526 SECTION 13.

§13.2 Green’s Functions for Dirichlet Boundary Value Problems

Dirichlet problems for the two-dimensional Helmholtz equation take the form

Lu = ∇2 u + k 2 u = F (x, y), (x, y) in A, (13.15a)

u(x, y) = K(x, y), (x, y) on β(A). (13.15b)

For k = 0, we have the special case of Poisson’s equation. When F (x, y) has
continuous first derivatives and piecewise continuous second derivatives in A, as
does K(x, y) on β(A), this problem has a unique solution. The special case in
which A is a rectangle was discussed in Section 6.7 (see problem 6.70). In practical
situations when F (x, y) and K(x, y) may not satisfy these conditions, verification
of uniqueness is much more difficult, as is finding the solution by previous methods.
Green’s functions provide an excellent alternative.
We define the Green’s function G(x, y; X, Y ) for problem 13.15 as the solution
of

LG = ∇2 G + k 2 G = δ(x − X, y − Y ), (x, y) in A, (13.16a)

G(x, y; X, Y ) = 0, (x, y) on β(A). (13.16b)

It is the solution of problem 13.15 due to a unit source at the point (X, Y ) when
boundary conditions are homogeneous. In Section 13.3, we shall prove that the
solution of boundary value problem 13.15 can be expressed in the form
ZZ Z
∂G(x, y; X, Y )
u(x, y) = G(x, y; X, Y )F (X, Y ) dA + K(X, Y ) ds, (13.17)
A β(A) ∂N

where ∂G/∂N is the outward normal derivative of G with respect to the (X, Y )
variables along β(A). The solution is expressed in terms of integrals of the associated
Green’s function and source and boundary terms F (x, y) and K(x, y). We shall also
interpret these integrals physically. In this section, we concentrate on methods for
finding Green’s functions.
If we substitute u = G(x, y; X, Y ) and v = G(x, y; R, S) into Green’s identity
13.14a,
ZZ
[G(x, y; R, S)∇2 G(x, y; X, Y ) − G(x, y; X, Y )∇2 G(x, y; R, S)] dA = 0
A

(since G(x, y; R, S) and G(x, y; X, Y ) satisfy boundary condition13.16b). But be-

cause G is a solution of PDE 13.16a, we may write
ZZ
0= {G(x, y; R, S)[δ(x − X, y − Y ) − k 2 G(x, y; X, Y )]
A
− G(x, y; X, Y )[δ(x − R, y − S) − k 2 G(x, y; R, S)]} dA
= G(X, Y ; R, S) − G(R, S; X, Y ).

In other words, the Green’s function is symmetric under the interchange of first and
second variables with third and fourth,

G(x, y; X, Y ) = G(X, Y ; x, y). (13.18)

 SECTION 13.2 527

This result is also valid when boundary condition 13.15b is replaced by either a
Neumann or a Robin condition.
For boundary value problems associated with ODEs, we derived general for-
mulas (equations 12.33 and 12.34 in Section 12.3) for Green’s functions. This was
possible because boundaries for ODEs consist of two points. For PDEs, boundaries
consist of curves for two-dimensional problems and surfaces for three-dimensional
problems. As a result, it is impossible to find formulas for Green’s functions as-
sociated with multivariable boundary value problems. What we can do is develop
general techniques useful in large classes of problems. In this section, we illustrate
four of these techniques for finding the Green’s function for Dirichlet problem 13.15
in the case of Poisson’s equation,

Lu = ∇2 u = F (x, y), (x, y) in A, (13.19a)

u(x, y) = K(x, y), (x, y) on β(A). (13.19b)

The Green’s function for this problem satisfies

LG = ∇2 G = δ(x − X, y − Y ), (x, y) in A, (13.20a)

G(x, y; X, Y ) = 0, (x, y) on β(A). (13.20b)

These techniques may also be appropriate for boundary value problems with Neu-
mann or Robin conditions or mixed problems (problems with different types of
boundary conditions on different parts of the boundary), and also for the Helmholtz
equation.
Full Eigenfunction Expansion
In this method, the Green’s function is expanded in terms of orthonormal eigen-
functions of the associated eigenvalue problem

Lu + λ2 u = 0, (x, y) in A, (13.21a)
u(x, y) = 0, (x, y) on β(A). (13.21b)

We illustrate with the following example.

