 1 2
r 0 dr0 d( cos θ0 )dφ0
X l
ψ(r) = δ(r − c)δ( cos θ )× r <
4π 2π c2 ε0 r l+1
>
l=0 0 −1 0
or,
∞
Q X l 1
ψ(r) = P l (0) r < P l ( cos θ)
4π ε0 r l+1
>
l=0
where r < (r > ) is now the smaller (larger) of r and c. Making use of
(−1 )l (2l)!
P2l+1 (0) = 0 and P2l (0) = 2
,
22l (l! )
this becomes
∞ l
Q X ( − 1 ) (2l)! 2l 1
ψ(r) = 2
r < P2l ( cos θ).
4π ε0 22l (l! ) r2l+1
>
l=0
Notice that for b → ∞ and r > c, this reduces to

∞ l
Q X ( − 1 ) (2l)! c2l
ψ(r) = 2 2l+1 P 2l
( cos θ)
4π ε0 22l (l! ) r
l=0
which for a point on the z-axis converges to the well-known consequence of Coulomb’s
law,
Q 1
ψ(z, 0) = √ .
4π ε0 z2 + c2
12.3 The Non-Homogeneous Wave and Diffusion Equations
12.3.1 The Non-Homogeneous Helmholtz Equation
The non-homogeneous wave equation is
2 1 ∂2 ψ(r, t)
∇ ψ(r, t) − =σ(r, t). (12.3.1)
c2 ∂ t2
As in the bowed string problem, if the source is monochromatic and harmonic,
that is if σ(r, t) =σ(r) e−iωt , we can assume the same time dependence for ψ(r, t),
Unauthenticated
ψ(r, t) =ψ(r) e−iωt . This means that the problem is effectively time-independent
and so no initial conditions are needed. We proceed by substitution into the original
PDE which yields
(∇2 + k2 )ψ(r) =σ(r), k =ω/c. (12.3.2)
From equation (12.2.4) we know that the solution of the non-homogeneous

Helmholtz equation is
Z Z
ψ(r) = G(r 0 ;r)σ(r 0 )dV 0 + [ψ(r 0 )n·∇0 G(r 0 ;r) − G(r 0 ;r)n·∇0 ψ(r 0 )]dS0 (12.3.3)
V S
where
(∇2 + k2 )G(r;r 0 ) =δ(r − r 0 ) (12.3.4)
and G(r;r 0 ) is subject to the homogeneous counterparts of whatever boundary con-

ditions are imposed on ψ(r). One expression for G(r;r 0 ) is provided by the bilinear
formula,
X u n (r)( u n (r 0 ) )*
G(r;r 0 ) = (12.3.5)
n k2 − k2n
where the functions u n (r) are the normalized normal modes defined by
(∇2 + k2 ) u n (r) = 0 with boundary conditions u n (r) = 0 or n·∇ u n (r) = 0 for r on S.
Another is provided by the same kind of partial expansion and direct construction
technique that we used for Poisson’s equation.
12.3.2 The Forced Drumhead
If the external force per unit area applied normal to a drumhead is F(r, t), its trans-
verse displacement will obey the equation
2 1 ∂2 1
∇ ψ(r, t) − ψ(r, t) = − 2 F(r, t) ≡ f (r, t) (12.3.6)
c2 ∂ t2 c µ
where, as before, µ is the mass per unit area. We assume that f (r, t) = f (r) e−iωt and
set the forced displacement of the drumhead equal to ψ(r, t) =ψ(r) e−iωt . Substituting
into (10.3.5), we get the two dimensional non-homogeneous Helmholtz equation
ω
(∇2 + k2 )ψ(r) = f (r), k= .
c
This is to be solved subject to the (Dirichlet) boundary condition ψ(r) = 0 for r on the
perimeter of the drumhead. Thus, if the drumhead is circular with radius a, we have
the condition ψ(a, θ) = 0.
Unauthenticated
With both ψ(r) and G(r;r 0 ) satisfying homogeneous Dirichlet conditions, the so-
lution is provided by the integral
Z2π Za
ψ(r) = G(r; r 0 )f (r 0 )r0 dr0 dθ0 .
0 0
Thus, all we need to do is construct the Green’s function.

