Pde

Second Order Linear Partial Differential Equations
Part I
Second linear partial differential equations; Separation of Variables; 2-

point boundary value problems; Eigenvalues and Eigenfunctions
Introduction
We are about to study a simple type of partial differential equations (PDEs):

the second order linear PDEs. Recall that a partial differential equation is
any differential equation that contains two or more independent variables.
Therefore the derivative(s) in the equation are partial derivatives. We will
examine the simplest case of equations with 2 independent variables. A few
examples of second order linear PDEs in 2 variables are:
α2 uxx = ut (one-dimensional heat conduction equation)
a2 uxx = utt (one-dimensional wave equation)
uxx + uyy = 0 (two-dimensional Laplace/potential equation)
In this class we will develop a method known as the method of Separation of

Variables to solve the above types of equations.
© 2008, 2012 Zachary S Tseng E-1 - 1

(Optional topic) Classification of Second Order Linear PDEs
Consider the generic form of a second order linear partial differential

equation in 2 variables with constant coefficients:
a uxx + b uxy + c uyy + d ux + e uy + f u = g(x,y).
For the equation to be of second order, a, b, and c cannot all be zero. Define
its discriminant to be b2 – 4ac. The properties and behavior of its solution
are largely dependent of its type, as classified below.
If b2 – 4ac > 0, then the equation is called hyperbolic. The wave

equation is one such example.
If b2 – 4ac = 0, then the equation is called parabolic. The heat

conduction equation is one such example.
If b2 – 4ac < 0, then the equation is called elliptic. The Laplace

equation is one such example.
In general, elliptic equations describe processes in equilibrium. While the

hyperbolic and parabolic equations model processes which evolve over time.
Example: Consider the one-dimensional damped wave equation

9uxx = utt + 6ut.
It can be rewritten as: 9uxx − utt − 6ut = 0. It has coefficients a = 9, b = 0,

and c = −1. Its discriminant is 9 > 0. Therefore, the equation is hyperbolic.
© 2008, 2012 Zachary S Tseng E-1 - 2

The One-Dimensional Heat Conduction Equation
Consider a thin bar of length L, of uniform cross-section and constructed of

homogeneous material. Suppose that the side of the bar is perfectly
insulated so no heat transfer could occur through it (heat could possibly still
move into or out of the bar through the two ends of the bar). Thus, the
movement of heat inside the bar could occur only in the x-direction. Then,
the amount of heat content at any place inside the bar, 0 < x < L, and at any
time t > 0, is given by the temperature distribution function u(x, t). It
satisfies the homogeneous one-dimensional heat conduction equation:
α2 uxx = ut
Where the constant coefficient α2 is the thermo diffusivity of the bar, given
by α2 = k / ρs. (k = thermal conductivity, ρ = density, s = specific heat, of the
material of the bar.)
Further, let us assume that both ends of the bar are kept constantly at 0
degree temperature (abstractly, by connecting them both to a heat reservoir
© 2008, 2012 Zachary S Tseng E-1 - 3

of the same temperature; more practically, say they are immersed in iced
water). This assumption imposes explicit restriction on the bar’s ends, in
this case:
u(0, t) = 0, and u(L, t) = 0. t > 0
Those two conditions are called the boundary conditions of this problem.
They literally specify the conditions present at the boundaries between the
bar and the outside. Think them as the “environmental factors” of the given
problem.
In addition, there is an initial condition: the initial temperature distribution

within the bar, u(x, 0). It is a snapshot of the temperature everywhere inside
the bar at t = 0. Therefore, it is an (arbitrary) function of the spatial variable
x only. That is, the initial condition is u(x, 0) = f (x).
Hence, what we have is a problem given by:
(Heat conduction eq.) α2 uxx = ut , 0 < x < L, t > 0,
(Boundary conditions) u(0, t) = 0, and u(L, t) = 0,
(Initial condition) u(x, 0) = f (x).
This is an example of what is known, formally, as an initial-boundary value

problem. Although it is still true that we will find a general solution first,
then apply the initial condition to find the particular solution. A major
difference now is that the general solution is dependent not only on the
equation, but also on the boundary conditions. In other words, the given
partial differential equation will have different general solutions when paired
with different sets of boundary conditions.
© 2008, 2012 Zachary S Tseng E-1 - 4