Example 13.1 Find the Green’s function associated with the Dirichlet problem for the two-dimen-
sional Laplacian on a rectangle A: 0 ≤ x ≤ L, 0 ≤ y ≤ L0 .
Solution Separation of variables on

∇2 u + λ2 u = 0, (x, y) in A, (13.22a)
u(x, y) = 0, (x, y) on β(A), (13.22b)

leads to normalized eigenfunctions

2 nπx mπy
umn (x, y) = √ sin sin 0 ,
LL0 L L
corresponding to eigenvalues λ2mn = (nπ/L)2 + (mπ/L0 )2 (see Section 6.5). The
eigenfunction expansion of G(x, y; X, Y ) in terms of these eigenfunctions is
∞ X
X ∞
G(x, y; X, Y ) = cmn umn (x, y), (13.23)
m=1 n=1
 528 SECTION 13.2

and this representation satisfies the boundary condition that G vanish on the edges
of the rectangle. To calculate the coefficients cmn , we substitute this representation
into the PDE ∇2 G = δ(x − X, y − Y ) for G and expand the delta function in terms
of the umn (x, y),
X∞ X ∞  2 2 
n π m2 π 2
cmn − 2 − umn (x, y)
m=1 n=1
L L02
= δ(x − X, y − Y )
∞ X ∞
"Z Z 0 #
X L L
= δ(x − X, y − Y )umn (x, y) dy dx umn (x, y)
m=1 n=1 0 0

X∞ X ∞
= umn (X, Y )umn (x, y).
m=1 n=1

Consequently, cmn = umn (X, Y )/(−λ2mn ), and

X∞ X ∞
umn (X, Y )
G(x, y; X, Y ) = umn (x, y)
m=1 n=1
−λ2mn
∞ ∞
−4 X X 1 nπX mπY nπx mπy
= 0  nπ 2  mπ 2 sin sin 0
sin sin 0 . (13.24)
LL m=1 n=1 L L L L
+
L L0
In Exercise 1 it is shown that this representation can also be obtained using Green’s
identity 13.14a. This avoids the interchange of the Laplacian and summation oper-
ations and the eigenfunction expansion of δ(x − X, y − Y ).•
A general formula for full eigenfunction expansions can be found in Exercise
2, but such representations are of limited calculational utility. First, they are pos-
sible only when the eigenvalue problem can be separated, and this requires that
the boundary of A consist of coordinate curves (or coordinate surfaces, in three-
dimensional problems), Second, in the case in which the full eigenfunction expansion
is available, a partial eigenfunction expansion that converges more rapidly is also
available.
Partial Eigenfunction Expansion
Like the full eigenfunction expanison, this method requires that region A be bounded
by coordinate curves (or coordinate surfaces, in three-dimensional problems). It
differs in that separation is considered on the homogeneous problem

Lu = 0, (x, y) in A, (13.25a)
u(x, y) = 0, (x, y) on β(A), (13.25b)

and is carried out until one variable remains. An eigenfunction expansion for the
Green’s function is then found in terms of normalized eigenfunctions already deter-
mined, with coefficients that are functions of the remaining variable. We illustrate
once again with the problem in Example 13.1.
Example 13.2 Find a partial eigenfunction representation for the Green’s function in Example
13.1.
 SECTION 13.2 529

Solution Separation of variables on

∇2 u = 0, (x, y) in A, (13.26a)
u(x, y) = 0, (x, y) on β(A), (13.26b)
r
2 nπx
leads to normalized eigenfunctions fn (x) = sin . We expand G(x, y; X, Y )
L L
in terms of these,
∞
X
G(x, y; X, Y ) = an (y)fn (x). (13.27)
n=1