Trying the direct construction method first, we expand G(r;r 0 ) in a Fourier series
in θ:
∞
G(r;r 0 ) = 0 0 imθ
X
G m (r; r ,θ ) e . (12.3.7)
m=−∞
Next, we use closure to expand the delta function in θ:

∞
1 1 1 X im(θ−θ0 )
δ(r − r 0 ) = δ(r − r0 )δ(θ − θ0 ) = δ(r − r0 ) e . (12.3.8)
r r 2π
m=−∞
Substituting these expansions into (12.3.4) expressed in terms of two dimensional po-
lar coordinates and using the orthogonality of the Fourier functions, we find that the
coefficients G m (r; r0 , θ0 ) factor according to
0 0 1 −imθ0
G m (r; r , θ ) = e g m (r; r0 ) (12.3.9)
2π
where
2
m2

d g (r; r0 ) + 1 d g (r; r0 ) + 2
g m (r; r0 ) =
1
δ(r − r0 ).
k − (12.3.10)
d r2 m r dr m r2 r
The homogeneous
( ) version of this is Bessel’s equation with general solution u(r) =
J|m| (kr)
. A solution that satisfies the boundary condition | u < (0)| < ∞ is
N|m| (kr)
u < (r) = J|m| (kr), 0 ≤ r, while one that satisfies u > (a) = 0 is u > (r) = N|m| (ka) J|m| (kr) −
J|m| (ka) N|m| (kr), r ≤ a. Their Wronskian is
W(r) = −k J|m| (ka)[J|m| (kr) N|0m| (kr) − N|m| (kr) J|0m| (kr)].
We can show from their small x behaviour that the Wronskian of the Bessel and
Neumann functions is
0 0 2
J m (x) N m (x) − J m (x) N m (x) = .
πx
2
Therefore, W(r) = − πr J|m| (ka) which with p(r) = r and equation (12.1.12) gives us
 J|m| (kr)[N|m| (ka)J|m| (kr0 ) − J|m| (ka)N|m| (kr0 )], 0 ≤ r < r0


0 π 1
g m (r; r ) = −
2 J|m| (ka) 
[N|m| (ka)J|m| (kr) − J|m| (ka)N|m| (kr)]J|m| (kr0 ), r0 < r ≤ a
Unauthenticated
Substituting back into (12.3.7), this yields the Green’s function
1 J0 (k r < )
G(r;r 0 ) = [ J (ka) N0 (k r > ) − N0 (ka) J0 (k r > )]
4 J0 (ka) 0
∞
1 X J m (k r < )
+ [ J (ka) N m (k r > ) − N m (ka) J m (k r > )] cos m(θ − θ0 )
2
m=1
J m (ka) m
(12.3.11)
where, as usual, r < (r > ) is the smaller (larger) of r and r0 .

To find the equivalent bilinear formula expression for the Green’s function is quite
straight forward since we already know what are the circular drumhead eigenfunc-
tions or normal modes. Specifically, from Section 11.3.2 we have
(
J m (k mn r) cos mθ
u mn (r, θ) = m = 0, 1, 2, . . . , n = 1, 2, . . . .
J m (k mn r) sin mθ
To normalize these functions over the area of the drumhead, we multiply them by the
normalization constants for the Bessel and Fourier functions which means multiply-
ing them by
r
1 2 1
N0n = √ for m = 0 and N mn = for m = 1, 2, . . . .
πa J00 (k0n a) π a J 0m (k mn a)
Substituting into (12.3.5), we obtain the bilinear formula