If the boundary conditions specify u, e.g. u(0, t) = f (t) and u(L, t) = g(t), then
they are often called Dirichlet conditions. If they specify the (spatial)
derivative, e.g. ux(0, t) = f (t) and ux(L, t) = g(t), then they are often called
Neumann conditions. If the boundary conditions are linear combinations of
u and its derivative, e.g. α u(0, t) + β ux(0, t) = f (t), then they are called Robin
conditions. Those are the 3 most common classes of boundary conditions.
If the specified functions in a set of condition are all equal to zero, then they
are homogeneous. Our current example, therefore, is a homogeneous
Dirichlet type problem.
But before any of those boundary and initial conditions could be applied, we
will first need to process the given partial differential equation. What can
we do with it? There are other tools (by Laplace transforms, for example),
but the most accessible method to us is called the method of Separation of
Variables. The idea is to somehow de-couple the independent variables,
therefore rewrite the single partial differential equation into 2 ordinary
differential equations of one independent variable each (which we already
know how to solve). We will solve the 2 equations individually, and then
combine their results to find the general solution of the given partial
differential equation. For a reason that should become clear very shortly, the
method of Separation of Variables is sometimes called the method of
Eigenfunction Expansion.
© 2008, 2012 Zachary S Tseng E-1 - 5

Separation of Variables
Start with the one-dimensional heat conduction equation α2 uxx = ut .

Suppose that its solution u(x, t) is such a function that it can be expressed as
a product, u(x, t) = X(x)T(t), where X is a function of x alone and T is a
function of t alone. Then, its partial derivatives can also be expressed
simply by:
u=XT uxx = X ″ T
ux = X ′ T utt = X T ″
ut = X T ′ uxt = utx = X ′ T ′
Hence, the heat conduction equation α2 uxx = ut can be rewritten as
α2 X ″ T = X T ′.
2
Dividing both sides by α X T :
X ′′ T ′
=
X α2T
2
(α is a constant, so it could go to either side of the equation, but it is usually,
and more conveniently, moved to the t side.) The equation is now
“separated”, as all the x-terms are on the left and t-terms are on the right.
Note: The above step would not have been possible if either X = 0 or T = 0.
However, if either part is zero, then u = XT = 0, which will trivially satisfy
the given equation α2 uxx = ut . This constant zero solution is called the
trivial solution of the equation. We know this is going to be the case.
Therefore, we will assume from this point onward that X ≠ 0 and T ≠ 0. We
look for only the nonzero solutions.
© 2008, 2012 Zachary S Tseng E-1 - 6

But how do we completely pull it apart into 2 equations? The critical idea
here is that, because the independent variables x and t can, and do, vary
independently, in order for the above equation to hold for all values of x and
t, the expressions on both sides of the equation must be equal to the same
constant. Let us call the constant −λ. It is called the constant of separation.
(The negative sign is optional, of course, since λ is an arbitrary number and
it could be either positive or negative or even zero. But putting a negative
sign here right now makes our later calculation a little easier.) Thus,
X ′′ T ′
= =−λ .
X α2T
Why must the two sides be the same constant? Well, think what would
happen if one of the sides isn’t a constant. The equation would not be true
for all x and t if that were the case, because then one side/variable could be
held at a fixed value while the other side/variable changes.
Next, equate first the x-term and then the t-term with −λ. We have
X ′′
=−λ → X ″ = −λX → X ″ + λX = 0,
X
and,
T′
=−λ → T ′ = −α2 λ T → T ′ + α2 λ T = 0.
α2T
Consequently, the single partial differential equation has now been separated
into a simultaneous system of 2 ordinary differential equations. They are a
second order homogeneous linear equation in terms of x, and a first order
linear equation (it is also a separable equation) in terms of t. Both of them
can be solved easily using what we have already learned in this class.
© 2008, 2012 Zachary S Tseng E-1 - 7