In actual fact, coefficients an (y) must also be functions of X and Y , but we shall
understand this dependence implicitly rather than express it explicitly. To deter-
mine the an (y), we substitute this expression into the PDE ∇2 G = δ(x − X, y − Y )
for G and expand the delta function in terms of the fn (x),
X∞ X∞
−n2 π 2 d2 an
an fn (x) + fn (x) = δ(x − X, y − Y )
n=1
L2 n=1
dy 2
∞
"Z #
X L
= δ(x − X, y − Y )fn (x) dx fn (x)
n=1 0

X∞
= fn (X)δ(y − Y )fn (x).
n=1

This equation and the boundary conditions G(x, 0; X, Y ) = 0 = G(x, L0 ; X, Y )

require the an (y) to satisfy

d2 an n2 π 2
− an = δ(y − Y )fn (X), 0 < y < L0 ,
dy 2 L2
an (0) = 0, an (L0 ) = 0.

We can solve this boundary value problem most easily by using our theory of Green’s
functions for ODEs. Since a solution of the homogeneous equation that satisfies
the first boundary condition is sinh (nπy/L), and one that satisfies the second is
sinh [nπ(L0 − y)/L], equation 12.34 in Section 12.3 gives
 
1 nπy nπ(L0 − Y ) nπY nπ(L0 − y)
an (y) = sinh sinh h(Y − y) + sinh sinh h(y − Y ) ,
J L L L L
where J is the conjunct of sinh (nπy/L) and sinh [nπ(L0 − y)/L],
   
1 nπy −nπ nπ(L0 − y)  nπ  nπy nπ(L0 − y)
J= sinh cosh − cosh sinh
fn (X) L L L L L L
0
nπ sinh (nπL /L)
= −√ .
2L sin (nπX/L)

Thus, an alternative to the double-series, full eigenfunction expansion is the single-

series, partial eigenfunction expansion
 530 SECTION 13.2
√ nπX "
X∞ − 2L sin
L nπy nπ(L0 − Y )
G(x, y; X, Y ) = sinh sinh h(Y − y)
nπL0 L L
n=1 nπ sinh
L #r
nπY nπ(L0 − y) 2 nπx
+ sinh sinh h(y − Y ) sin
L L L L

 nπX nπx nπy nπ(L0 − Y )
X
 ∞ −2 sin sin sinh sinh

 L L L L

 , 0≤y≤Y

 nπL0
 n=1 nπ sinh
= L (13.28)

 nπX nπx nπY nπ(L0 − y)
 X −2 sin
 ∞ sin sinh sinh

 L L L L , Y ≤ y ≤ L0 .

 0

 n=1 nπL
nπ sinh
L
It is clear that we could find an equivalent solution by expanding G in a Fourier
sine series in y. The result would be


 nπY nπy nπx nπ(L − X)

 X∞ −2 sin sin 0 sinh 0 sinh

 L 0 L L L0 , 0≤x≤X

 nπL

 n=1 nπ sinh 0
G(x, y; X, Y ) = L (13.29)

 nπY nπy nπX nπ(L − x)

 X∞ −2 sin sin sinh sinh

 L0 L0 L0 L0 , X ≤ x ≤ L.

 nπL

 n=1 nπ sinh 0
L
A natural question to ask is: In which problems, should each of these expres-
sions for G(x, y; X, Y ) be used? Since each is a Fourier series, rates of convergence
of the series will depend on the relative magnitudes of coefficients. The coefficient
of sin (nπx/L) in representation 13.28 for y > Y is
nπX nπY nπ(L0 − y)
−2 sin sinh sinh
L L L ,
nπL0
nπ sinh
L
and for large n we may drop the negative exponentials in the hyperbolic functions
and approximate this quantity with
0
enπY /L enπ(L −y)/L nπX −1 nπ(Y −y)/L nπX
− nπL 0 /L sin = e sin .
nπe L nπ L
Similarly, when y < Y , the coefficient can, for large n, be approximated by
−1 nπ(y−Y )/L nπX
e sin .
nπ L
Corresponding coefficients in representation 13.29 are approximated for large n by
−1 nπ|X−x|/L0 nπY
e sin 0 .
nπ L
It follows that to calculate G(x, y; X, Y ) at a value of x that is substantially different
from X, it would be wise to use representation 13.29, and, conversely, when y is
markedly different from Y , representation 13.28 would provide faster convergence.
 SECTION 13.2 531

