22 Brief Introduction to Green’s Functions:

PDEs
In a previous section we discussed Laplace’s equation in the disk with Dirich-
let boundary conditions, namely

Example:

∇2 u = 0 , in Ω := {(r, θ) : 0 ≤ r < a , 0 ≤ θ < 2π}
u(a, θ) = f (θ) , for 0 ≤ θ < 2π, f is continuous
and derived Poisson’s formula for the solution,
a2 − r2 2π
Z
f (ψ) dψ
u(r, θ) = .
2π 0 a − 2ar cos(θ − ψ) + r2
2

1 a2 −r2
If we define G(r, ψ) := 2π a2 −2ar cos(ψ)+r2
, then we can rewrite u as
Z 2π
u(r, θ) = G(r, θ − ψ)f (ψ) dψ . (1)
0

Thus, the form of solution is an integral, actually a convolution integral, with

kernel G(·, ·) being considered the Green’s function for the problem.

Example: For the IVP for the heat equation we found that
Z ∞ −(x−y)2 /4Dt Z ∞
e
u(x, t) = √ f (y) dy = S(x − y, t)f (y) dy ,
−∞ 4πDt −∞

which again a convolution integral for the solution. Sometimes S(x, t), which
we called the fundamental solution to the heat equation, is also called the
Green’s function for the heat equation.

Example: Consider now the Poisson problem

 2
∇ u = −f (x, y) in Ω = {(x, y) : 0 < x < π, 0 < y < π}
u=0 on ∂Ω
For the homogeneous problem we have the eigenfunctions
P∞ {sin(nx)}n≥1 and
from
P∞ the Fourier transform idea, u(x, y) = n=1 bn (y) sin(nx), f (x, y) =
f
n=1 n (y) sin(nx), with
2 π 2 π
Z Z
0 0 0
bn (y) = u(x , y) sin(nx )dx and fn (y) = f (x0 , y) sin(nx0 )dx0 ,
π 0 π 0

1
and upon substitution of the series into the equation, we obtain

d2 bn
− n2 bn = −fn (y) 0 < y < π , bn (0) = bn (π) = 0 .
dy 2
Thus,
Z y
1
bn (y) = sinh(n(π − y)) sinh(nz)fn (z)dz
n sinh(nπ) 0
Z π 
+ sinh(ny) sinh(n(π − z))fn (z)dz
y

(For details, see Appendix at end of this section)

Therefore, in the Fourier sine series for u(x, y) above, substitute the defi-
nition of fn (y) and interchange integration and differentiation to obtain
Z πZ π
u(x, y) = G(x, y; ξ, η)f (ξ, η)dξdη (2)
0 0

where
( P
∞ 2 sinh(n(π−y)) sinh(nη)
n=1 sin(nx) sin(nξ)
nπ sinh(nπ)
η≤y
G(x, y; ξ, η) =
same series but with the roles of η and y switched y ≤ η

Remark: This series can be expressed in closed form only in terms of elliptic
functions, a calculation we will not pursue here. Still, the form of (2) leads
one to think of this G(x, y; ξ, η) as a Green’s function for this Poisson prob-
lem.

Boundary value problems above have solutions that end up being expressed
in terms of integrals whose integrands are either the boundary data or source
functions times a kernel function we will call Green’s function, G. The ques-
tion arises whether such a Green’s function and solution representation of a
PDE in terms of an integral can be derived more directly. This question is
motivated from ODE boundary value problems and associated Green’s func-
tions. We go through the construction of Green’s functions for the solution
of boundary value problems for the ODE case in Appendix J.

2
Exercise: Consider the Poisson problem defined on a square:
 2
∇ u = −1 Ω = {(x, y) : 0 < x, y < π}
u=0 on ∂Ω

Let v(x) = x2 (π − x) and note that v(0) = v(π) = 0 and ∇2 v = −1. Set
solution to the above problem u(x, y) = v(x) + w(x, y) and show that w
satisfies ∇2 w = 0 in Ω, w(0, y) = w(π, y) = 0, 0 < y < π, and w(x, 0) =
w(x, π) = − 12 x(π − x), 0 < x < π. This is a standard Laplace’s equation
problem, so solve it via separation of variables method.

