You are on page 1of 20

Partial Differential Equations – MATH 304

Dr. Florian Beyer

Semester 1, 2024
Department of Mathematics and Statistics
University of Otago

Version: 14 March 2024


Contents

0 Introduction and preliminaries 4


0.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
0.2 Partial derivatives and the chain rule . . . . . . . . . . . . . . . . . . . . . 5
Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13

1 Basic concepts 14
1.1 What are partial differential equations? . . . . . . . . . . . . . . . . . . . 14
1.2 Multiple integrals and the Gauss-Green theorem . . . . . . . . . . . . . . 15
1.2.1 Double and multiple integrals . . . . . . . . . . . . . . . . . . . . . 15
1.2.2 Coordinate systems and the transformation formula . . . . . . . . 17
1.2.3 Surfaces and surface integrals . . . . . . . . . . . . . . . . . . . . . 20
1.2.4 Examples: Balls, spheres and other surfaces . . . . . . . . . . . . . 22
1.2.5 The Gauss-Green theorem and its consequences . . . . . . . . . . . 24
1.3 Under which assumptions can all of this be justified? . . . . . . . . . . . . 26

2 Laplace’s and Poisson’s equations I 30


2.1 The equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
2.2 Motivations and applications . . . . . . . . . . . . . . . . . . . . . . . . . 30
2.2.1 Electrostatics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
2.2.2 Fluids and gases . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
2.2.3 Complex analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
2.3 Mean-value formulas for harmonic functions . . . . . . . . . . . . . . . . . 33
2.4 The Dirichlet energy and boundary value problems . . . . . . . . . . . . . 36
2.4.1 Boundary value problems . . . . . . . . . . . . . . . . . . . . . . . 36
2.4.2 Dirichlet energy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
2.4.3 Consequences . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38

3 Laplace’s and Poisson’s equations II 42


3.1 The fundamental solution of Laplace’s equation . . . . . . . . . . . . . . . 42
3.2 The Green’s function method . . . . . . . . . . . . . . . . . . . . . . . . . 46
3.3 Green’s function of the unit ball . . . . . . . . . . . . . . . . . . . . . . . . 50
3.4 Interpretation and illustrations . . . . . . . . . . . . . . . . . . . . . . . . 52
3.5 Separation of variables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54

2
4 The heat equation 59

5 The transport and wave equation 64


5.1 The equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64
5.2 Basic phenomenology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64
5.3 Examples and applications . . . . . . . . . . . . . . . . . . . . . . . . . . . 66
5.4 Initial value problem of the transport equation . . . . . . . . . . . . . . . 66
5.5 Local wave propagation . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68
5.6 Representation formulas . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70
5.6.1 The initial value problem of the wave equation . . . . . . . . . . . 70
5.6.2 D’Alembert’s formula for n = 1 . . . . . . . . . . . . . . . . . . . . 70
5.6.3 Kirchhoff’s formula for n = 3 . . . . . . . . . . . . . . . . . . . . . 73
5.6.4 Poisson’s formula for n = 2 . . . . . . . . . . . . . . . . . . . . . . 76
5.7 Further examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78

6 The method of characteristics 79


6.1 Generalised transport equations . . . . . . . . . . . . . . . . . . . . . . . . 79
6.2 Characteristic equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81
6.3 Initial value problem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83
6.4 Analysis of the characteristic curves . . . . . . . . . . . . . . . . . . . . . 84
6.5 Consequences and applications . . . . . . . . . . . . . . . . . . . . . . . . 90

3
Chapter 3

Laplace’s and Poisson’s equations II

3.1 The fundamental solution of Laplace’s equation


The aim is now to construct solutions of the Dirichlet boundary value problem of Poisson’s
equation. In this section here, we lay the foundation for this by deriving the so-called
fundamental solution of Laplace’s equation.
Since Laplace’s equation is rotationally symmetric (as discussed on one of the assign-
ments), there should exist a particular solution u(x) of Laplace’s equation which only
depends on the distance q
r(x) = x21 + . . . + x2n = |x| (3.1)
of the point x in Rn from the origin 0 = (0, . . . , 0) ∈ Rn . By this we mean a solution
u(x) which can be written as
u(x) = v(r(x)) (3.2)
for some function v which only depends on a single variable. Notice that it is of course
not clear a-priori whether such a function v exists (and is non-trivial) and if yes, on which
domain it can be defined. Nevertheless we attempt now to construct solutions u(x) of
Laplace’s equation of the form (3.2). We assume that v is at least twice differentiable
wherever it is defined. Then the chain rule Eq. (0.15) yields
xi
∂xi u(x) = v ′ (r(x)) p 2
x1 + . . . + x2n
for each i = 1, . . . , n. In the same way, we find
xi
p
x21 + . . . + x2n − xi √
x2i x21 +...+x2n
∂x2i u(x) = v ′′ (r(x)) + v ′ (r(x))
x21 + . . . + x2n x21 + . . . + x2n
x2i x21 + . . . + x2n − x2i
= v ′′ (r(x)) 2 + v ′
(r(x))
x1 + . . . + x2n (x21 + . . . + x2n )3/2
x2i r2 (x) − x2i
= v ′′ (r(x)) + v ′
(r(x)) .
r2 (x) r3 (x)