In addition, when boundary integrals arise for the solution of Dirichlet prob-
lem 13.19 (and this occurs for nonhomogeneous boundary conditions 13.19b), it is
advantageous to use representation 13.28 for integrations along y = 0 and y = L0 ,
but use representation 13.29 along x = 0 and x = L.
Splitting Technique
Sometimes it is convenient to split G into two parts, G = U +g, where U contains the
singular part of G due to the delta function in PDE 13.20a and g guarantees that G
satisfies the boundary conditions associated with L. This splitting technique permits
consideration of the singular nature of the Green’s function without the annoyance
of boundary conditions. (The technique could have been used for ODEs, but it was
unnecessary because formulas 12.33 and 12.34 in Section 12.3 were presented for
Green’s functions.) To be more specific, for the Green’s function satisfying problem
13.20, we set G = U + g, where U (x, y; X, Y ) satisfies the PDE
LU = δ(x − X, y − Y ) (13.30)
and g satisfies the boundary value problem
Lg = 0, (x, y) in A, (13.31a)
g = −U, (x, y) on β(A). (13.31b)
Because U (x, y; X, Y ) is not required to satisfy boundary conditions, it is often
called the free-space Green’s function for the operator L. Free-space Green’s
functions for the Laplace and Helmholtz operators in two and three dimensions are
listed in Table 13.1. Each is singular at the source point (X, Y ).

∇2 ∇2 + k 2
Laplacian Helmholtz

xy 1 p 1 p
plane ln (x − X)2 + (y − Y )2 Y0 [k (x − X)2 + (y − Y )2 ]
2π 4
√ 2 2 2
eik (x−X) +(y−Y ) +(z−Z)
− p ,
xyz −1 4π (x − X)2 + (y − Y )2 + (z − Z)2
space 4π p(x − X)2 + (y − Y )2 + (z − Z)2 √ 2 2 2
e−ik (x−X) +(y−Y ) +(z−Z)
− p
4π (x − X)2 + (y − Y )2 + (z − Z)2

Table 13.1
We justify the first entry here; the other three are discussed in the exercises.
The two-space Green’s function G(x, y; X, Y ) for the Laplacian is the solution of
∇2 G = δ(x − X, y − Y ).
It is the effect at point (x, y) due to a unit source at (X, Y ). Because the function
should be symmetric about the source point, we switch to polar coordinates centred
at (X, Y ), and search for a function G(r; 0) satisfying
d2 G 1 dG δ(r)
2
+ = , (13.32)
dr r dr 2πr
 532 SECTION 13.2

where we have used equation 13.9 for the delta function at the origin. Multiplication
by r leads to
 
d dG δ(r)
r = .
dr dr 2π
Integration with respect to r from r = 0 to an arbitrary value of r gives
dG 1 dG 1 1
r = =⇒ = =⇒ G(r; 0) = ln r + C.
dr 2π dr 2πr 2π
We take C = 0. This shows that the effect at a point due to a unit source is 1/(2π)
times the logarithm of the distance from point to source. It follows that the effect
at point (x, y) due to a source at (X, Y ) is
1 p
G(x, y; X, Y ) = ln (x − X)2 + (y − Y )2 .
2π
A similar derivation gives the free-space Green’s function for the three-dimen-
sional Laplacian (Exercise 25). Unfortunately, the same technique does not work
for the Helmholtz operator. In Exercise 26, we provide an alternative derivation for
free-space Green’s functions associated with the Laplacian and this technique does
extend to Helmholtz operators (Exercises 28 and 29).
We now return to the splitting technique by illustrating it in the following
example.
Example 13.3 Find the Green’s function for the Dirichlet problem associated with Laplace’s equa-
tion on a circle 0 ≤ r ≤ a.
Solution The Green’s function associated with the Dirichlet problem for the
Laplacian on a circle centred at the origin with radius a satisfies