∞
1 X 1 J0 ( k0n r) J0n ( k0n r0 )
G(r;r 0 ) = 2 2
πa 0
n=1 [ J 0 ( k 0n a) ] k2 − k20n
∞ ∞
2 XX 1 J m ( k mn r) J m ( k mn r0 ) cos m(θ − θ0 )
+ 2 2
(12.3.12)
πa [ J m 0 ( k mn a) ]
m=1 n=1 k2 − k2mn
where we have used the identity cos mθ cos mθ0 + sin mθ sin mθ0 = cos m(θ − θ0 ).
12.3.3 The Non-Homogeneous Helmholtz Equation With Boundaries at Infinity
When there are no finite boundaries (the source is isolated), the boundary condition
that accompanies the Green’s function PDE
(∇2 + k2 )G(r;r 0 ) =δ(r − r 0 )
is simply that the solution be bounded and, in particular, that lim |G(r;r 0 )| <∞.
|r−r 0 |→∞
This can be solved by using Fourier transforms in much the same way that we did in
Section 12.1.5.
Unauthenticated
We start by putting a subscript on the wave number in the Green’s function PDE
to distinguish it from the transform variable:
(∇2 + k20 )G(r;r 0 ) =δ(r − r 0 ) (12.3.13)
and so the time dependence associated with G(r;r 0 ) is now e−i ω0 t , ω0 = c k0 . Taking
the three-dimensional transform of (12.3.13), we obtain
0
1 e ik·r
g(k;r 0 ) = 3/2 2 − 2
(2π ) k0 k
where k2 = k·k and
Z2π Zπ Z∞
0 0 1
g(k;r ) ≡ F{G(r;r )} = 3/2
e ik·r G(r;r 0 ) r2 dr sin θdθdφ.
(2π )
0 0 0
Thus,
Z2π Zπ Z∞ 0
0 1 e ik·(r −r) 2
G(r;r ) = 3 k dk sin θ k d θ k d φ k (12.3.14)
(2π ) k20 − k2
0 0 0
which we recognize as the bilinear formula for the Green’s function.

To evaluate this integral, we take the k3 axis in the direction of r − r 0 so that
k·(r − r 0 ) = kR cos θ k , R = |r − r0 |.
Then,
Z2π Zπ Z∞
0 1 e−ikR cos θ k 2
G(r;r ) = − 3 k dk sin θ k d θ k d φ k
(2π ) k2 − k20
0 0 0
Z∞
1 2 sin kR
=− 2
kdk
(2π ) R k2 − k20
0
Z∞
1 1 e ikR − e−ikR
=− 2 iR
kdk
(2π ) k2 − k20
0
Z∞
1 1 e ikR
=− 2
kdk.
(2π ) iR k − k20
2
−∞
The final integral can be evaluated by contour integration. When combined with time
dependence e−i ω0 t , inclusion of the pole at k = − k0 within the contour results in an
incoming wave. On the other hand, the residue of the pole at k = k0 yields an outgoing
Unauthenticated
wave. Therefore, we use the same contours that were used in Section 12.1.5 and obtain
the solutions
0
0 1 e i k0 |r−r |
Gout (r;r ) = − , (12.3.15)
4π |r − r 0 |
and
0
0 1 e−i k0 |r−r |
G in (r;r ) = − . (12.3.16)
4π |r − r 0 |
These are used to generate purely outgoing or incoming wave solutions, respectively.
Thus, for example, if the source σ(r, t) =σ(r) e−i ω0 t is an isolated loudspeaker or
acoustic antenna, the waves that it emits will be described by
0
e−i ω0 t e i k0 |r−r |
Z
ψ(r, t) = − σ(r)dV 0 .
4π |r − r0 |
all space
Far from the source, r r0 , we have |r − r 0 | ∼ r − n·r 0 where n= rr and so