42
3.1. THE FUNDAMENTAL SOLUTION OF LAPLACE’S EQUATION

Hence,
n Pn 2
Pn 2
X
i=1 xi i=1 (r (x) − x2i )
∆u(x) = ∂x2i u(x) ′′
= v (r(x)) ′
+ v (r(x))
r2 (x) r3 (x)
i=1
r2 (x) nr2 (x) − r2 (x)
= v ′′ (r(x)) + v ′
(r(x))
r2 (x) r3 (x)
n−1
= v ′′ (r(x)) + v ′ (r(x)) .
r(x)
It follows that if v : (0, ∞) → R is a solution of the ODE
n−1 ′
v ′′ (r) +v (r) = 0, (3.3)
r
then u(x) = v(r(x)) is a solution of Laplace’s equation defined on the domain Rn \{0}.
Observe here that Eq. (3.3) is singular at r = 0 and hence we must exclude the origin 0
of Rn since r(0) = 0 (see Eq. (3.1)).
It is straightforward to solve the ODE (3.3). Ignoring the trivial case n = 1, the general
solution is (
b ln r + c n = 2,
v(r) = b
(3.4)
rn−2
+ c n ≥ 3,
for r > 0 and for arbitrary constants c and b. We see that, unless we make the special
choice b = 0 and v is therefore constant, these solutions are all singular at r = 0. The
type of singularity depends on the dimension n. Setting b = −1/(2Vol2 ) = −1/(2π) and
c = 0 in the case n = 2 (recall that Vol2 = π), or, b = 1/n(n − 2)Voln and c = 0 for
n ≥ 3 in Eq. (3.4) and plugging this into Eq. (3.2) defines the so-called fundamental
solution of Laplace’s equation.
Definition 3.1. The function
(
1
− 2Vol ln |x| n = 2,
Φ(x) := 2
1 (3.5)
n(n−2)Voln |x|n−2
n ≥ 3,

which is defined for all x ∈ Rn \{0} is called fundamental solution of Laplace’s


equation.
Observe that Φ(x) is a particular solution of Laplace’s equation of the form Eq. (3.2)
which is smooth everywhere on Rn except at the origin 0. The motivation for choosing
these particular values for b and c in Eq. (3.4) will become apparent shortly.
As a preparation for the following discussion, we need to derive a few technical prop-
erties of the fundamental solution Φ(x).
Lemma 3.2.
Let U be a region in Rn and f : U → R a ∂-smooth function (not necessarily harmonic).
Pick ξ ∈ U and ϵ > 0 so that B(ξ, ϵ) ⊂ U . Then
ˆ
DΦ(x − ξ) · ν(x)f (x)dS(x) = − f (x)dS(x), (3.6)
∂B(ξ,ϵ) ∂B(ξ,ϵ)

43
CHAPTER 3. LAPLACE’S AND POISSON’S EQUATIONS II

where ν(x) is an outward pointing unit normal of ∂B(ξ, ϵ) given by, for example,
Eq. (1.43).

Observe carefully that when we write DΦ(x − ξ) we mean the vector-valued function
DΦ evaluated at x − ξ. Recall from Eq. (3.5) that Φ is singular only at the origin x = 0.
The integrand on the left side of Eq. (3.6) is therefore singular only at x = ξ (which is
not in ∂B(ξ, ϵ)) and hence the surface integral over ∂B(ξ, ϵ) is well-defined and finite.
Since the same argument that we used in Step 3 of the proof of Theorem 2.1 yields

f (ξ) = lim f (x)dS(x),


ϵ→0+ ∂B(ξ,ϵ)

we can conclude from Eq. (3.6) that


ˆ
f (ξ) = − lim DΦ(x − ξ) · ν(x)f (x)dS(x). (3.7)
ϵ→0+ ∂B(ξ,ϵ)

Exercise 3.1.
Prove Lemma 3.2.

We also consider the following estimates.


Lemma 3.3.
Let U be a region in Rn and f : U → R a ∂-smooth function (not necessarily harmonic).
Fix an arbitrary ξ ∈ U . Then, for every ϵ ∈ (0, 1] which is sufficiently small so that
B(ξ, ϵ) ⊂ U , we have
ˆ (
M ϵ |ln ϵ|, n = 2,
f (x)Φ(x − ξ)dS(x) ≤ M
(3.8)
∂B(ξ,ϵ) n−2 ϵ, n ≥ 3,

where M := maxx∈Ū |f (x)|. Moreover, the following integral is well-defined and satisfies
the estimate
ˆ (
M 2
ϵ (1 − 2 ln ϵ) , n = 2,
f (x)Φ(x − ξ)dx ≤ 4 M 2 (3.9)
B(ξ,ϵ) 2(n−2) ϵ , n ≥ 3.

This lemma is a statement about the behaviour of the integrals on the left sides of
Eqs. (3.8) and (3.9) as functions of ϵ for fixed ξ. These estimates are not supposed to be
particularly “accurate”. The point is that they guarantee that both integrals on the left
sides are always well-defined and finite despite the singularity of the function Φ(x − ξ) at
x = ξ. Especially in Eq. (3.9) this is not trivial in since the singularity x = ξ is indeed
in the integration domain. We conclude that
ˆ ˆ
lim f (x)Φ(x − ξ)dS(x) = lim f (x)Φ(x − ξ)dx = 0. (3.10)
ϵ→0+ ∂B(ξ,ϵ) ϵ→0+ B(ξ,ϵ)