δ(r − R)δ(θ − Θ)
∇2 G = , 0 < r < a, −π < θ ≤ π, (13.33a)
r
G(a, θ; R, Θ) = 0, −π < θ ≤ π. (13.33b)

The free-space Green’s function for the two-dimensional Laplacian with singularity
at (R, Θ) is
1 p
U (r, θ; R, Θ) = ln (r cos θ − R cos Θ)2 + (r sin θ − R sin Θ)2
2π
1
= ln [r 2 + R2 − 2rR cos (θ − Θ)]
4π

(see Table 13.1). When we split G into G = U + g, function g must satisfy

∇2 g = 0, 0 < r < a, −π < θ ≤ π, (13.34a)

1
g(a, θ; R, Θ) = − ln [a2 + R2 − 2aR cos (θ − Θ)], −π < θ ≤ π. (13.34b)
4π
Separation of variables on the PDE, together with boundedness at r = 0, leads to
a solution of the form
X∞  
A0 n cos nθ n sin nθ
g(r, θ; R, Θ) = √ + An r √ + Bn r √
2π n=1 π π
 SECTION 13.2 533

(see equation 6.31a in Section 6.3). Boundary condition 13.34b requires

X∞  
A0 n cos nθ n sin nθ −1
√ + An a √ + Bn a √ = ln [a2 + R2 − 2aR cos (θ − Θ)]
2π n=1 π π 4π
"  2   #
−1 1 R R
= ln a2 − ln 1 + −2 cos (θ − Θ) .
4π 4π a a

With the result

X∞
αn cos nφ 1
= − ln (1 + α2 − 2α cos φ), (|α| < 1), (13.35)
n=1
n 2

we may write
X∞  
A0 n cos nθ n sin nθ
√ + An a √ + Bn a √
2π n=1 π π
∞
−1 1 X (R/a)n
= ln a2 + cos n(θ − Θ)
4π 2π n=1 n
∞
−1 1 X (R/a)n
= ln a2 + (cos nθ cos nΘ + sin nθ sin nΘ).
4π 2π n=1 n

Comparison of coefficients requires

A −1 An an (R/a)n Bn an (R/a)n
√0 = ln a2 , √ = cos nΘ, √ = sin nΘ,
2π 4π π 2πn π 2πn
and therefore
X∞  n

−1 n (R/a) (R/a)n
g(r, θ; R, Θ) = ln a + r cos nθ cos nΘ + sin nθ sin nΘ
2π n=1
2πnan 2πnan
∞
−1 1 X (rR/a2 )n
= ln a + cos n(θ − Θ).
2π 2π n=1 n

But identity 13.35 permits evaluation of this series in closed form,

"  2   #
−1 1 rR rR
g(r, θ; R, Θ) = ln a − ln 1 + −2 cos (θ − Θ)
2π 4π a2 a2
1 1
= ln a − ln [a4 + R2 r 2 − 2a2 Rr cos (θ − Θ)].
2π 4π
Finally,
1 1
G(r, θ; R, Θ) = U + g = ln [r 2 + R2 − 2Rr cos (θ − Θ)] + ln a
4π 2π
1
− ln [a4 + R2 r 2 − 2a2 Rr cos (θ − Θ)]
 4π 
1 2 r 2 + R2 − 2Rr cos (θ − Θ)
= ln a 4 . (13.36)
4π a + R2 r 2 − 2a2 Rr cos (θ − Θ)
 534 SECTION 13.2

