Boundary value problems for

partial differential equations

Henrik Schlichtkrull
March 11, 2013

1
Boundary value problem 2

1 Introduction
This note contains a brief introduction to linear partial differential equations.
Partial differential equations differ from ordinary differential equations by being
equations for functions that can depend on more than one variable.
We shall only focus on one particular method of solution, based on the theory
of Fourier series. It is a special case of a more general method called separation
of variables. The idea of the general method is to look for solutions that are
products of functions of one variable, with the hope that all other solutions are
obtained from taking, in general, infinite sums of such product functions. In the
special case that we shall consider, these infinite sums will be Fourier series.
As with ordinary differential equations, the partial differential equations are
usually given with an additional constraint on the solution. In the theory of
ordinary equations, where the solution depends only on one variable, the extra
constraint consists of the initial conditions which impose some given values for
the function (and possibly its derivatives up to some order), at some given initial
time t0 . For partial differential equations, the typical additional constraint is a
so-called boundary condition, in which specified values are imposed at points on
the boundary of the domain where the solution is supposed to be defined. This
explains the title boundary value problems of this note.
There are three main types of partial differential equations of which we shall
see examples of boundary value problems - the wave equation, the heat equation
and the Laplace equation. They are all defined in terms of the Laplacian ∆. By
definition this is the differential operator
2 2
∂ ∂
∆= 2
+ ··· + 2 ,
∂x1 ∂xn
which acts on (two times differentiable) functions of x ∈ Rn .
The wave equation
∂ 2u
= ∆u,
∂t2
and the heat equation
∂u
= ∆u,
∂t
both for functions u(x, t) of x ∈ Rn and t ∈ R, and the Laplace equation
∆u(x) = 0
for functions u(x) of x ∈ Rn . In each case, the differential equation is considered
for elements (x, t) in some domain Ω, in which the solution is required to be
defined and two times differentiable, and it is accompagnied by some conditions
for u on the boundary of Ω. The exact type of boundary conditions can vary
from case to case.
In this note we shall just look at one of these equations, the wave equation,
and we shall only consider the simplest boundary value problem for it.
Boundary value problem 3

2 The wave equation

Let us consider the wave equation in (space) dimension one,

∂ 2u ∂ 2u
= ,
∂t2 ∂x2
for (x, t) in the domain [0, a] × [0, ∞[, where a > 0. A typical boundary condition
consists of
u(0, t) = u(a, t) = 0
for all t ≥ 0, and
∂u
u(x, 0) = f (x), (x, 0) = g(x)
∂t
for all x ∈ [0, a], for some prescribed functions f and g on the interval [0, a].
Clearly, a solution satisfying these boundary conditions can only exist if we have
f (0) = f (a) = 0. Similarly we need g(0) = g(a) = 0.
The physical interpretation is that u(x, t) describes the displacement at lo-
cation x and time t of a freely vibrating string of length a (and with a suitably
normalized tension). The first condition above signifies that the endpoints of the
string are fixed, and the second condition specifies an initial position together
with an initial velocity of the string. For simplicity we assume in the following
that the initial velocity is g = 0. For simplicity we assume in the following also
that a = π. This can easily be achieved by a change of variables, in which x and
t are both scaled by the common factor π/a.
The idea of our approach is to seek a solution in the form of a Fourier series
X
cn (t)einx (1)
n∈Z

with t-dependent coefficients. The series is assumed to constitute the Fourier

series of the function x 7→ u(x, t) at all times t. In order to solve the equation, one
then needs to determine the coefficient functions cn (t) such that the differential
equation and the boundary conditions are satisfied by the sum of the series.
Recall from the Fourier theory that a function defined for x ∈ [0, π] allows
two natural 2π-periodic extensions, given by the even and the odd extension to
[−π, π]. For the even extension, the Fourier series becomes a pure cosine series,
and for the odd extension it becomes a pure sine series. In our case, where the
boundary conditions prescribe that the solution u(x, t) vanishes for x = 0 and
x = π, it is natural to use the odd extension, since the sine functions sin(nx)
already satisfy this condition for all n. For this reason it is advantageous to
replace the Fourier series (1) by a Fourier sine series
∞
X
bn (t) sin(nx) (2)
n=1
Boundary value problem 4

again with t-dependent coefficients.