44
3.1. THE FUNDAMENTAL SOLUTION OF LAPLACE’S EQUATION

Proof of Lemma 3.3. Let ξ ∈ U be fixed. Let us start by analysing the function
ˆ
I : (0, r) → R, I(ϵ) = f (x)Φ(x − ξ)dS(x)
∂B(ξ,ϵ)

which agrees with the integral on the left side of Eq. (3.8). Here, r > 0 is any sufficiently
small constant so that B(ξ, r) ⊂ U . For each value ϵ ∈ (0, r), this integral is well-defined
and finite since the integration variable x never equals ξ.
Let us now derive the estimate Eq. (3.8). Because Φ(x) only depends on the distance
|x| from the origin, see Eq. (3.5), it follows that the function Φ(x − ξ) is constant on the
surface ∂B(ξ, ϵ) and its value is 1/(nVoln (n − 2)ϵn−2 ) for n ≥ 3 (let us restrict to this
case now; the case n = 2 is treated in Exercise 3.2). We hence find
ˆ
1
I(ϵ) = f (x)dS(x).
nVoln (n − 2)ϵn−2 ∂B(ξ,ϵ)
We rewrite this as
ˆ
ϵ 1 ϵ
I(ϵ) = f (x)dS(x) = f (x)dS(x).
n − 2 nVoln ϵn−1 ∂B(ξ,ϵ) n−2 ∂B(ξ,ϵ)

Now we wish to estimate |I(ϵ)|. To this end, we use standard basic properties of integrals
(see footnote 4 on page 41) to find

ϵ ϵ
|I(ϵ)| = f (x)dS(x) ≤ |f (x)|dS(x).
n−2 ∂B(ξ,ϵ) n−2 ∂B(ξ,ϵ)

Since f is ∂-smooth, the function |f (x)| is continuous on Ū and hence attains its absolute
maximum
ffl value M on Ū . This implies that the mean value of |f (x)| over ∂B(ξ, ϵ), i.e.,
∂B(ξ,ϵ) |f (x)|dS(x), must be at most M , and we have therefore found
ϵ
|I(ϵ)| ≤ M. (3.11)
n−2
This completes the proof of Eq. (3.8) in the case n ≥ 3. Now in order to show Eq. (3.9)
(in the case n ≥ 3 only here), we first realise that the function I(ϵ) is continuous1 .
Hence |I(ϵ)| is continuous. Since it is bounded as a consequence of Eq. (3.11) it can be
integrated with respect to ϵ. This yields
ˆ ϵ
M ϵ2
|I(ϵ̃)|dϵ̃ ≤ .
0 2(n − 2)
Since this is finite for all ϵ in (0, r), it makes sense to use Eq. (1.41) and we find
ˆ ˆ ϵ ˆ ϵ
M ϵ2
f (x)Φ(x − ξ)dx = I(ϵ̃)dϵ̃ ≤ |I(ϵ̃)| dϵ̃ ≤ .
B(ξ,ϵ) 0 0 2(n − 2)
1
One can use a theorem from integration theory which, under the given conditions here, asserts that
the integral depending on the variable ϵ is a continuous function of that variable. Again, we are not able
to discuss the technical details here.

45
CHAPTER 3. LAPLACE’S AND POISSON’S EQUATIONS II

Figure 3.1: Illustration of the region Uϵ used in Section 3.2.

This completes the proof in the case n ≥ 3.


Exercise 3.2.
Prove Lemma 3.3 in the case n = 2.

3.2 The Green’s function method


We will now use the fundamental solution of Laplace’s equation to first derive a useful
integral identity for any ∂-smooth function u (not necessarily harmonic etc.) on any
region U ⊂ Rn . To this end, fix ξ ∈ U and choose any ϵ > 0 which is sufficiently small
so that B(ξ, ϵ) ⊂ U . Then define Uϵ := U \B(ξ, ϵ); see Figure 3.1. We do not prove here
that Uϵ is also a region in Rn in the sense of Definition 1.5. Let us consider the integral
ˆ
(u(x)∆(Φ(x − ξ)) − Φ(x − ξ)∆u(x)) dx. (3.12)

Observe carefully here that because we integrate over Uϵ (and not over U ) the integrand is
regular on the integration domain. Recall that when we write ∆Φ(x − ξ) (or DΦ(x − ξ))
we mean the function ∆Φ (or DΦ) evaluated at x − ξ. However, when we write ∆(Φ(x −
ξ)) (or D(Φ(x − ξ))) as in Eq. (3.12), we actually define a new function Φ̃(x) := Φ(x − ξ)
(often not written down explicitly) and apply the Laplacian ∆ (or the gradient D) to Φ̃,
i.e.,
∆(Φ(x − ξ)) = ∆Φ̃(x), D(Φ(x − ξ)) = DΦ̃(x).

46
3.2. THE GREEN’S FUNCTION METHOD

This is quite similar to what is discussed at the very end of Chapter 0. In this case here
we can now show using the chain rule that these two different ways to apply D and ∆
to these functions are in fact the same

∆(Φ(x − ξ)) = ∆Φ(x − ξ) D(Φ(x − ξ)) = DΦ(x − ξ). (3.13)

In general, this would not always be the case. In any case, we can now apply Green’s
formula Eq. (1.48) to write
ˆ
(u(x)∆(Φ(x − ξ)) − Φ(x − ξ)∆u(x)) dx

ˆ ˆ (3.14)
= D(Φ(x − ξ)) · ν(x) u(x) dS(x) − Du(x) · ν(x) Φ(x − ξ) dS(x),
∂Uϵ ∂Uϵ

where ν(x) is an outward pointing unit normal of ∂Uϵ .