Lastly, now that the partial differential equation becomes two ordinary
differential equations, we need to similarly rewrite the boundary conditions.
The boundary conditions can be rewritten as:
u(0, t) = 0 → X(0)T(t) = 0 → X(0) = 0 or T(t) = 0

u(L, t) = 0 → X(L)T(t) = 0 → X(L) = 0 or T(t) = 0
If we choose T(t) = 0, both conditions would be satisfied. However, it

would mean that the temperature distribution function, u(x, t) =
X(x)T(t), would be the constant zero function (the trivial solution).
That is a totally uninteresting solution that would not give us the
general solution (it could not satisfy any initial condition, except when
it is also constant zero). Hence, we have to let the new boundary
conditions to be: X(0) = 0 and X(L) = 0.
Therefore, at the end of this process, we have two ordinary differential

equations, together with a set of two boundary conditions that go with the
equation of the spatial variable x:
X ″ + λX = 0, X(0) = 0 and X(L) = 0,
T ′ + α2 λ T = 0 .
The general solution (that satisfies the boundary conditions) shall be solved
from this system of simultaneous differential equations. Then the initial
condition u(x, 0) = f (x) could be applied to find the particular solution.
© 2008, 2012 Zachary S Tseng E-1 - 8

3 3
Example: Separate t uxx + x utt = 0 into an equation of x and an equation
of t.
Let u(x, t) = X(x)T(t) and rewrite the equation in terms of X and T:
t 3 X ″ T + x3 X T ″ = 0 ,
t3 X ″ T = − x3 X T ″.
Divide both sides by X ″ T ″, we have separated the variables:
t 3 T − x3 X
=
T ′′ X ′′ .
Now insert a constant of separation:
t 3 T − x3 X
= = −λ .
T ′
′ X ′′
Finally, rewrite it into 2 equations:
t3 T = −λ T ″ → λ T ″ + t3 T = 0,
− x3 X = −λ X ″ → λ X ″ − x3 X = 0.
© 2008, 2012 Zachary S Tseng E-1 - 9

Example: Separate
ux + 2 utx − 10 utt = 0, u(0, t) = 0, ux(L, t) = 0.
Let u(x, t) = X(x)T(t) and rewrite the equation in terms of X and T:
X ′ T + 2 X ′ T ′ − 10 X T ″ = 0,
X ′ T + 2 X ′ T ′ = 10 X T ″.
Divide both sides by X ′ T ″, and insert a constant of separation:

T + 2T ′ 10 X
= =−λ .
T ′′ X′
Rewrite it into 2 equations:
T + 2 T ′ = −λ T ″ → λ T ″ + 2 T ′ + T = 0,
10 X = −λ X ′ → λ X ′ + 10 X = 0.
The boundary conditions also must be separated:
u(0, t) = 0 → X(0)T(t) = 0 → X(0) = 0 or T(t) = 0

ux(L, t) = 0 → X ′(L)T(t) = 0 → X ′(L) = 0 or T(t) = 0
As before, setting T(t) = 0 would result in the constant zero solution

only. Therefore, we must choose the two (nontrivial) conditions in
terms of x: X(0) = 0, and X ′(L) = 0.
© 2008, 2012 Zachary S Tseng E-1 - 10

The Two-Point Boundary Value Problems
What we have done thus far is to separate the heat conduction equation, with
2 independent variables, into 2 equations of one variable each. Meanwhile
we have also rewritten the boundary conditions, so that they now associate
with the spatial variable x only.
(1) X ″ + λX = 0, X(0) = 0 and X(L) = 0,