Let us postpone the question of convergence of (2) for the moment, and
analyze what other conditions are needed in order that the sum u(x, t) solves
our boundary value problem. Assuming that the convergence of the series is
strong enough, we can perform the partial differentiations with respect to t and
x termwise in the series. Inserting the series into the wave equation
∂ 2u ∂ 2u
=
∂t2 ∂x2
we thus obtain
∞ ∞
X ∂ 2 bn (t) X
sin(nx) = −bn (t)n2 sin(nx)
n=1
∂t2 n=1

for all (x, t) and by uniqueness of the coefficients in such a series, we conclude
that
∂ 2 bn
= −n2 bn
∂t2
for all n. This is an ordinary differential equation for the function bn , which we
can solve. We find, for each n,

bn (t) = Cn cos(nt) + Dn sin(nt) (3)

with arbitrary constants Cn , Dn . These constants will now be determined from

the boundary conditions.
The vanishing at x = 0 and x = π has been taken care of by choosing a series
(2) for which the individual terms already vanish at these points. To satisfy the
boundary condition that u(x, 0) = f (x), we need that the series
∞
X
bn (0) sin(nx)
n=1

sums to f (x) for all x. From the Fourier theory we recall the following.
2.1 Theorem. Let f ∈ C 1 ([0, π]) with f (0) = f (π) = 0 and let B1 , B2 , . . . be
given by
2 π
Z
Bn = f (x) sin(nx) dx. (4)
π 0
Then ∞
P
n=1 |Bn | < ∞ and the Fourier sine series

∞
X
Bn sin(nx), (x ∈ [0, π]),
n=1

converges uniformly to f (x).

Boundary value problem 5

Proof. Let F : R → C be the 2π-periodic extension of of the odd extension

fO : [−π, π] → C of f . Thus
(
f (x) if x ∈ [0, π]
F (x + n2π) =
−f (−x) if x ∈ [−π, 0]

for n ∈ Z. It is easily seen that F is continuous and a piecewise C 1 -function. 1

Now Theorem 5.1 from the notes about Fourier series can be applied to F .
Returning to the analysis of when the series (2) is a solution to our boundary
value problem, we see that in order to satisfy u(x, 0) = f (x) we must choose the
coefficients bn (t) such that bn (0) = Bn , where Bn is defined by (4). For each n
this fixes the value of the first constant in (3) to Cn = Bn .
For the other boundary condition, that ∂u ∂t
(x, 0) = g(x) = 0 for all x, we
consider the termwise differentiated series
∞
X
b0n (t) sin(nx),
n=1

Although we have not yet argued for the validity of termwise differentiation, it
appears that we need this series to represent the zero function g(x) = 0. This
will be the case if b0n (0) = 0 for all n, and this is achieved if Dn = 0 in (3) for
all n.
The conclusion of this analysis is thus that we expect that the Fourier sine
series ∞
X
u(x, t) = Bn sin(nx) cos(nt)
n=1

will produce a solution to the boundary value problem if we define the coeffi-
cients by (4). However, we have not yet considered the delicate questions of
convergence of the series and the validity of termwise differentiation in it. Some
extra regularity of f is needed, and the precise result is stated in the following
theorem.

2.2 Theorem. Assume f ∈ C 3 ([0, π]), and that

f (0) = f (π) = 0, f 00 (0) = f 00 (π) = 0. (5)

Let B1 , B2 , . . . be given by
Z π
2
Bn = f (x) sin(nx) dx. (6)
π 0
1
In fact, F is C 1 : This is clear away from the multiples of π. At x = 0 the derivative from
the left of −f (−x) is equal to the derivative from the right of f (x). Futhermore, the derivative
at π from the left of f (x) is equal to the derivative at −π from the right of −f (−x).
Boundary value problem 6