With the help of Figure 3.1, we notice that ∂Uϵ consists of two disconnected pieces,
namely ∂U and ∂B(ξ, ϵ). On ∂U , the vector field ν(x) is an outward pointing unit
normal of ∂U . On ∂B(ξ, ϵ), however, it agrees with the inward pointing unit normal of
∂B(ξ, ϵ). We keep this in mind in the following discussion. We now split up the boundary
terms in Eq. (3.14). While we write this down, we use Eq. (3.13) and give special names
to certain terms as follows:
ˆ ˆ
u(x)∆Φ(x − ξ)dx − Φ(x − ξ)∆u(x)dx
| Uϵ {z } | Uϵ {z }
=:I =:II
ˆ ˆ
= DΦ(x − ξ) · ν(x) u(x) dS(x) − Du(x) · ν(x) Φ(x − ξ) dS(x) (3.15)
∂B(ξ,ϵ) ∂B(ξ,ϵ)
| {z } | {z }
=:III =:IV
ˆ ˆ
+ DΦ(x − ξ) · ν(x) u(x) dS(x) − Du(x) · ν(x) Φ(x − ξ) dS(x).
∂U ∂U

Term I is zero for all ϵ > 0 because Φ is harmonic, i.e., the function ∆Φ(x) vanishes
for all x ∈ Rn \{0}. Regarding II, we first notice that it is well-defined and finite since
we do not integrate over the singularity at ξ = x. Moreover we can write,
ˆ ˆ
Φ(x − ξ)∆u(x)dx = II + Φ(x − ξ)∆u(x)dx.
U B(ξ,ϵ)

This is a meaningful decomposition of the potentially singular integral on the left-hand


side because also the second term on the right hand side is well-defined and finite as a
consequence of Lemma 3.3. In fact Lemma 3.3 implies that the last integral approaches
zero in the limit ϵ → 0+ . Hence
ˆ
lim II = Φ(x − ξ)∆u(x)dx.
ϵ→0+ U

47
CHAPTER 3. LAPLACE’S AND POISSON’S EQUATIONS II

In term III, the integral needs to be evaluated with respect to the inward pointing unit
normal ν(x) of ∂B(ξ, ϵ). By changing the sign in Lemma 3.2 we find

III = u(x)dS(x).
∂B(ξ,ϵ)

In the limit ϵ → 0+ , we use Eq. (3.7) with the opposite sign and find

lim III = u(ξ).


ϵ→0+

Concerning the term IV , it follows from Eq. (3.10) that

lim IV = 0.
ϵ→0+

In total, by rearranging Eq. (3.15) and taking the limit ϵ → 0+ , we have established that

ˆ ˆ
u(ξ) = Du(x) · ν(x) Φ(x − ξ)dS(x) − DΦ(x − ξ) · ν(x) u(x)dS(x)
ˆ
∂U ∂U
(3.16)
− Φ(x − ξ)∆u(x)dx.
U

This integral identity holds for all ∂-smooth functions u : U → R.


Suppose now that u is some so far unknown solution of the Dirichlet boundary value
problem of Poisson’s equation with ∂-smooth source term f and boundary data g. If we
could evaluate the right-hand side of Eq. (3.16) for any ξ ∈ U , then Eq. (3.16) would
be a formula to compute the value of the unknown solution u at ξ. Since u satisfies
this boundary value problem, we have enough information to compute, in principle, the
second and third integral of Eq. (3.16) because the fundamental solution Φ is given
explicitly by Eq. (3.5), the value of u equals the given function g on ∂U (which allows
us to compute the second integral), and ∆u equals the known function −f (which allows
us to compute the third integral). However, since u is unknown, we do not have enough
information to calculate the first integral since we do not know Du · ν on the boundary.
Let us therefore try to “get rid of this first integral” (which we refer to as the “bad term”
of (3.16) in the following).
Let us start again from an arbitrary region U in Rn (it is not yet necessary to choose
boundary data, source terms etc.). For each ξ ∈ U we consider the Dirichlet boundary
value problem (
∆Kξ (x) = 0, for all x ∈ U ,
(3.17)
Kξ (x) = Φ(x − ξ), for all x ∈ ∂U ,
for a ∂-smooth unknown function Kξ : U → R whose index ξ reminds us that its defining
properties Eq. (3.17) depend on the choice of ξ. Notice here that since ξ ∈ U , the vector
x − ξ does not vanish on the boundary x ∈ ∂U . The boundary condition in Eq. (3.17) is
therefore well-defined. If the boundary value problem Eq. (3.17) has a ∂-smooth solution

48
3.2. THE GREEN’S FUNCTION METHOD

Kξ for each ξ ∈ U , we call Kξ corrector function of U and define Green’s


function of U as
G(ξ, x) := Φ(x − ξ) − Kξ (x), (3.18)
for each ξ ∈ U and for all x ∈ Ū \{ξ}. It is important to notice that this function is
singular at x = ξ and that this singularity is caused by the presence of Φ(x − ξ) in
the definition Eq. (3.18); recall that Kξ (x) is ∂-smooth and therefore completely regular
everywhere.
´ In any case, it is a consequence of Lemma 3.3, in particular Eq. (3.9), that
U G(ξ, x)f (x)dx is well-defined and finite for any ∂-smooth function f : U → R.
The significance of Eqs. (3.17) and (3.18) becomes clear when we return to the task of
removing the “bad term” in Eq. (3.16) now. By means of Green’s formula Eq. (1.48) we
find for any fixed ξ ∈ U :
ˆ ˆ
Kξ (x)∆u(x)dx − ∆Kξ (x)u(x)dx
U ˆ U ˆ
= Du(x) · ν(x) Kξ (x)dS(x) − DKξ (x) · ν(x) u(x)dS(x).
∂U ∂U

We reemphasise that Kξ is considered as a function of x only (not of ξ) and hence