(2) T ′ + α2 λ T = 0.
The next task is to solve this system of two simultaneous ordinary

differential equations, one of them with boundary conditions. We will look
the 2 equations one at a time, and consolidate their solutions at the end. We
will start off by solving the more interesting (and more complex) of the two,
namely the second order linear equation where x is the independent variable.
It is our first taste of a boundary value problem (BVP). (It is not an initial
value problem; as mentioned earlier in this course, an initial value problem
requires that both of its data points be taken at the same time/place, but here
we have 2 data points taken at different instances of x: at x = 0 and x = L.)
A little background first: unlike an initial value problem of a second order

linear equation, whose solution’s existence and uniqueness (under certain
well-understood conditions) are guaranteed, there is no such guarantee for a
boundary value problem. Indeed, take an arbitrary pairing of a differential
equation and a set of boundary conditions, the odds are good that there is not
a solution satisfying them, or that there are multiple solutions satisfying
them. Notice that the boundary value problem in (1) arises from a
homogeneous linear differential equation, which always has at least one
solution (the constant zero solution, or the trivial solution) which would also
satisfy the homogeneous boundary conditions given. Thus, (in this case, at
least) the existence of a solution is not the issue. The constant zero solution,
X(t) = 0, however, is not usable for us. Because, if X(t) = 0, then u(x, t) =
X(x)T(t) = 0, which is not the general solution. (Why not?) Hence, what we
are looking to find presently is a second, nonzero, solution to the given
boundary value problem. That is, we are looking for instance(s) where the
uniqueness of solution fails to hold.
© 2008, 2012 Zachary S Tseng E-1 - 11

Somewhat fortunately for us, that in (1) above while the boundary
conditions are fixed, the equation itself is not − note that the coefficient λ,
which was just the arbitrary constant of separation, has not been determined.
Thus, what we need to do here is somewhat a reverse of what you’d have
expected to be doing. Namely, we will start with the fixed boundary
conditions and try to find an equation (by finding an appropriate coefficient
λ) that has a nonzero solution satisfying the given boundary conditions. (In
other words, we are starting with the answer and then go looking for the
correct question that would give that answer!) This kind of reversed
boundary value problems is called an Eigenvalue problem. The specific
value(s) of λ that would produce a nonzero solution of the boundary value
problem is called an eigenvalue of the boundary value problem. The
nonzero solution that arises from each eigenvalue is called a corresponding
eigenfunction of the boundary value problem.
Note: Why are λ and X(t) called, respectively, eigenvalue and

d2
eigenfunction? Let D = − denotes the second derivative
dx 2
differential operator. The boundary value problem (1) on the previous
page can be rewritten into
DX = λX,
where X is restricted to functions that satisfy the boundary conditions

X(0) = 0 and X(L) = 0. Notice the similarity with the defining relation,
Ax = λx, of eigenvalue and eigenvector in linear algebra, hence the
similar naming scheme. Like eigenvectors, every constant multiple of
an eigenfunction is another eigenfunction corresponding to the same
eigenvalue.
© 2008, 2012 Zachary S Tseng E-1 - 12

Eigenvalues and Eigenfunctions of a Two-Point BVP
Hence, the next goal is to find the eigenvalues λ such that the boundary
value problem (1)
X ″ + λX = 0, X(0) = 0 and X(L) = 0,
will have a nonzero solution satisfying both boundary conditions. Since the
form of the general solution of the second order linear equation is dependent
on the type of roots that its characteristic equation has. In this example, the
characteristic equation is r 2 + λ = 0. The type of roots it has is dependent on
its discriminant, which is simply −4λ. We will attempt to find λ by
separately considering the 3 possible types of the solution arise from the
different roots of the characteristic equation.
Case 1: If λ < 0 (−4λ > 0, distinct real roots of characteristic equation):
Let us denote λ = −σ 2, where σ = − λ > 0 . The characteristic

equation becomes r 2 + λ = r 2 − σ 2 = 0, which has roots r = ± σ.
σx −σx
The general solution is then X(x) = C1 e + C2 e . Applying the
first of the boundary conditions gives
X(0) = 0 = C1 + C2 → C2 = − C1
The second boundary condition gives