This result is also obtained with a partial eigenfunction expansion in Exercise 13.•
The splitting technique points out a distinct difference between Green’s func-
tions for one-dimensional problems and those for multidimensional problems. The
Green’s function g(x; X) for a one-dimensional boundary value problem (associ-
ated with a second-order ODE) is a continuous function of x (or can be made
so) with a jump discontinuity in its first derivative. Green’s functions for multi-
dimensional boundary value problems can always be represented as the sum of a
free-space Green’s function U and a regular part g, and, according to Table 13.1,
free-space Green’s functions are always singular at the source point. Thus, multi-
variable Green’s functions always have discontinuities at source points.
Method of Images
The method of images is simply physical reasoning and intelligent guesswork in
arriving at the function g in the splitting technique, and as such it works only on
Laplace’s equation with very simple geometries. When the Green’s function G for a
domain A is split into U + g, the free-space Green’s function U can be regarded as
the potential due to a unit point source interior to A. This source, by itself, induces
a nonzero potential on β(A). What is needed is a source distribution exterior to A
whose potential g will cancel the effect of U on β(A). (The fact that this distribution
is exterior to A guarantees that G = U + g satisfies ∇2 G = δ interior to A.)
We illustrate with the following three-dimensional problem.
Example 13.4 Find the Green’s function associated with the three-dimensional Dirichlet problem
for Laplace’s equation in a sphere of radius a.
Solution The Green’s function satisfies

δ(r − R)δ(θ − Θ)δ(φ − Φ)

∇2 G = , 0 < r < a, 0 < φ < π, −π < θ ≤ π,(13.37a)
r 2 sin φ
G(a, φ, θ; R, Φ, Θ) = 0, 0 < φ < π, −π < θ ≤ π. (13.37b)

Accordingpto Table 13.1, the free-space Green’s function with source point (X, Y, Z)
is −1/[4π (x − X)2 + (y − Y )2 + (z − Z)2 ]. When (R, Φ, Θ) are the spherical co-
ordinates of (X, Y, Z), this function becomes
−1
U (r, φ, θ; R, Φ, Θ) = p .
4π r2 + R2 − 2Rr[cos φ cos Φ + sin φ sin Φ cos (θ − Θ)]
What the method of images suggests is
finding a source distribution exterior
to the sphere, the potential g for
z
which is such that G = U + g vanishes
on r = a. We might first consider a (R *, F, Q)
whether a single source of magnitude q
at a point (R∗ , Φ∗ , Θ∗ )(R∗ > a) might (a, f, q )
suffice. Symmetry would suggest that a a
such a source could eliminate U on (R , F,Q)
x y
r = a, which is symmetric around the
line through the origin, and (R, Φ, Θ) Figure 13.1
 SECTION 13.2 535

(Figure 13.1) only if (R∗ , Φ∗ , Θ∗ ) were to lie on the line also. We assume, therefore,
that Θ∗ = Θ and Φ∗ = Φ, in which case the condition that G = U + g vanish on
r = a is
−1
0= p
4π a2 + R2 − 2aR[cos φ cos Φ + sin φ sin Φ cos (θ − Θ)]
−q
+ q
4π a2 + R∗ 2 − 2aR∗ [cos φ cos Φ + sin φ sin Φ cos (θ − Θ)]

or,
p
−q a2 + R2 − 2aR[cos φ cos Φ + sin φ sin Φ cos (θ − Θ)]
q
= a2 + R∗ 2 − 2aR∗ [cos φ cos Φ + sin φ sin Φ cos (θ − Θ)].

Since this condition must be valid for all φ and θ, we set φ = 0 and φ = π,
p p
−q a2 + R2 − 2aR cos Φ = a2 + R∗ 2 − 2aR∗ cos Φ,
p p
−q a2 + R2 + 2aR cos Φ = a2 + R∗ 2 + 2aR∗ cos Φ.