DKξ = (∂x1 Kξ , . . . , ∂xn Kξ ). Now, since Kξ is a solution of Eq. (3.17), the second
integral on the left side vanishes and in the first integral on the right side, we can replace
Kξ (x) by Φ(x − ξ):
ˆ ˆ ˆ
Kξ (x)∆u(x)dx = Du(x)·ν(x) Φ(x−ξ)dS(x)− DKξ (x)·ν(x) u(x)dS(x).
U ∂U ∂U

The first term on the right-hand side is our “bad term” which can hence be expressed as
ˆ ˆ ˆ
Du(x)·ν(x) Φ(x−ξ)dS(x) = Kξ (x)∆u(x)dx+ DKξ (x)·ν(x) u(x)dS(x).
∂U U ∂U

This allows us now to finally eliminate the “bad term” in Eq. (3.16) as follows
ˆ ˆ
u(ξ) = Kξ (x)∆u(x)dx + DKξ (x) · ν(x) u(x)dS(x)
Uˆ ∂U ˆ
− DΦ(x − ξ) · ν(x) u(x)dS(x) − Φ(x − ξ)∆u(x)dx.
∂U U

Thanks to the definition of Green’s function Eq. (3.18), we have therefore found
ˆ ˆ
u(ξ) = − DG(ξ, x) · ν(x) u(x)dS(x) − G(ξ, x)∆u(x)dx. (3.19)
∂U U

We recall here that both integrals in Eq. (3.19) are well-defined and finite despite the
singularity of Green’s function at x = ξ. In writing Eq. (3.19) we understand that
n
X ∂G(ξ, x)
DG(ξ, x) = Dx G(ξ, x) = , (3.20)
∂xi
i=1

49
CHAPTER 3. LAPLACE’S AND POISSON’S EQUATIONS II

i.e., derivatives of G are taken with respect to the variables x = (x1 , . . . , xn ) only, but not
with respect to ξ = (ξ1 , . . . , ξn ). Notice that Eq. (3.19) holds for any ∂-smooth function
u : U → R, whether u satisfies Poisson’s equation or not. However, if u is a solution
the Dirichlet boundary value problem of Poisson’s equation, then Eq. (3.19) becomes a
useful representation formula for u.
Theorem 3.4.
Let U ⊂ Rn be a region. Suppose Green’s function G(ξ, x) of the region U is known.
Then every ∂-smooth solution u : U → R of the Dirichlet boundary value problem
(
−∆u(x) = f (x), for all x ∈ U ,
u(x) = g(x), for all x ∈ ∂U ,
with ∂-smooth source term f : U → R and ∂-smooth boundary data g : U → R can be
represented as
ˆ ˆ
u(ξ) = − DG(ξ, x) · ν(x) g(x)dS(x) + G(ξ, x)f (x)dx, (3.21)
∂U U

for each ξ ∈ U . Here, ν(x) is an outward pointing unit normal of ∂U .

This theorem yields the representation formula (3.21) for solutions u of the Dirichlet
boundary value problem of Poisson’s equation. This formula, however, is useful only if
Green’s function G is known for the given region U (and sufficiently simple so that the
integrals can be calculated in practice). It turns out that this is not always the case even
for simple regions U . In Section 3.3 we discuss an example now where it is.

3.3 Green’s function of the unit ball


As a concrete example, let us consider the “simple” case U = B(0, 1) in Rn in this section.
For this region, one can indeed find Green’s function explicitly. As we will discuss now,
Green’s function is
(
Φ(x − ξ) − Φ(|ξ|x − ξ/|ξ|), for ξ ∈ B(0, 1)\{0},
G(ξ, x) = (3.22)
Φ(x) − Cn , for ξ = 0,

for all ξ ∈ B(0, 1) and x ∈ B(0, 1)\{ξ} where


(
0 n = 2,
Cn := 1 (3.23)
n(n−2)Voln , n ≥ 3.

Let us demonstrate that Eq. (3.22) is indeed the correct formula for Green’s function
of U . According to Eqs. (3.18) and (3.22), the task is to check that
(
Φ(|ξ|x − ξ/|ξ|), for ξ ∈ B(0, 1)\{0},
Kξ (x) = (3.24)
Cn , for ξ = 0,
is the corrector function of U for each ξ ∈ U :

50
3.3. GREEN’S FUNCTION OF THE UNIT BALL

1. ∂-Smoothness: First we demonstrate that Kξ : U → R is ∂-smooth for each ξ ∈ U .


In the case ξ = 0, we have that Kξ (x) = const which is a ∂-smooth function
trivially. Let now ξ ̸= 0. We first recall from Eq. (3.5) that Φ is singular only
at the origin. We therefore show now that |ξ|x − ξ/|ξ| = 0 implies that either
x ̸∈ B(0, 1) or ξ ̸∈ B(0, 1). In fact, this vector vanishes if and only if

x = ξ/|ξ|2 .

This has the particular consequence that

|x| = 1/|ξ|.

So if |ξ| < 1, then |x| > 1, or, if |x| ≤ 1, then |ξ| ≥ 1. We have therefore established
that the function Kξ : U → R in Eq. (3.24) is ∂-smooth for each ξ ∈ B(0, 1).

2. Harmonicity: Next we establish that Kξ is harmonic for each ξ ∈ B(0, 1) as the


first part of showing that it satisfies the boundary value problem Eq. (3.17). In
the case ξ = 0, the function Kξ is a constant function and is therefore clearly
harmonic. In the case ξ ̸= 0, this is a consequence of the following exercise.

Exercise 3.3.
Pick any ξ ∈ Rn \{0}. Show that the function Φ̃(x) = Φ(|ξ|x − ξ/|ξ|) is non-
singular on Rn \{ξ/|ξ|2 } and satisfies Laplace’s equation there.