σL −σL σL −σL
X(L) = 0 = C1 e + C2 e = C1 ( e −e )
Hence, either C1 = 0 = C2 , (this would be the zero, or the trivial

solution, which does not lead to the general solution later on); or
σL −σL
e −e = 0. But this second case does not have a solution (why?).
Therefore, there is no negative value of λ that would give a nonzero
solution. Thus, there is no negative eigenvalue for this problem.
© 2008, 2012 Zachary S Tseng E-1 - 13

Case 2: If λ = 0 (−4λ = 0, repeated real root of characteristic equation):
The equation becomes X″ = 0. It has the general solution (either by

integrating both sides twice, or from the characteristic equation r 2 = 0)
X(x) = C1 + C2 x.
Applying boundary conditions to get
X(0) = 0 = C1 → C1 = 0
X(L) = 0 = C1 + C2 L = C2 L → C2 = 0 (because L > 0)
Hence, X(x) = 0 + 0 x = 0, the trivial solution is the only possibility.

Therefore, zero is not an eigenvalue for this problem either.
Case 3: If λ > 0 (−4λ < 0, complex roots of characteristic equation):
Let us denote λ = σ 2, where σ = λ > 0 . The characteristic equation

becomes r 2 + λ = r 2 + σ 2 = 0, which has roots r = ± σ i.
The general solution is then X(x) = C1 cos(σx) + C2 sin(σx). Applying

boundary conditions to get
X(0) = 0 = C1 cos(0) + C2 sin(0) = C1 → C1 = 0

X(L) = 0 = C1 cos(σL) + C2 sin(σL) = C2 sin(σL)
The second equation has 2 possible solutions: C2 = 0 (= C1), which

results in the trivial solution again; or, more importantly, it could be
that sin(σL) = 0, which means σL = π, 2π, 3π, …, nπ, … That is, there
are infinitely many values σ = π/L, 2π/L, 3π/L, …, nπ/L, … such that
there exists a nonzero solution of this boundary value problem.
The (positive) values of λ for which the equation will have a solution
satisfying the specified boundary conditions, i.e. the eigenvalues of this BVP,
are
2 n2 π 2
λ =σ = 2 , n = 1, 2, 3, …
L
© 2008, 2012 Zachary S Tseng E-1 - 14

All such values of λ are called the eigenvalues of the given boundary value
problem. Recall that the actual solution of this problem corresponding to
each eigenvalue λ is called an eigenfunction of the problem.
What are the eigenfunctions, then? Retracing our work above, we see that
they occur only when λ are positive (therefore, the general solution in the
form of X(x) = C1 cos(σx) + C2 sin(σx)), that C1 must be zero, and C2 could
2 2 2
be any nonzero constant. Lastly, λ = n π /L , where n = 1, 2, 3, … are all
the positive integers.
Therefore, the eigenfunctions corresponding to the eigenvalues found above

− that is, they are the actual nonzero solutions that satisfy the given set of
boundary conditions when the original differential equation has λ = n2π2/L2
as its coefficient − are
nπ x
X n = sin
L , n = 1, 2, 3, …
© 2008, 2012 Zachary S Tseng E-1 - 15