Then the sum ∞

X
u(x, t) = Bn sin(nx) cos(nt) (7)
n=1

converges to a C 2 -function u(x, t) on [0, π]×[0, ∞[, which solves the wave equation
∂ 2u ∂ 2u
= , (8)
∂t2 ∂x2
with the boundary conditions
u(0, t) = u(π, t) = 0 (9)
and
∂u
u(x, 0) = f (x), (x, 0) = 0. (10)
∂t
Conversely, if a C 2 -function of (x, t) solves this boundary value problem, then it
is the unique function given by the sum (7).
Proof. Let F be the odd 2π-periodic extension of f , as defined in the proof of
Theorem 2.1. It was seen in the footnote to that proof that F is C 1 . The
present assumption on the second derivatives of f at 0 and π implies that F 00 is
continuous at these points, hence F is C 2 . It is in fact C 3 , since continuity of the
third derivative can be seen as in thePfootnote.
It follows from Theorem 2.1 that |Bn | < ∞ and that the Fourier Psine series
of f converges uniformly to f . It follows from the convergence of |Bn | that
the series (7) is convergent for all (x, t). It is then clear that (9) and the first
boundary condition in (10) hold for the sum u(x, t) of (7).
It remains to be verified that u(x, t) is two times continuously differentiable
and solves both the wave equation (8) and the second boundary condition in (10).
Here we need to be able to differentiate u(x, t) twice. We shall do this by termwise
differentiation of the series. From the theorem about termwise differentiation of
a series it is known that this is allowed, provided that both the series itself
and the series of derivatives converge uniformly in a neighborhood of the given
point. Note that each termwise differentiation of the series (7) (with respect to
either t or x) worsens its convergence by multiplying P 2a factor n to the n-th term.
3
However, in our case, since F ∈ C (R), we have n |Bn | < ∞ (apply Theorem
4.3 of the Fourier notes to the second derivative). This implies that after up to
two termwise differentiations, the series (7) is still uniformly convergent. Hence
u(x, t) is C 2 and it suffices to verify the remaining conditions for each individual
term sin(nx) cos(nt). This is a straightforward verification.
To establish the final statement of uniqueness, we assume that u(x, t) is some
C 2 -function which solves the boundary value problem. We have to show that it
satisfies (7). We now define the functions bn (t) by
2 π
Z
bn (t) = u(x, t) sin(nx) dx, (11)
π 0
Boundary value problem 7

then we know from Theorem 2.1 that

∞
X
u(x, t) = bn (t) sin(nx)
n=1

for all (x, t). We have to show

bn (t) = Bn cos(nt)

where Bn are the coefficents in (6). All we have to show is that bn (t) solves the
ordinary differential equation
∂ 2 bn
= −n2 bn (12)
∂t2
with the initial conditions

bn (0) = Bn , b0n (0) = 0. (13)

The first condition in (13) follows immediately from the condition on u that
u(x, 0) = f (x). We are going to determine the derivative b0n of bn from (11). Here
we use the theorem from analysis about differentiation under the integral.
Rb Ac-
cording to this theorem, the derivative with respect to t of an integral a ϕ(x, t) dx
of a function of two variables (x, t) can be determined as

d b
Z Z b
∂ϕ
ϕ(x, t) dx = (x, t) dx
dt a a ∂t

provided ϕ is C 1 . We apply this to (11) and obtain

2 π ∂u
Z
0
bn (t) = (x, t) sin(nx) dx.
π 0 ∂t
In particular, Z π
2 ∂u
b0n (0) = (x, 0) sin(nx) dx = 0
π 0 ∂t
since u(x, t) is assumed to satisfy the boundary condition. This shows the second
condition in (13). Furthermore, again applying differentiation under the integral,

2 π ∂ 2u 2 π ∂ 2u
Z Z
00
bn (t) = (x, t) sin(nx) dx = (x, t) sin(nx) dx
π 0 ∂t2 π 0 ∂x2
since u(x, t) is assumed to satisfy the wave equation. By two consequtive partial
integrations, the latter integral can be rewritten as
2 π d2
Z
00
bn (t) = u(x, t) 2 sin(nx) dx, (14)
π 0 dx
Boundary value problem 8

where in the first partial integration we use that sin(nx) vanishes in the end-
points 0 and π, and in the second we use that u(x, t) vanishes in these points.
Finally, from (14) we obtain
Z π
00 22
bn (t) = −n u(x, t) sin(nx) dx = −n2 bn
π 0

as claimed in (12).

2.3 Remark. The result in Theorem 2.2 can be improved slightly. In fact,
it suffices to assume f is two times differentiable and satisfies (5). However,
the argument for the improvement does not apply in general to other partial
differential equations, and for this reason the proof given above for the weaker
result is preferable.
The improvement is based on the observation that due to the trigonometric
formula
sin(θ) cos(ϕ) = 21 (sin(θ + ϕ) + sin(θ − ϕ)),
the series (7) can be rewritten as a sum of two absolutely convergent Fourier sine
series ∞ ∞
X X
1 1
u(x, t) = 2 Bn sin(n(x + t)) + 2 Bn sin(n(x − t))
n=1 n=1
1 1
= 2
F (x + t) + 2
F (x − t).
That the last expression is a solution to the wave equation when F is two times
differentiable, is an immediate consequence of the chain rule. It is also easy to
verify the boundary conditions (the expression above is known as d’Alembert’s
solution).