We have already seen in the previous step that ξ/|ξ|2 is not in B(0, 1) if ξ is in
B(0, 1). This proves that Kξ is harmonic for each ξ ∈ U .

3. Boundary data: Finally we confirm that Kξ satisfies the correct boundary condition

Kξ (x) = Φ(x − ξ) for x ∈ ∂B(0, 1),

as the second part of verifying the boundary value problem Eq. (3.17). In the case
ξ = 0, this is to show that Cn = Φ(x) for all x ∈ ∂B(0, 1), which follows easily
from Eqs. (3.5) and (3.23). If ξ ∈ B(0, 1)\{0}, we need to show that

Φ(|ξ|x − ξ/|ξ|) = Φ(x − ξ) for all x ∈ ∂B(0, 1). (3.25)

Because Φ only depends on the distance from the origin, recall Eq. (3.5), it suffices
to show that
|ξ|x − ξ/|ξ| = |x − ξ| (3.26)

for all x ∈ ∂B(0, 1). The square of the left side is


2
|ξ|x − ξ/|ξ| = |ξ|2 |x|2 − 2x · ξ + 1 = |ξ|2 − 2x · ξ + 1,

51
CHAPTER 3. LAPLACE’S AND POISSON’S EQUATIONS II

where we have used |x| = 1 in the last step as a consequence of x ∈ ∂B(0, 1). The
square of the right side of Eq. (3.26) is

|x − ξ|2 = |x|2 − 2x · ξ + |ξ|2 = 1 − 2x · ξ + |ξ|2

where we have used |x| = 1 in the last step again. This establishes Eq. (3.25) and
hence confirms that Eq. (3.24) satisfies the correct boundary condition.
We can thus conclude that Eq. (3.22) is indeed Green’s function of U = B(0, 1).

Exercise 3.4.
Show
1 1 − |ξ|2
DG(ξ, x) · ν(x) = − , (3.27)
nVoln |x − ξ|n
for each ξ ∈ B(0, 1) and all x ∈ ∂B(0, 1), where G is Green’s function of the unit ball
given in Eq. (3.22). Recall Eq. (3.20).

The result of this exercise can be used in Eq. (3.21). In the case f = 0 for example, this
yields ˆ
u(ξ) = P (ξ, x)g(x)dS(x), (3.28)
∂B(0,1)

for all ξ ∈ B(0, 1), where P (ξ, x) is so-called Poisson’s kernel for the ball B(0, 1)
derived from (3.27) given by
1 1 − |ξ|2
P (ξ, x) = , (3.29)
nVoln |x − ξ|n
defined for each ξ ∈ B(0, 1) and all x ∈ ∂B(0, 1). We have thus found that Eqs. (3.28)
and (3.29) represent any ∂-smooth solution of the Dirichlet boundary value problem of
Laplace’s equation with ∂-smooth boundary data g for the region U = B(0, 1) in Rn .
We will discuss applications and examples in class and in the assignments.

3.4 Interpretation and illustrations


Green’s function G(ξ, x) is often interpreted as the solution of the Dirichlet boundary
value problem of Laplace’s equation with zero boundary data which has a “puncture” (or
“singularity”) at x = ξ. This is often written for an arbitrary fixed ξ ∈ U as follows
(
−∆x G(ξ, x) = δξ (x), for all x ∈ U ,
(3.30)
G(ξ, x) = 0, for all x ∈ ∂U .

Here ∆x denotes the Laplacian with respect to the x-variables, i.e.,


n
X ∂ 2 G(ξ, x)
∆x G(ξ, x) = . (3.31)
i=1
∂x2i

52
3.4. INTERPRETATION AND ILLUSTRATIONS

The so-called Dirac δ-distribution δξ (x) is quite a complicated mathematical object,


which we are not going to define here. The only thing we need to know about it here is
that δξ (x) = 0 for all x ̸= ξ. This means that ∆x G(ξ, x) = 0 for all x ̸= ξ, which means
that G is therefore a “punctured solution of Laplace’s equation” as required.

1.5
1.5

7
.0
-0
0
.1

1.0
-0

1.0

-0
.0
7
-0
.0
3
0.
04 0.5
0.5 7

-0
0.0 -0

.1
.1
8

5
0.

0.1
1

0.
2

19 8

0.22
0.3
23

x2
0 0.0
0.
x2

0.0 0.1

1
-0.2

0.1
5 2
0.26

04
0.
5
0.1 −0.5 -0
0.

.1
08

1
−0.5
0.
00

0.0
1
−1.0
−1.0
7
.0
-0

-0.04
-0.0
-0.1

7
−1.5
4

−1.5 −2.0 −1.5 −1.0 −0.5 0.0 0.5 1.0 1.5 2.0
−2.0 −1.5 −1.0 −0.5 0.0 0.5 1.0 1.5 2.0 x1
x1

Figure 3.2: Contour plot of Φ(x − ξ), i.e., Figure 3.3: Contour plot of G(ξ, x) inside
the first term in the definition of Green’s and outside the region U = B(0, 1) in R2
function, inside and outside the region U = for ξ = (3/4, 0). Comparing this with Fig-
B(0, 1) in R2 for ξ = (3/4, 0). ure 3.2 illustrates the effect of the corrector
function.