Example: Find the eigenvalues and eigenfunctions of the two-point
boundary value problem
X ″ + λX = 0, X(0) = 0 and X ′(L) = 0.
Case 1: If λ < 0:
Again denote λ = −σ 2, where σ = − λ > 0 . The characteristic

equation becomes r 2 + λ = r 2 − σ 2 = 0, which has roots r = ± σ.
σx −σx
The general solution is then X(x) = C1 e + C2 e . Its derivative is
σx −σx
X ′(x) = C1 σ e − C2 σ e . Apply the boundary conditions and we
get
X(0) = 0 = C1 + C2 → C2 = − C1
σL −σL σL −σL
X ′(L) = 0 = C1 σ e − C2 σ e = C1 σ ( e +e )
Since σ > 0, once again we encounter the situation where either C1 = 0

σL −σL
= C2 (thus giving us the trivial solution), or e + e = 0 (which is
impossible, because each exponential term is always positive, and as a
result their sum cannot be zero). As before, this second case does not
have a solution, either. Therefore, there is no negative eigenvalue for
this problem.
Case 2: If λ = 0:
Again, the equation becomes X ″ = 0. The general solution is,

therefore, X(x) = C1 + C2 x. Its derivative is X ′(x) = C2.
Applying boundary conditions to get
X(0) = 0 = C1 → C1 = 0
X ′(L) = 0 = C2 → C2 = 0
Hence, X(x) = 0 + 0 x = 0, the trivial solution once again. Therefore,

zero is not an eigenvalue for this problem.
© 2008, 2012 Zachary S Tseng E-1 - 16

Case 3: If λ > 0:
As before, denote λ = σ 2, where σ = λ > 0 . The characteristic

equation is then r 2 + λ = r 2 + σ 2 = 0, which has roots r = ± σ i.
The general solution is X(x) = C1 cos(σx) + C2 sin(σx). Its derivative is

X ′(x) = −C1 σ sin(σx) + C2 σ cos(σx). Applying boundary conditions to
get
X(0) = 0 = C1 cos(0) + C2 sin(0) = C1 → C1 = 0

X ′(L) = 0 = −C1 σ sin(σL) + C2 σ cos(σL) = C2 σ cos(σL)
Since σ > 0 always, the second equation has 2 possible solutions:

C2 = 0 (= C1), which results in the trivial solution; or, it could be that
cos(σL) = 0, which means σL = π/2, 3π/2, 5π/2, …, (2n − 1)π/2, …
That is, there are infinitely many values σ = π/2L, 3π/2L, 5π/2L, …,
(2n − 1)π/2L, … such that there exists a nonzero solution of this
boundary value problem.
Therefore, yes, there are positive eigenvalues in the form
2(2n − 1) 2 π 2
λ =σ = , n = 1, 2, 3, …
4 L2
Retracing the steps in the above calculation, we see that the

eigenvalues have corresponding eigenfunctions
(2n − 1) π x
X n = sin , n = 1, 2, 3, …
2L
© 2008, 2012 Zachary S Tseng E-1 - 17

Let us return to solving the heat conduction problem. After finding the
eigenvalues and corresponding eigenfunctions by solving the 2-point
boundary value problem, the next step in the process is to solve the second
equation (that of the time variable t):
T ′ + α2 λ T = 0. (2)
Since it is the second half of a system of simultaneous equations, the

eigenvalues λ from the first equation have to be used here. Hence, the
equation becomes:
n 2π 2
2
T′+α 2
T =0.
L
The equation is both a first order linear equation, as well as a separable
equation. Solve it using either method (solving it as a separable equation is
probably quicker) to get the general solution
−α 2 n 2π 2 t / L2
Tn (t ) = Cn e , n = 1, 2, 3, …
We can now assemble the result from the two ordinary differential equations
to find the solutions of the partial differential equation. Recall that the
assumption in the separation of variables method is that the PDE has
solutions in the form u(x, t) = X(x)T(t). Therefore, we see that the solutions
of the one-dimensional heat conduction equation, with the boundary
conditions u(0, t) = 0 and u(L, t) = 0, are in the form
−α 2 n 2π 2 t / L2 nπ x
u n ( x, t ) = X n ( x) Tn (t ) = Cn e sin
L ,
n = 1, 2, 3, …
© 2008, 2012 Zachary S Tseng E-1 - 18