The solution found in Theorem 2.2,

∞
X
u(x, t) = Bn sin(nx) cos(nt),
n=1

has the following physical interpretation. It represents a superposition of oscil-

n 1
lations with frequencies νn = 2π . The lowest frequency, ν1 = 2π , is called the
fundamental frequency. The higher frequencies ν2 , ν3 , . . . are called overtones.
Each overtone is exactly one octave higher than the preceding. The correspond-
ing displacements sin(nx) cos(nt) of the string are called pure oscillations. The
sound obtained from the string will thus be a superposition of pure oscillations
with frequencies equal to the overtones, and the coefficients in this superposition
are determined from the Fourier series of the initial displacement function f (x).
Boundary value problem 9

3 Separation of variables
As described in the introduction the technique used in the previous chapter is an
instance of a more general technique called separation of variables, which we will
briefly describe.
The starting point for the investigation above was the Ansatz that a solution
u(x, t) could be obtained in the form of a Fourier series
X
cn (t)einx
n∈Z

or a Fourier sine series X

bn (t) sin(nx).
n∈Z

The method we are about to describe explains why we made this Ansatz, and it
can be used to find reasonable analogues in some more general situations.
In the more general approach one first seeks a solution u(x, t) of arbitrary
product form. That is, we assume

u(x, t) = X(x)T (t)

for some one-variable functions X and T to be determined. With this Ansatz,

and still aiming for solutions of the wave equation

∂ 2u ∂ 2u
= ,
∂t2 ∂x2
we obtain
X(x)T 00 (t) = X 00 (x)T (t)
for all x and t. Dividing by X(x)T (t) (if allowed) we obtain

T 00 (t) X 00 (x)
=
T (t) X(x)

in which the variables have been separated. Since t and x are independent vari-
ables, the two sides can agree for all t and all x only if both sides are independent
of both variables. In other words, T 00 (t)/T (t) and X 00 (x)/X(x) both have to be
equal to a constant κ. We conclude that X and T have to satisfy the following
ordinary differential equations

X 00 − κX = 0, T 00 − κT = 0. (15)

We keep in mind that we also want the boundary conditions

u(0, t) = u(a, t) = 0
Boundary value problem 10

for all t ≥ 0, and

∂u
u(x, 0) = f (x), (x, 0) = 0,
∂t
to be satisfied.
Let us concentrate on the function X(x). If T is nontrivial the first boundary
condition applied to u(x, t) = X(x)T (t) implies

X(0) = X(a) = 0.

Suppose first that κ > 0. Then the differential equation X 00 − κX implies

that X is a linear combination
√ √
X(x) = c1 e κx
+ c2 e− κx

and the boundary condition implies

√ √
c1 + c2 = 0, c1 e κa
+ c2 e − κa
= 0.
√ √
Since e κa 6= e− κa , the only solution is c1 = c2 = 0 and hence X = 0. If κ = 0
we obtain similarly that X is a linear combination

X(x) = c1 + c2 x

which only satisfies X(0) = X(a) = 0 if X = 0. We conclude that the constant

κ has to be negative. √
We assume κ < 0 and let ω = −κ. Then the solution to X 00 − κX reads

X(x) = c1 eiωx + c2 e−iωx

and the boundary condition now becomes

c1 + c2 = 0, c1 eiωa + c2 e−iωa = 0.

Here nontrivial solutions are obtained precisely when eiωa = e−iωa , or equivalently,
when ωa = nπ for some n ∈ N. In that case we obtain with c1 = −c2 , that X is
(a constant multiple of) the function
nπx
X(x) = sin(ωx) = sin( ).
a
When a = π as above, the analysis thus leads to the conclusion that it is reason-
able to expect the product form of a solution only if X(x) is a multiple of sin(nx)
for some n ∈ N.
Since the differential equation is linear, a more general solution is obtained
by taking linear combinations of products X(x)T (t) where X(x) = sin(nx) with
Boundary value problem 11

different values of n ∈ Z. The factor T (t) is allowed to vary with n, and writing
it as T (t) = bn (t) this leads exactly to the Ansatz
∞
X
u(x, t) = bn (t) sin(nx)
n=1

of the previous section. Notice that with κ = −ω 2 = −n2 the equation for T (t)
in (15) becomes
T 00 = −n2 T,
which is exactly the equation (12) which we solved in order to find bn .