0| 0
e i k0 |r−r i k0 r
·r 0
|r−r 0 |
∼e r e−ik where k0 = k0 n. In that case,
e ikr
Z
0
·r 0
ψ(r) ∼ f (k0 ) where f (k0 ) = e−ik σ(r 0 )dV 0 . (12.3.17)
r
all space
This is analogous to the formalism we introduced in Section 11.4.2 in our discussion

of spherical waves.
As one would expect, (12.3.15) and (12.3.16) both become the Green’s function
(12.2.25) for Poisson’s equation in the limit as k0 → 0.
12.3.4 General Time Dependence
If the source term σ(r, t) has a general time dependence, we cannot separate the
space and time dependence so easily and we do have to worry about initial conditions.
Let us start by re-stating the problem. It consists of solving
2 1 ∂2 ψ(r, t)
∇ ψ(r, t) − =σ(r, t) (12.3.18)
c2 ∂ t2
where the source term σ(r, t) has a general time dependence and the solution ψ(r, t)
is subject to
ψ(r, t) or n·∇ψ(r, t) given for r on S (12.3.19)
plus initial conditions

∂ψ(r, t)
ψ(r, t) and given at an initial time t = τ throughout V . (12.3.20)
∂t
Unauthenticated
To tackle this we need the Green’s function G(r, t;r 0 , t0 ) which is the solution of
0 0 1 ∂2
2
∇ G(r, t;r , t ) − G(r, t;r 0 , t0 ) =δ(r − r 0 )δ(t − t0 ) (12.3.21)
c2 ∂ t2
that satisfies the boundary condition G = 0 or n·∇G = 0 for r on S plus the initial
condition G(r, t;r 0 , t0 ) = 0 for t < t0 . The latter condition follows from the prin-
ciple of causality: G is a response to a stimulus at t = t0 and so should be zero
prior to that time. With the usual Green’s theorem manipulations, one can show that
G(r, t;r 0 , t0 ) = G(r 0 , − t0 ;r, − t) and that the solution to our original problem is
Zt Z Zt Z
0 0 0 0 0 0
ψ(r, t) = G(r, t; r , t )σ(r , t )dV dt − [G(r, t; r 0 , t0 )n · ∇0 ψ(r 0 , t0 )
τ V τ S
0 0 0 0 0 0 0
− ψ(r , t )n · ∇ G(r, t; r , t )dS dt ]
Z
1 ∂ ∂

− 2 [G(r, t; r 0 , T) 0 ψ(r 0 , t0 ) − ψ(r 0 , T) 0 G(r, t; r 0 , t0 ) ]dV 0

c ∂t t0 =τ ∂t t0 =τ
V
(12.3.22)
An analogous approach applies to the diffusion equation. To solve
2 1 ∂ψ(r, t)
∇ ψ(r, t) − =σ(r, t) (12.3.23)
κ ∂t
with ψ(r, t) or n · ∇ψ(r, t) given for r on S plus the initial condition ψ(r, t) given at
t = τ, we first determine the Green’s function G(r, t;r 0 , t0 ) which is the solution of
0 0 1 ∂
2
∇ G(r, t;r , t ) − G(r, t;r 0 , t0 ) =δ(r − r 0 )δ(t − t0 ) (12.3.24)
κ ∂t
that satisfies the boundary condition
G(r, t;r 0 , t0 ) = 0 or n · ∇G = 0 for r on S
plus the initial condition G(r, t;r 0 , t0 ) = 0 for t < t0 .

One can again show that G(r, t;r 0 , t0 ) = G(r 0 , − t0 ;r, − t) and
Zt Z
ψ(r, t) = G(r, t;r 0 , t0 )σ(r 0 , t0 )dV 0 dt0
τ V
Zt Z
− [G(r, t;r 0 , t0 )n · ∇0 ψ(r 0 , t0 ) − ψ(r 0 , t0 )n · ∇0 G(r, t;r 0 , t0 )]dS0 dt0
τ S
Z
1
− G(r, t;r 0 ,τ)ψ(r 0 ,τ)dV 0 . (12.3.25)
κ
V
Unauthenticated
12.3.5 The Wave and Diffusion Equation Green’s Functions for Boundaries at
Infinity
The Green’s function associated with the wave equation PDE and Dirichlet boundary
conditions at infinity is a solution of
1 ∂2

∇
2
− G(r, t;r 0 , t0 ) =δ(r − r 0 )δ(t − t0 ) (12.3.26)
c2 ∂ t2
that is everywhere bounded as a function of both r and t. The solution by means of