The general solution, of the temperature distribution (recall that this general
solution is only valid for the given set of boundary conditions) within a bar
that has both ends kept at 0 degree, is just the linear combination of all the
above (linearly independent) functions un(x, t). That is,
∞
−α 2 n 2π 2 t / L2 nπ x
u ( x, t ) = ∑ Cn e sin
n =1 L .
Once the general solution has been found, we can now apply the initial
condition in order to find the particular solution. Set t = 0 in the general
solution above, and equate it with the initial condition u(x, 0) = f (x):
∞
nπ x
u ( x,0) = ∑ Cn sin = f ( x) .
n =1 L
Take a few seconds to grasp what this equation says, and we quickly realize
that we are not out of the woods yet. The equation above specifies that the
(arbitrary) initial condition must be equal to an infinite series of sine terms,
and there are infinitely many coefficients cn that need to be solved. Can this
equation even be solved, in general? The answer is a reassuring “yes”. But
to get there we still need to know several things. Namely, what kind of
functions could be expressed as a series of sines (and, more generally, sines
and/or cosines)? Even if such functions exist, how does the arbitrary initial
condition f (x) fit in? Lastly, how could we possibly solve for the infinitely
many coefficients with just this one equation given?
© 2008, 2012 Zachary S Tseng E-1 - 19

Summary (thus far)
The Method of Separation of Variables:
1. Separate the PDE into ODEs of one independent variable each.

Rewrite the boundary conditions so they associate with only one of
the variables.
2. One of the ODEs is a part of a two-point boundary value problem.

Solve this problem for its eigenvalues and eigenfunctions.
3. Solve the other ODE.
4. Multiply the results from steps (2) and (3), and sum up all the
products to find the general solution.
© 2008, 2012 Zachary S Tseng E-1 - 20

The “Rosetta Stone” of Separation of Variables.
© 2008, 2012 Zachary S Tseng E-1 - 21

Exercises E-1.1:
1 − 4 Determine whether each PDE can be separated. Then separate it into

two ODEs if it is possible to separate.
1. x2 uxx − t2 utt = 0
2. x uxx − π utt = 5 uxt
3. uxx − 3u = ut , u(0, t) = 0, u(π, t) = 0.
4. uxx + 2t utx = 4 u, ux(0, t) = 0, u(9, t) = 0.
5. Show that the following boundary value problem has no solutions:

y″ + 4y = 0, y(0) = 2, y(π) = 3.
6 − 11 Find all eigenvalues and their corresponding eigenfunctions of each

two-point boundary value problem.
6. X ″ + λX = 0, X(0) = 0 , X(2π) = 0.
7. X ″ + λX = 0, X(0) = 0 , X ′(2π) = 0.
8. X ″ + λX = 0, X ′(0) = 0 , X(2π) = 0.
9. X ″ + λX = 0, X ′(0) = 0 , X ′(2π) = 0.
10. X ″ − λX = 0, X(0) = 0 , X(1) = 0.
11. X ″ − λX = 0, X ′(0) = 0 , X(1) = 0.
12. (a) Show that any positive eigenvalue of the boundary value problem
X ″ + λX = 0, X(0) + X ′(0) = 0, X(L) = 0,
2
must be in the form λ = σ , where σ satisfies the equation σ = tan(σL). (b) Is
0 an eigenvalue of this problem?
13. Show that 0 is not an eigenvalue, and that any positive eigenvalue of the
boundary value problem
X ″ + λX = 0, X(0) − X ′(0) = 0, X(L) + 2X ′(L) = 0,
2 2σ 2 − 1
must be in the form λ = σ , where σ satisfies the equation cot(σL ) = .
3σ
© 2008, 2012 Zachary S Tseng E-1 - 22