4 The wave equation, higher dimension

The vibrations of a circular drum are solutions of the wave equation in two space
dimensions
∂ 2u ∂ 2u ∂ 2u
= + .
∂t2 ∂x2 ∂y 2
Here (x, y) describes a position on the membrane of the drum. We denote by
a > 0 the radius of the membrane, then positions belong to

{(x, y) ∈ R2 | x2 + y 2 ≤ a2 }.

The variable t represents time, and the function u(x, y, t) describes the displace-
ment of the membrane in location (x, y) at time t. Assuming that the circular
frame of the drum head is fixed we have the natural boundary condition

u(t, x, y) = 0, for x2 + y 2 = a2 .

In this example we shall discuss some solutions to this problem. Since the
drum head is circular it is natural to transform (x, y) into polar coordinates

(x, y) = (r cos θ, r sin θ),

where 0 ≤ r ≤ a and θ ∈ R. In these coordinates the differential equation reads

∂ 2u ∂ 2 u 1 ∂u 1 ∂ 2u
= + +
∂t2 ∂r2 r ∂r r2 ∂θ2
and the boundary condition reads

u(t, a, θ) = 0

for all t and θ.

Boundary value problem 12

The simplest vibrations are those for which the displacement u(t, r, θ) is in-
dependent of θ (called radial vibrations), and we shall confine ourselves to the
consideration of these. The problem we want to solve is thus
∂ 2u ∂ 2 u 1 ∂u
= +
∂t2 ∂r2 r ∂r
with
u(t, a) = 0
for all t. Additional boundary conditions
∂u
u(0, r) = f (r), (0, r) = g(r)
∂t
describe the initial displacement and velocity at time t = 0 by some given func-
tions f and g. We assume f = 0 in the following.
We follow the general strategy outlined in the preceding section and look for
solutions in separated variables, u(r, t) = R(r)T (t). Substituting in the equation
above and dividing both sides by R(r)T (t) yields
T 00 (t) R00 (r) + 1r R0 (r)
=
T (t) R(r)
for all t and all r. As before we conclude that both sides of this equation must
be independent of both t and r, say equal to κ. We obtain the equations
T 00 − κT = 0
and
1
R00 + R0 − κR = 0.
r
As before we separate in cases depending on the sign of κ, and motivated
√ by the
previous case we concentrate immediately on negative κ and put ω = −κ. We
conclude that R(r) must satisfy
1
R00 + R0 + ω 2 R = 0
r
and R(a) = 0. Changing variables to s = ωr we obtain for the function S(s) =
R(s/ω):
1
S 00 + S + S = 0
s
which we recognize as the Bessel equation of order p = 0. The equation was
treated in the notes about power series solutions, where the solution J0 (s) was
found. There is a second independent solution (usually denoted Y0 (s)), but it
behaves badly for s → 0 and can be excluded for this reason. We conclude that
R(r) = J0 (ωr)
Boundary value problem 13

and from the boundary condition R(a) = 0 we deduce that only the values of
ω > 0 for which J0 (ωa) = 0 are relevant. Let 0 < γ1 < γ2 < . . . be the positive
roots of J0 (γ) = 0, then ω = γn /a for some n.
Finally, we solve the equation T 00 = κT = −ω 2 T , subject to the boundary
condition (from the assumption f = 0) that T (0) = 0. We obtain

T (t) = sin(ωt) = sin(γn t/a)

and conclude that the separated solutions are

u(t, r) = J0 (γn r/a) sin(γn t/a).

A reasonable Ansatz for solving the problem is thus that

∞
X
u(t, r) = cn J0 (γn r/a) sin(γn t/a)
n=1

for some constants c1 , c2 , · · · ∈ R.

The next step would be to determine the coefficients cn in terms of the given
function g, and prove that the infinite sum converges to a solution. The frequen-
cies ωn = γn /a, which are determined from the zeros of the Bessel function and
the radius of the drum, are the frequencies that one will hear when the drum is
vibrating (provided it vibrates radially). Since these are not integral multiples
of a fixed frequency, the series of sine functions above is not a series that can be
treated within ordinary Fourier theory. One needs a more general theory, which
is beyond the scope of this course.