Exercise: First, in the cartesian coordinate system, show the following:

1. Fix the point (ξ, η) in the plane and show that G = G(x, y : ξ, η) =
1
2π
ln( R1 ), where R is the distance between (x, y) and (ξ, η), i.e. R =
p
(x − ξ)2 + (y − η)2 , is a solution to the problem ∇2 u = uxx + uyy = 0
in the plane, except at (x, y) = (ξ, η).

2. Similarly, in 3-space, fix x0 = (ξ, η, ζ) ∈ R3 , and let ρ = |x − x0 | be the

distance between x0 and x = (x, y, z), and show that G(x, x0 ) = − 4π 1 1
ρ
is a solution in R3 \ {x0 } to ∇2 u = uxx + uyy + uzz = 0.

The G0 s in the above exercise are the free-space Green’s functions for
R2 and R3 , respectively. But in bounded domains Ω where we want to solve
the problem ∇2 u =R −f (x), x ∈ Ω, u = 0 on ∂Ω, and be able to write the
solution as u(x) = Ω G(x, x0 )f (x0 )dx0 , we need G = 0 on ∂Ω. Therefore, we
want G, the Green’s function associated with the domain Ω, to have the form
G = 2π1
ln( R1 ) + g(x, x0 ) in the 2D case, and G = − 4π
1 1
ρ
+ g(x, x0 ) in the 3D
case. Thus, g must be found so that G vanishes on the boundary ∂Ω, and g
is harmonic in Ω. This is difficult to do in general, but in some simpler cases
it can be done via a reflection principle. (In 2D, there are also complex
variable methods to find Green’s functions, but we will not delve into that
methodology in these Notes.)
The reflection principle is motivated by some physics experiments where a
potential field due to positive and negative charges placed at a given distance
apart gives rise to curves of constant potential; so there exists a boundary
of points between the two charges where there is zero potential. As a test

3
problem consider the Poisson problem on the upper half-plane, namely
 2
∇ u = −f (x, y) in Ω = {(x, y); y > 0}
u=0 for (x, 0) ∈ ∂Ω

Fix x0 = (ξ, η) ∈ Ω (maybe think of it as the location of a point positive

charge). Then, can we locate a “negative” point charge outside Ω and use
this to construct the appropriate g? Yes, let x00 = (ξ, −η). Then let

G(x, x0 ) = G(x, y, ξ, η) =

1 1 1 1
ln( p )− ln( p )=
2π 2
(x − ξ) + (y − η)2 2π (x − ξ) + (y + η)2
2

(x − ξ)2 + (y + η)2
 
1
ln .
4π (x − ξ)2 + (y − η)2

It is clear that G = 0 for (x, y) = (x, 0), and is harmonic for (x, y) 6= (ξ, η).
Thus, Z ∞Z ∞
u(x, y) = G(x, y, ξ, η)f (ξ, η) dξ dη .
0 −∞

Observe that we first fix a point (ξ, η) = x0 ∈ Ω and then construct G as

a function of x = (x, y) ∈ Ω such that G satisfies the boundary condition.
We do this by using a version of ln(1/R) (or 1/ρ in the 3D case) to preserve
the harmonic character of G. In the above example, the reflection point x00
1
is outside Ω = upper half-plane, so ln( |x−x 00 | ) is harmonic everywhere in Ω.

Of course, when we write out u(x, y), then for each (fixed) (x, y) ∈ Ω, G is
integrated against the problem’s non-homogeneity with respect to x0 = (ξ, η).
A slight variation of this is to rotate and translate the above problem so
that Ω = {(x, y) : x < 1, |y| < ∞}. Then

Example 2: Again fix x0 = (ξ, η) ∈ Ω (so ξ < 1). We need to reflect about the
boundary {x = 1} a point x00 = (ξ 00 , η) so that ln(1/|x−x0 |)−ln(1/|x−x00 |) =