Fourier transforms proceeds exactly as in Section 12.3.3 and yields
Z∞ 0
c2 e−iω(t−t )
Z
0
0 0
G(r, t;r , t ) = − 4
e−ik·(r−r ) 2
k dk sin θ k d θ k d φ k dω.
(2π ) c k2 − ω2
2
−∞ all space
(12.3.27)
As we saw in Chapter 3, the integral

Z∞ 0
e−iω(t−t )
∆≡ dω (12.3.28)
c2 k2 − ω2
−∞
has four different values depending on how one avoids the poles at ω = ±ck. However,
only one of these four satisfies our (causal) initial condition G(r, t;r 0 , t0 ) = 0 for t < t0 .
This is the so-called retarded solution
 0
 2π sin ck(t − t ) for t > t0
∆ ret = ck (12.3.29)
0 for t < t0

which arises from deforming the contour of integration to pass above both poles. The
name is a reflection of the correspondence to a signal emitted at a time t0 that is earlier
or retarded compared to the time of arrival t.
Inserting (12.3.29) into (12.3.27) yields the (retarded) Green’s function
0

 − c −ik·(r−r 0 ) sin ck(t − t ) 2
R
e k dkd( cos θ k )d φ k

G(r, t;r 0 , t0 ) = 2π all space k (12.3.30)
0


for t > t0 and t < t0 , respectively. Choosing the k3 axis to be in the direction of r − r 0 ,
the angular integration is easy to perform and gives us
Z∞
sin ck(t − t0 ) 2
Z
−ik·(r−r 0 ) 4π
e k dkd( cos θ k )d φ k = sin kR sin ckTdk
k R
all space 0
Unauthenticated
where R = |r − r 0 | and T = t − t0 . Converting the sines on the right hand side into
exponentials, we obtain
Z∞ Z∞
1
sin kR sin ckTdk = − [e ik(R+cT) + e−ik(R+cT) − e ik(R−cT) − e−ik(R−cT) ]dk
4
0 0
Z∞
1
=− [e ik(R+cT) − e ik(R−cT) ]dk
4
−∞
1
= [δ(R − cT) − δ(R + cT)].
4
Since the second delta function does not contribute for T > 0 and since

1 R
δ(R − cT) = δ −T ,
c c
substitution back into (12.3.30) gives us

0

 − 1 δ |r − r | − (t − t0 )

for t > t0
G(r, t;r 0 , t0 ) = 4π c (12.3.31)
0

 0 for t < t
as our final result for the (retarded) Green’s function.

Given the unusual appearance of this Green’s function, it is helpful to keep in
mind its physical interpretation. It is a spherical wave produced by an instantaneous
disturbance at the single point r=r 0 and time t = t0 . Prior to t = t0 , at times t < t0 ,
there is no wave. After t = t0 , at times t > t0 , the wave spreads out (propagates) to the
location |r − r 0 | = c(t − t0 ) and as it spreads, its amplitude decreases as |r−r
1
0 | . A close
analogue is the wave produced by dropping a small but massive pebble into a quiet
pool of water.
Equation (12.3.22) is the prescription for superposing all of these elementary
waves, part originating from the continuous source described by the function σ(r, t)
and part from the boundary and initial conditions. Maximal simplification is obtained
for the case of boundaries at infinity, an initial time τ → −∞, and the requirement
that ψ(r, t) → 0 as r → ∞ and t → −∞. What results is a single integral of the form
0