These two equations imply that R∗ = a2 /R and q = −a/R, and with these, U + g
vanishes identically on r = a. Thus, the Green’s function for the Laplacian inside a
sphere of radius a is
−1
G(r, φ, θ; R, Φ, Θ) = p
4π r2 + R2
− 2Rr[cos φ cos Φ + sin φ sin Φ cos (θ − Θ)]
a
+ s  2 2  2
a a
4πR r 2 + − 2r [cos φ cos Φ + sin φ sin Φ cos (θ − Θ)]
R R
−1
= p
2 2
4π r + R − 2Rr[cos φ cos Φ + sin φ sin Φ cos (θ − Θ)]
a
+ p .
2 2 4 2
4π R r + a − 2a Rr[cos φ cos Φ + sin φ sin Φ cos (θ − Θ)]

(13.38)

EXERCISES 13.2
1. Show that coefficients cmn in representation 13.23 can be obtained by substituting v = umn (x, y)
and u = G(x, y; X, Y ) in Green’s identity 13.14a.
2. Show that when un (x, y) are orthonormal eigenfunctions of the eigenvalue problem

∇2 u + λ2 u = 0, (x, y) in A, (13.39a)
u(x, y) = 0, (x, y) on β(A), (13.39b)

associated with the Dirichlet problem

 536 SECTION 13.2

∇2 u = F (x, y), (x, y) in A, (13.40a)

u(x, y) = K(x, y), (x, y) on β(A), (13.40b)
the full eigenfunction expansion for the Green’s function is
X∞
un (X, Y )un (x, y)
G(x, y; X, Y ) = . (13.41)
n=1
−λ2n

(This expansion should be compared with that in Exercise 25 of Section 12.3 for the Green’s
function of an ODE.)
In Exercises 3–8 use Exercise 2 (and its extension to three dimensions) to find
full eigenfunction expansions for the Green’s function associated with the Dirichlet
problem for Poisson’s equation on the given domain.
3. 0 ≤ r < a, −π < θ ≤ π 4. 0 ≤ r < a, 0 < θ < π
5. 0 ≤ r < a, 0 < θ < L 6. 0 < x < L, 0 < y < L0 , 0 < z < L00
7. 0 ≤ r < a, −π < θ ≤ π, 0 < z < L 8. 0 ≤ r < a, 0 ≤ φ ≤ π, −π < θ ≤ π
9. Use the method of images and the result of Example 13.4 to find the Green’s function for the
Dirichlet problem associated with Poisson’s equation in a hemisphere of radius a.
10. Use a “modified” method of images to find the Green’s function for the Dirichlet problem
associated with the two-dimensional Laplacian on a circle of radius a. Assume that g consists
of a potential due to an exterior, negative unit point source plus a constant potential.
11. Use the result of Exercise 10 and the method of images to find the Green’s function for the
Dirichlet problem associated with Poisson’s equation on a semicircle 0 < r < a, 0 < θ < π.
How does it compare with the representation in Exercise 4.
12. Use the method of images to find the Green’s function for the Dirichlet problem for the Laplacian
on the rectangle 0 < x < L, 0 < y < L0 .
13. In this exercise we use a partial eigenfunction expansion to find Green’s function 13.36 for
problem 13.33.
(a) Show that the partial eigenfunction expansion for G(r, θ; R, Θ) is
∞  
A0 (r) X cos nθ sin nθ
G(r, θ; R, Θ) = √ + An (r) √ + Bn (r) √ .
2π n=1
π π

(b) Substitute the expansion in part (a) into PDE 13.33a, and expand δ(r − R)δ(θ − Θ)/r in a
Fourier series to obtain the following boundary value problems for the coefficients:
 
d dA0 δ(r − R)
r = √ , A0 (a) = 0;
dr dr 2π
 
d dAn n2 cos nΘ
r − An = δ(r − R) √ , An (a) = 0;
dr dr r π
  2
d dBn n sin nΘ
r − Bn = δ(r − R) √ , Bn (a) = 0.
dr dr r π