4
Figure 1: For Example 3, with two boundary pieces, we have to have three
reflected points.

0 when x = (1, y). Hence, ξ 00 − 1 = 1 − ξ, or ξ 00 = −ξ + 2 and

G(x, x0 ) = G(x, y, ξ, η) =

1 1 1 1
ln( p )− ln( p )=
2π (x − ξ)2 + (y − η)2 2π (x + ξ − 2)2 + (y − η)2

(x + ξ − 2)2 + (y − η)2
 
1
ln .
4π (x − ξ)2 + (y − η)2

Example 3: Consider the quarter-plane problem, that is

 2
∇ u = −f (x) Ω = {(x, y) : x > 0, y > 0}
u=0 on ∂Ω
To have G|y=0 = 0 we do what we did in the first example, namely reflect
about the boundary {y = 0} to obtain
1 1 1 1
ln( 0
)− ln( )
2π |x − x | 2π |x − x00 |
where x = (x, y), x0 = (ξ, η), and x00 = (ξ, −η). But with the second piece of
boundary, {x = 0}, we must reflect about that boundary also: x000 = (−ξ, η)

5
(see Figure 1). Now we have
1 1 1 1 1 1
ln( ) − ln( ) − ln( ).
2π |x − x0 | 2π |x − x00 | 2π |x − x000 |
The problem now is that if we have x on one boundary segment or the other
segment, two of these log terms cancel out, but not three of them. To get
G = 0 on both segments of the boundary of the quarter-plane Ω we need to
add/subtract another term (that is harmonic in Ω). Pick the reflection point
x∗ = (−ξ, −η) and write

G(x, x0 ) = G(x, y, ξ, η) =

 
1 1 1 1 1
ln( ) − ln( ) − ln( ) + ln( ) =
2π |x − x0 | |x − x00 | |x − x000 | |x − x∗ |

1
{ln(1/R) − ln(1/R00 ) − ln(1/R000 ) + ln(1/R∗ )} .
2π
It should be clear from Figure 1 that when x = (0, y), then R000 = R, R∗ = R00 ,
so G|x=0 = 0. Similarly, when x = (x, 0), then R00 = R and R∗ = R000 , so
G|y=0 = 0. Since all the terms combined are harmonic in Ω, except when
(x, y) = (ξ, η), then G is the Green’s function for the quarter-plane that we
seek.

Example 4: Now consider the Poisson problem in the unit disk:

 2
∇ u = −f (r, θ) in Ω = {(r, θ) : 0 ≤ r < 1, 0 ≤ θ < 2π}
u=0 on ∂Ω = {(1, θ) : 0 ≤ θ < 2π}

Again fix x0 = (ξ, η) = (s, φ) ∈ Ω. Since the boundary can be described by

a single function (r = 1), we should be able to use a single reflection point.
A straight forward calculation using (x, y) = (r, θ) = (r cos(θ), r sin(θ)) and
(ξ, η) = (s, φ) = (s cos(φ), s sin(φ)) gives

|x − x0 |2 = (x − ξ)2 + (y − η)2 = r2 + s2 − 2rs cos(θ − φ) .

Let the reflected point (outside of the unit disk Ω) be x00 = (1/s, φ), which is
on the same “ray” as (ξ, η) = (s, φ), then |x − x00 |2 = r2 + s12 − 2 rs cos(θ − φ).

6
Then, when r = 1, we have
 
1 1 1
ln( ) − ln( ) =
2π |x − x0 | |x − x00 |

(s )
1 1 − (2/s) cos(θ − φ) + 1/s2
ln =
2π 1 − 2s cos(θ − φ) + s2

(s )
1 1 − (2/s) cos(θ − φ) + 1/s2
ln =
2π s2 (1/s2 − (2/s) cos(θ − φ) + 1)

1
ln(1/s)
2π
which is a constant. But a constant is trivially a harmonic function, so write
 
0 1 1 1 1
G(x, x ) = G(r, θ, s, φ) = ln( ) − ln( ) − ln( )
2π |x − x0 | |x − x00 | s
00
 
1 s|x − x |
= ln .
2π |x − x0 |

Thus, we have the solution

Z 2π Z 1  2 2 
1 s r − 2rs cos(θ − φ) + 1
u(r, θ) = ln f (s, φ) sds dφ . (3)
4π 0 0 r2 − 2rs cos(θ − φ) + s2