1
Z σ r 0 − |t−tc |
ψ(r, t) = − dV 0 (12.3.32)
4π |r − r0 |
all space
which is referred to in electrodynamics as the “retarded potential” solution of the wave

equation.
Let us now turn our attention to the diffusion equation. Using Fourier transforms
again to solve
0 0 1 ∂
2
∇ G(r, t;r , t ) − G(r, t;r 0 , t0 ) =δ(r − r 0 )δ(t − t0 ),
κ ∂t
Unauthenticated
we find
Z∞ 0 0
e−ik·(r−r )−iω(t−t ) 2
Z
0 0 iκ
G(r, t;r , t ) = − 4 k dkd( cos θ k )d φ k dω.
(2π ) ω+iκ k2
−∞ all space
The integral over ω can be evaluated by a simple application of the residue calculus.
The integrand has a single simple pole located at ω = −iκ k2 in the lower half of the
complex ω plane. If t < t0 , the contour will have to be closed in the upper half plane
excluding the pole and resulting in a null result. But, for t > t0 , we must close the con-
tour in the lower half plane and thus pick up a non-zero contribution from its residue.
Thus,
0 2 0

 − κ3 e−ik·(r−r )−κ k (t−t ) k2 dkd( cos θ k )d φ k
R
0 0 (2π )
G(r, t;r , t ) = all space (12.3.33)
0

for t > t0 and t < t0 , respectively. This result inherently meets the initial condition that
we wanted to impose on this Green’s function.
The integral is a three-dimensional version of one we evaluated in Chapter 3. If we
set R=r − r 0 and T = t − t0 and complete the square in the exponent, it becomes
Z 2
Z
0 2
)−κ k2 (t−t0 ) R
e−κT (k−i 2κT ) d k1 d k2 d k3
R
e−ik·(r−r 2 −
k dkd( cos θ k )d φ k = e 4κT
all space all space
3
Z∞

2 X 2
e−κT ( ) dk 
R
− 4κT k1 −i 2κT
=e  1
−∞
where we have switched from spherical to rectangular coordinates and set x − x0 =

X. The rintegral in the second line was evaluated in Chapter 3 and we found that it
π
equals . Therefore, substituting all this information back into (12.3.33), we obtain
κT
as our final expression
 κ
0 0
 −
3/2
exp [ − |r − r 0 |2 /4κ(t − t0 )] for t > t0
G(r, t;r , t ) = 0
[4πκ(t − t ) ] (12.3.34)
 0 for t < t0
This is a sharply peaked (Gaussian) function of |r − r 0 | for small values of t − t0

and broad for large values of t − t0 . In fact, in the limit as t → t0 , the Green’s function
G(r, t;r 0 , t0 ) → − κδ(r − r 0 ). Thus, it describes how a delta function impulse at t = t0
diffuses through the medium at later times.
We can now use (12.3.25) to write down the solution to a diffusion problem involv-
ing a source σ(r, t), boundary condition ψ → 0 (and G → 0) as |r − r 0 | → ∞, and
initial condition ψ(r, t) given at t = τ: it is
Unauthenticated
Zt Z
1 1
ψ(r, t) = √ 3/2
exp [ − |r − r 0 |2 /4κ(t − t0 )]σ(r0 , t0 )dt0 dV 0
4πκ (t − t0 )
τ all space
Z
1
+ 3/2
exp [ − |r − r 0 |2 /4κ(t − τ)]ψ(r0 ,τ)dV 0 . (12.3.35)
[4πκ(t − τ) ]
all space
In particular, if the source σ(r, t) ≡ 0 and ψ(r, 0) =δ(x − a), the solution is
1 (x−a )2
ψ(r, t) =ψ(x, t) = √ e− 4κt
4πκt
which describes the diffusion in both x-directions of a substance initially confined in
the yz-plane at x = a.
Unauthenticated

(9783110409475 - Mathematics For The Physical Sciences) 12 Non-Homogeneous Boundary Value Problems Greens Functions

Cargado por

Información del documento

Título original

Derechos de autor

Formatos disponibles

Compartir este documento

Compartir o incrustar documentos

Opciones para compartir

¿Le pareció útil este documento?

¿Este contenido es inapropiado?

Copyright:

Formatos disponibles

(9783110409475 - Mathematics For The Physical Sciences) 12 Non-Homogeneous Boundary Value Problems Greens Functions

Cargado por

Copyright:

Formatos disponibles

12 Non-Homogeneous Boundary Value Problems:

12.1.1 Definition of a Green’s Function

for given values of the constants α1 , α2 , α3 , β1 , β2 , and β3 .