Let us illustrate what this means in the case U = B(0, 1) in R2 . Recall from Eq. (3.22)
that in this case
(
Φ(x − ξ) − Φ(|ξ|x − ξ/|ξ|), for ξ ∈ B(0, 1)\{0},
G(ξ, x) = (3.32)
Φ(x) − Cn , for ξ = 0,

for constants Cn given by Eq. (3.23). We see that the “puncture” at x = ξ mentioned
above is generated by the first term only. In Figure 3.2 we provide a contour plot of just
this first term for ξ = (3/4, 0) ∈ U . It is clear that while this first term by itself satisfies
the first condition in Eq. (3.30), it in general violates the second condition there, i.e., the
boundary condition. The role of the corrector function Kξ (x), i.e., the second term in
Eq. (3.32), is now to enforce the correct boundary values, see Figure 3.3. The presence
of this second term distorts the contour lines of the first term so that the 0-contour line
agrees with the boundary. In this way, both conditions in Eq. (3.30) are satisfied. With
the original puncture still at ξ = (3/4, 0) as in Figure 3.2, Figure 3.3 demonstrates that
the corrector function is nothing but a second punctured solution of Laplace’s equation
with the puncture located at exactly the right place outside the region U (at the point
ξ/|ξ|2 = (4/3, 0) in this case).

53
CHAPTER 3. LAPLACE’S AND POISSON’S EQUATIONS II

3.5 Separation of variables


Next we discuss a different method to solve the Dirichlet boundary value problem of
Laplace’s equation. The idea of so-called separation of variables is to write the
unknown u(x) in the form of a product

u(x) = u1 (x1 ) · · · un (xn ), (3.33)

for n unknown functions u1 , . . . , un , each of which depends only on a single variable


x1 , . . . , xn , respectively. It is of course a-priori not clear at all if there are any solutions
of Laplace’s equation of this particular form Eq. (3.33). In any case, when we plug
Eq. (3.33) into Laplace’s equation, we get

∆u(x) =u′′1 (x1 )u2 (x2 ) · · · un (xn ) + u1 (x1 )u′′2 (x2 ) · · · un (xn )
+ . . . + u1 (x1 )u2 (x2 ) · · · u′′n (xn ) = 0.

Under the assumption that none of these functions vanishes, it follows

u′′1 (x1 ) u′′ (xn )


+ ... + n = 0. (3.34)
u1 (x1 ) un (xn )

Let us write this as


u′′1 (x1 ) u′′ (x2 ) u′′ (xn )
=− 2 − ... − n .
u1 (x1 ) u2 (x2 ) un (xn )
When we vary the variable x1 but keep all the other variables x2 , . . . , xn fixed, then the
right side of this equation is constant. But this means that the left side, being a function
of x1 only, must be a constant, too. There therefore must exist a constant c1 , which we
allow to be complex in general at this stage, such that

u′′1 (x1 )
= c21 . (3.35)
u1 (x1 )

The same argument applied to all the other functions ui (xi ) for i = 2, . . . , n yields that
there must also exist constants ci (again allowed to be complex for the time being) such
that
u′′i (xi )
= c2i , for all i = 1, . . . , n. (3.36)
ui (xi )
Laplace’s equation in the form (3.34) therefore becomes

c21 + . . . + c2n = 0. (3.37)

In summary, by writing the unknown u(x) as the product Eq. (3.33), we have there-
fore decomposed Laplace’s equation into n ordinary differential equations of the form
Eq. (3.36) with so far unspecified complex constants c1 , . . . , cn subject to the algebraic
condition Eq. (3.37).

54
3.5. SEPARATION OF VARIABLES

Figure 3.4: The boundary value problem for U = (0, π)3 discussed in Section 3.5.

The general solution of each ordinary differential equation (3.36) is


(
Aj ecj xj + Bj e−cj xj if cj ̸= 0,
uj (xj ) = (3.38)
Aj + Bj xj if cj = 0,

for arbitrary, so far, complex constants Aj and Bj , for each j = 1, . . . , n. It is clear that,
since u(x) given by Eq. (3.33) is supposed to be real, the (so far free) complex constants
Aj and Bj will in general need to satisfy reality conditions depending on the choices of
c1 , . . . , cn and the particular choice of boundary conditions on ∂U .
Let us for the remainder of this discussion now restrict to the following relatively
simple exemplary Dirichlet boundary value problem. To this end, we restrict to the
region U = (0, π)3 , i.e., the “cube” or “box” in R3 with side length π (the reason for
choosing this particular side length will become clear in a moment); see Figure 3.4. As
the boundary conditions, we only consider the case here where the unknown u(x) is
zero on all faces of the cube apart from the face given by x3 = π (the blue face in
Figure 3.4). On this latter face, the unknown is supposed to agree with some given
smooth function g(x1 , x2 ). In order to match this boundary condition to the ones on
the adjacent faces, the function g is required to vanish at the boundary of that face so
that the total boundary function (previously labelled g) has a chance to be ∂-smooth as
required. We could imagine that the cube represents a box which is immersed into ice
water (at a temperature of zero degrees) so that only the upper face in Figure 3.4 is not
submerged. The boundary data function g can then be interpreted as the air temperature
function and the unknown u as the temperature function inside the box.

55
CHAPTER 3. LAPLACE’S AND POISSON’S EQUATIONS II

These boundary conditions imply that u1 (0) = u1 (π) = 0 where u1 is the first function
in Eq. (3.33). We conclude from Eq. (3.38) that the only solution corresponding to
the case c1 = 0 would be u1 (x1 ) = 0 for all x1 ∈ [0, π]. According to Eq. (3.33) the
function u would therefore need to vanish everywhere which is not compatible with all
the boundary conditions above (unless the function g happens to vanish). We therefore
have to consider the case c1 ̸= 0 in Eq. (3.38). The condition u1 (0) = 0 then implies that
A1 + B1 = 0 and hence B1 = −A1 . The condition u1 (π) = 0 gives

0 = A1 (ec1 π − e−c1 π ) = A1 e−c1 π (e2c1 π − 1).