Remark: It is possible to solve Laplace’s equation problem

 2
 ∇ u=0 in Ω = {(r, θ) : 0 ≤ r < 1, 0 ≤ θ < 2π}

u(1, θ) = h(θ) for 0 ≤ θ < 2π



by means of the Green’s function approach. In this case the solution is

Z 2π
∂G
u(r, θ) = (r, θ, 1, φ)h(φ) dφ ,
0 ∂r

7
where G is given above. By a tedious calculation, one can obtain from the
derivative of G that
Z 2π
1 1 − r2
u(r, θ) = h(φ) dφ (4)
2π 0 1 − 2r cos(θ − φ) + r2
which is exactly Poisson’s integral formula derived earlier.

Remark: Given the more general problem

 2
∇ u = −f (r, θ) in Ω
u = h(θ) on ∂Ω
where Ω = unit disk, the solution would just be the sum of the two solutions
(3),(4).

Exercises:
1. Show that the Green’s function is unique. (Hint: take the difference of
two of them.)
2. Find the Green’s function for the tilted half-space Ω = {(x, y) : ax +
by > 0}.
3. Find the Green’s function for the unit square Ω = {(x, y) : 0 < x <
1, 0 < y < 1}.
4. With G(x, x0 ) = − 4π
1 1
|x−x0 |
as the free space Green’s function in R3 ,
construct the Green’s function for the half-space Ω = {(x, y, z) : x, y ∈
R, z > 0}.
5. For the unit sphere Ω in R3 there is an analogue of the reflection point
given for the unit disk from our example 4. What would it be in the
3D case? Construct the Green’s function and write out the solution for
the problem defined in Ω = {(x, y, z) : x2 + y 2 + z 2 < 1}
 2
∇ u = −f (x, y, z) in Ω
u=0 on ∂Ω

Remark: Recall from calculus Green’s second identity

Z Z
2 2
(u∇ G − G∇ u) dx = (u∇G − g∇u) · νds
Ω ∂Ω

8
and again consider the Poisson problem
 2
∇ u = −f (x, y) in Ω ⊂ R2
(5)
u=0 on ∂Ω .

For arbitrary (ξ, η) ∈ Ω, we know that G = G(x, y, ξ, η) is harmonic for all

(x, y) ∈ Ω, (x, y) 6= (ξ, η). Another way to think of G is that it solves the
problem  2
∇ G = −δ(x − ξ, y − η) in Ω
(6)
G=0 on ∂Ω
where the δ is the standard 2D version of the Dirac distribution (see Appendix
D). Then substituting this into Green’s second identity, where u solves (5)
and G solves (6), we have, from the boundary conditions,
Z Z
0= (u∇ · ν − G∇u · ν)ds = (u∇2 G − G∇2 u)dx =
Z ∂Ω Ω

(−u(ξ, η)δ(x − ξ, y − η) + G(x, y, ξ, η)f (ξ, η))dξdη =

Ω Z
−u(x, y) + G(x, y, ξ, η)f (ξ, η)dξdη .
Ω
R
That is, u(x, y) = Ω
G(x, y, ξ, η)f (ξ, η)dξdη.

Summary: Know how to obtain reflection points for a given problem and
construct the Green’s function by the reflection principle. Then know how
to write the solution formula in terms of the Green’s function, being careful
to have the correct limits of integration.

Appendix: Solution of Poisson problem example, page 1

Problem:

uxx + uyy = −f (x, y) in Ω = {(x, y) : 0 < x < π, 0 < y < π}
u=0 on ∂Ω
P∞
Using
P∞ the finite Fourier transform u(x, y) = n=1 bn (y) sin(nx), and f (x, y) =
f
n=1 n (y) sin(nx), then, in the usual manner of finding Fourier coefficients
(and the orthogonality of the set {sin(nx)} on domain (0, π)), bn (y) =