LG(x; x0 ) = δ(x − x0 ). (12.1.4)

With these choices, the identity (12.1.3) gives us

© 2014 Leslie Copley

LG(x; x0 ) = δ(x − x0 ) (12.1.6)

12.1.2 Direct Construction of the Sturm Liouville Green’ s Function

The Green’s function DE (12.1.7) is as close to a homogeneous equation as a non- homo-

because G satisfies homogeneous boundary conditions. Thus, we conclude that

G(x; x0 ) = G(x0 ; x). (12.1.8)

Next, we note that integrating the DE from x0 − ε to x0 + ε gives us

Equation (12.1.8) tells us that G(x; x0 ) is symmetric about x = x0 ,

G(x0 + ε; x0 ) = G(x0 ; x0 + ε) = G(x0 − ε; x0 ),

Since we also have

G(x; x0 ) = c < u < (x) for a ≤ x < x0 (12.1.10)

where c < is some constant.

G(x; x0 ) = c > u > (x) for x0 < x ≤ b (12.1.11)

where c > is some other constant.

for the Wronskian of the DE

12.1.3 Application: The Bowed Stretched String

u < (x) = sin kx, 0 ≤ x.

Therefore, since p(x) ≡ 1 in this case, the desired Green’s function is

12.1.4 Eigenfunction Expansions : The Bilinear Formula

L u m (x) = − λ m ρ(x) u m (x) (12.1.16)

and the (homogeneous) boundary conditions

Because of the orthogonality of the {u m (x)}, this implies

This is called the bilinear formula for the Green’s function.

for that particular value of m.

12.1.5 Application: the Infinite Stretched String

Evidently, this is an inverse Fourier transform and so must be the solution of

we do indeed recover the bilinear formula (12.1.24).

and in the second,

12.2 Partial Differential Equations

12.2.1 Green’s Theorem and Its Consequences

In more than one dimension a non-homogeneous boundary value problem generally

Lψ(r) = f (r) with L ≡ ∇·[p(r)∇] + s(r) (12.2.1)

The proof of the theorem follows from a consideration of the integrals

LG(r;r 0 ) =δ(r − r 0 ) (12.2.3)

in V subject to suitable boundary conditions on S. Applying Green’s Theorem with u(r)

which is the multi-dimensional analogue of equation (12.1.5).

Case 1 (Dirichlet Boundary Conditions): ψ(r) is given on S.

G(r;r 0 ) = 0 for r on S. (12.2.5)

Equation (12.2.4) then becomes

Case 2 (Neumann Boundary Conditions): n·∇ψ(r) is given on S.

n·∇G(r;r 0 ) = 0 for r on S (12.2.7)

and if s(r) ≡ 0, we require the next best thing,

In the first instance, (12.2.4) becomes

and in the second,

G(r 0 ;r 00 ) = G(r 00 ;r 0 ). (12.2.11)

In other words, G(r;r 0 ) is symmetric under r ↔ r 0 .

12.2.2 Poisson’s Equation in Two Dimensions and With Rectangular Symmetry

G(0, y; x0 , y0 ) = G(a, y; x0 , y0 ) = G(x, 0; x0 , y0 ) = G(x, b; x0 , y0 ) = 0. (12.2.14)

Eigenfunction Expansion Method:

Since their closure relation must be

u λ (x, y) u*λ (x0 , y0 ) = δ(x − x0 )δ(y − y0 ),

Direct Construction Method:

Had our starting point been substitution of

This, of course, yields an expression for G(x, y; x0 , y0 ) that is equivalent to summing

12.2.3 Potential Problems in Three Dimensions and the Method of Images

This is confirmed by direct integration of (12.2.24). Integrating over a spherical

and substitute both expansions into the PDE of (12.2.29) to obtain

Invoking the orthogonality of the spherical harmonics to set up equations on a term

Far from the source, r r0 , we have |r − r 0 | ∼ r − n·r 0 where n= rr and so