We again ignore the trivial case given by A1 = 0 and hence conclude that

e2c1 π = 1.

This implies that


c1 = im1
for an arbitrary m1 ∈ Z where i is the imaginary unit. Imposing that the function u1 is
real, we find
u1 (x1 ) = Ã1 sin(m1 x1 )
for any Ã1 ∈ R and any positive integer m1 .
Now, since u2 (x2 ) is supposed to satisfy the same boundary conditions, we find in the
same way
u2 (x2 ) = Ã2 sin(m2 x2 )
for Ã2 ∈ R and any positive integer m2 . Eq. (3.37) therefore implies

c23 = −c21 − c22 = m21 + m22 .


p
The constant c3 = ± m21 + m22 is therefore real. The boundary condition u3 (0) = 0
implies together with Eq. (3.38) that A3 = −B3 and therefore, imposing that u3 is a real
function, q 
u3 (x3 ) = Ã3 sinh m21 + m22 x3 ,

for Ã3 ∈ R. Let us ignore the boundary condition on the remaining face x3 = π for the
moment. In total we have therefore constructed a solution of Laplace’s equation of the
form q 
u(x) = A sin m1 x1 sin m2 x2 sinh m21 + m22 x3 , (3.39)

for any A ∈ R and arbitrary positive integers m1 and m2 , which satisfies the first five of
our six boundary conditions.
It is clear that the remaining sixth boundary condition at x3 = π cannot be satisfied
by the particular solution Eq. (3.39) unless the boundary data function g happens to be

56
3.5. SEPARATION OF VARIABLES

very special. In order to address this issue we proceed as follows now. Since Laplace’s
equation is linear, the following function
N
X N
X q 
u(x) = Am1 ,m2 sin m1 x1 sin m2 x2 sinh m21 + m22 x3 , (3.40)
m1 =1 m2 =1

given by an arbitrary collection of real constants Am1 ,m2 is also a solution of Laplace’s
equation, and, it also satisfies the same five boundary conditions as before. The quantity
N here is an arbitrary positive integer. The remaining boundary condition at the face
x3 = π therefore yields the condition
g(x1 , x2 ) = u(x1 , x2 , π)
N N
(3.41)
X X q 
= Am1 ,m2 sin m1 x1 sin m2 x2 sinh m21 + m22 π .
m1 =1 m2 =1

cf. Eq. (3.40) evaluated at x3 = π.


The question is now, given the boundary data function g as above, can we always find
constants Am1 ,m2 ∈ R and an integer N > 0 so that Eq. (3.41) holds for all x1 , x2 ∈ [0, π]?
If yes, then Eq. (3.40) with this choice of constants Am1 ,m2 and integer N is indeed a
solution of our Dirichlet boundary value problem! (In fact, it is then the unique solution
because there can be at most one; cf. Theorem 2.3).
The answer to this question is “almost yes” and it is one of the major results of the
theory of Fourier analysis which we do not have much time to go into here. In
general though, we must allow N to go to infinity N → ∞ in Eqs. (3.40) in order to
satisfy (3.41). One can show that if g(x1 , x2 ) is a ∂-smooth function which vanishes
whenever x1 = 0, x1 = π, x2 = 0 or x2 = π, one can find constants Bm1 ,m2 ∈ R for all
integers m1 , m2 ≥ 1 such that

X
g(x1 , x2 ) = Bm1 ,m2 sin m1 x1 sin m2 x2 (3.42)
m1 ,m2 =1

holds at each (x1 , x2 ) ∈ [0, π]2 . These constants turn out to be uniquely determined as
we will see below. The infinite series (3.42) converges uniformly, and, one can swap the
sums with differentiation and integration.

Exercise 3.5.
Let k1 and k2 be arbitrary integers. Prove the orthogonality relation
ˆ π (
0 for k1 ̸= k2
sin(k1 x) sin(k2 x)dx = π .
0 2 for k1 = k2

Use this to show that the constants Bm1 ,m2 in Eq. (3.42) can be calculated as
ˆ πˆ π
4
Bm1 ,m2 = 2 g(x1 , x2 ) sin(m1 x1 ) sin(m2 x2 )dx1 dx2 (3.43)
π 0 0

57
CHAPTER 3. LAPLACE’S AND POISSON’S EQUATIONS II

under the previous conditions for all positive integers m1 , m2 . You can assume here
without justification that you can commute the infinite sums in Eq. (3.42) with integ-
ration.
Eq. (3.43) is a so-called Fourier integral. It is the basis for general Fourier series and,
even more generally, Fourier transforms.

Eq. (3.43) is the general formula which allows to find the constants Bm1 ,m2 so that
Eq. (3.42) holds for the given boundary data function g. In some examples, only a finite
number of these constants may be non-zero (in which case one can read them off directly
without using Eq. (3.43)). In any case, once the constants Bm1 ,m2 have been determined,
Eq. (3.41) implies that if we choose

B
Am1 ,m2 = pm1 ,m2  (3.44)
sinh m21 + m22 π

for all positive integers m1 and m2 , then the boundary condition Eq. (3.41) holds. Using
these constants Am1 ,m2 in Eq. (3.40) finally yields the harmonic function u(x) which
satisfies all the six boundary conditions of our Dirichlet boundary value problem.
We study examples of this in class and in the assignments.

58

You might also like