9
2
Rπ Rπ
π 0
u(ξ, y) sin(nξ) dξ, fn (y) = π2 0 f (ξ, y) sin(nξ) dξ. Substituting these
series for u, f into the PDE, we obtain

d 2 bn
− n2 bn = −fn (y) , bn (0) = bn (π) = 0 .
dy 2
A fundamental set of solutions, for any n, for the homogeneous equation is
cosh(ny), sinh(ny); to determine a particular solution to the non-homogeneous
equation by the variation-of-parameters method, let w(y) = w1 (y) sinh(ny)+
w2 cosh(ny). Carrying through the calculation of the method, we find that

1 y 1 y
Z Z
w1 (y) = − fn (η) cosh(nη) dη , w2 (y) = fn (η) sinh(nη) dη ,
n 0 n 0

so putting these back into w(y), we have

1 y
Z
w(y) = fn (η){sinh(nη) cosh(ny) − cosh(nη) sinh(ny)} dη =
n 0
1 y
Z
− fn (η) sinh(n(y − η)) dη
n 0

since sinh(nη) cosh(ny) − cosh(nη) sinh(ny) = − sinh(n(y − η)). The general

solution to the bn (y) equation is now

1 y
Z
bn (y) = An cosh(ny) + Bn sinh(ny) − fn (η) sinh(n(y − η)) dη .
n 0
Apply the boundary conditions:
Rπ
bn (0) = An = 0, and bn (π) = 0 = Bn sinh(nπ) − n1 0
fn (η) sinh(n(π − η)) dη.

10
Hence,
Z π
1
Bn = fn (η) sinh(n(π − η)) dη ⇒
n sinh(nπ) 0

Z π Z y
sinh(ny) 1
bn (y) = fn (η) sinh(n(π − η)) dη − fn (η) sinh(n(y − η)) dη =
n sinh(nπ) 0 n 0

Z y
1
fn (η)[sinh(ny) sinh(n(π − η)) − sinh(n(y − η)) sinh(nπ)] dη+
n sinh(nπ) 0

Z π 
fn (η) sinh(ny) sinh(n(π − η))] dη .
y

Using the hyperbolic function addition formulas to expand sinh(ny) sinh(n(π−

η)) − sinh(n(y − η)) sinh(nπ), it becomes

cosh(ny) sinh(nη) sinh(nπ)−cosh(nπ) sinh(nη) sinh(ny) = sinh(nη) sinh(n(π−y)) ,

so
Z y
1
bn (y) = fn (η) sinh(nη) sinh(n(π − y)) dη +
n sinh(nπ) 0

Z π 
fn (η) sinh(ny) sinh(n(π − η)) dη .
y

Now substitute into this solution the expression for fn (·) given above, giving
us
Z y Z π
2
bn (y) = f (ξ, η) sin(nξ) sinh(nη) sinh(n(π − y)) dξdη +
nπ sinh(nπ) 0 0

Z π Z π 
f (ξ, η) sin(nξ) sinh(ny) sinh(n(π − η)) dξdη .
y 0

11
Therefore,
∞
X
u(x, y) = bn (y) sin(nx)
n=1

Z π Z ∞
yX
2 sin(nx)
= f (ξ, η) sin(nξ) sinh(nη) sinh(n(π − y)) dηdξ +
0 0 n=1
nπ sinh(nπ)

Z π Z ∞
πX
2 sin(nx)
f (ξ, η) sin(nξ) sinh(ny) sinh(n(π − η)) dηdξ
0 y n=1
nπ sinh(nπ)

Z π Z y ∞
X 2 sinh(ny) sinh(nξ)
= f (ξ, η) sin(nx) sin(nξ) dηdξ
0 0 n=1
nπ sinh(nπ)

Z π Z π
+ same except roles of η and y are switched dηdξ .
0 y

Defining G by
 P∞ 2 sinh(nη) sinh(nξ)
 n=1 nπ sinh(nπ)
sin(nx) sin(nξ) if η ≤ y
G(x, y, ξ, η) =
 P∞
n=1 same series with roles of η and y switched if y ≤ η

then we can write u as

Z π Z π
u(x, y) = G(x, y, ξ, η)f (ξ, η) dξdη .
0 0

12