(c) The systems in part (b) are “singular” in the sense that there is only one boundary condition
and the coefficient r in the derivative term vanishes at r = 0. As a result, equations 12.33
and 12.34 in Section 12.3 cannot be used to find An and Bn . Instead, use properties 12.26a–c
from Section 12.3 and the one boundary condition to show that
 SECTION 13.2 537



 ln (R/a)
 √ , 0≤r≤R
A0 (r) = 2π

 ln (r/a)
 √ , R < r ≤ a,
2π
  n   

 cos nΘ rR r n
 √ 2
− , 0≤r≤R
2 πn  a  R
 n 
An (r) = n
 cos nΘ
 rR R
 √ − , R < r ≤ a,
2 πn a2 r
  n   

 sin nΘ rR r n
 √ 2
− , 0≤r≤R
2 πn  a  R 
Bn (r) = n n

 sin nΘ rR R
 √ − , R < r ≤ a.
2 πn a2 r

(d) Find G(r, θ; R, Θ) and use identity 13.35 to reduce the function to the form in equation
13.36.
14. Use the technique of Exercise 13 to find a partial eigenfunction expansion for the Green’s
function of the Dirichlet problem for the Laplacian on the semicircle 0 < r < a, 0 < θ < π.
Show that it can be expressed in the form of Exercise 11.
15. Use the technique of Exercise 13 to find the partial eigenfunction expansion for the Green’s
function of Exercise 5.
16. Find a partial eigenfunction expansion for the Green’s function of Exercise 6 using eigenfunctions
in x and y.
17. Show that when un (x, y) are orthonormal eigenfunctions of eigenvalue problem 13.21, the full
eigenfunction expansion for the Green’s function of the boundary value problem

∇2 u + k 2 u = F (x, y), (x, y) in A, (13.42a)

u(x, y) = K(x, y), (x, y) on β(A), (13.42b)

is
X∞
un (X, Y )un (x, y)
G(x, y; X, Y ) = , (13.43)
n=1
k 2 − λ2n

provided k 6= λn for any n. (The exceptional case is discussed in Exercise 8 of Section 13.3.)
In Exercises 18–24 use Exercise 17 to state Green’s functions for problem 13.42 on
the given domain. (See Example 13.1 and Exercises 3–8 for eigenpairs.)
18. 0 < x < L, 0 < y < L0 19. 0 ≤ r < a, −π < θ ≤ π
20. 0 ≤ r < a, 0 < θ < π 21. 0 < r < a, 0 < θ < L
0 00
22. 0 < x < L, 0 < y < L , 0 < z < L 23. 0 ≤ r < a, −π < θ ≤ π, 0 < z < L
24. 0 ≤ r < a, 0 ≤ φ ≤ π, −π < θ ≤ π
25. Derive the free-space Green’s function for the 3-dimensional Laplacian by taking the source at
the origin and using spherical coordinates centred there.
 538 SECTION 13.2

26. In this exercise we give a derivation of free-space Green’s functions for the two-dimensional
Laplacian that can be used to find free-space Green’s functions for Helmholtz operators.
(a) Show that G(r; 0) = C ln r + D is a general solution of the homogeneous version of equation
13.32.
(b) By substituting v = G(r; 0) and u = 1 in Green’s second identity 13.14a where A is a circle
of radius  centred at the source r = 0, show that
∂G(; 0) 1
 = .
∂r 2π
(c) Reason that G(r; 0) must satisfy
∂G(r; 0) 1 ∂G(r; 0) 1
r = and lim r = .
∂r 2π r→0 ∂r 2π
(d) Use the results of parts (a) and (c), to find G(r; 0).
27. Use the technique of Exercise 26 to derive the free-space Green’s function in Table 13.1 for the
three-dimensional Laplacian.
28. Use the technique of Exercise 26 to derive the free-space Green’s functions in Table 13.1 for the
three-dimensional Helmholtz operator ∇2 + k 2 . Hint: Set G(r; 0) = H(r)/r in the homogeneous
differential equation for G(r; 0).
29. Use the technique of Exercise 26 to derive the free-space Green’s function in Table 13.1 for the
two-dimensional Helmholtz operator ∇2 + k 2 .