Answers E-1.1:
1. Separable; there are many ways to separate, one possible result is

x2 X ″ + λX = 0, and t2 T ″ + λT = 0.
2. Not separable.
3. Separable; one possible result is
X ″ + (λ − 3)X = 0, and T ′ + λT = 0.
The boundary conditions become X (0) = 0 , X(π) = 0.
4. Separable; one possible result is
X ″ + λX ′ − 4X = 0, and 2t T ′ − λT = 0.
The boundary conditions become X ′(0) = 0 , X(9) = 0.
n2 nx
6. λ = , X n = sin , n = 1, 2, 3, …
4 2
( 2n − 1) 2 (2n − 1) x
7. λ = , X n = sin , n = 1, 2, 3, …
16 4
( 2n − 1) 2 (2n − 1) x
8. λ = , X n = cos , n = 1, 2, 3, …
16 4
n2 nx
9. λ = 0, X0 = 1; and λ = , X n = cos , n = 1, 2, 3, …
4 2
10. λ = − n 2π 2 , X n = sin nπ x , n = 1, 2, 3, …
− ( 2n − 1) 2 π 2 (2n − 1)π x
11. λ = , X n = cos , n = 1, 2, 3, …
4 2
12. (b) No, 0 is not an eigenvalue.
© 2008, 2012 Zachary S Tseng E-1 - 23

Part II
Fourier series; Euler-Fourier formulas; Fourier Convergence Theorem;

Even and odd functions; Cosine and Sine Series Extensions; Particular
solution of the heat conduction equation
Fourier Series
Suppose f is a periodic function with a period T = 2L. Then the Fourier

series representation of f is a trigonometric series (that is, it is an infinite
series consists of sine and cosine terms) of the form
a0 ∞  nπ x nπ x 
f ( x ) = + ∑  a n cos + bn sin 
2 n =1  L L 
Where the coefficients are given by the Euler-Fourier formulas:
mπ x
L
1
am = ∫ f ( x) cos dx , m = 0, 1, 2, 3, …
L −L L
nπ x
L
1
bn = ∫ f ( x) sin dx , n = 1, 2, 3, …
L −L L
The coefficients a’s are called the Fourier cosine coefficients (including a0,
the constant term, which is in reality the 0-th cosine term), and b’s are called
the Fourier sine coefficients.
© 2008, 2012 Zachary S Tseng E-2 - 1

Note 1: Thus, every periodic function can be decomposed into a sum of one
or more cosine and/or sine terms of selected frequencies determined solely
by that of the original function. Conversely, by superimposing cosines and/
or sines of a certain selected set of frequencies we can reconstruct any
periodic function.
Note 2: If f is piecewise continuous, then the definite integrals in the Euler-

Fourier formulas always exist (i.e. even in the cases where they are improper
integrals, the integrals will converge). On the other hand, f needs not to be
piecewise continuous to have a Fourier series. It just needs to be periodic.
However, if f is not piecewise continuous, then there is no guarantee that we
could find its Fourier coefficients, because some of the integrals used to
compute them could be improper integrals which are divergent.
Note 3: Even though that the “=” sign is usually used to equate a periodic
function and its Fourier series, we need to be a little careful. The function f
and its Fourier series “representation” are only equal to each other if, and
whenever, f is continuous. Hence, if f is continuous for −∞ < x < ∞, then f is
exactly equal to its Fourier series; but if f is piecewise continuous, then it
disagrees with its Fourier series at every discontinuity. (See the Fourier
Convergence Theorem below for what happens to the Fourier series at a
discontinuity of f .)
Note 4: Recall that a function f is said to be periodic if there exists a positive

number T, such that f (x + T ) = f (x), for all x in its domain. In such a case
the number T is called a period of f. A period is not unique, since if f (x + T )
= f (x), then f (x + 2T ) = f (x) and f (x + 3T ) = f (x) and so on. That is, every
integer-multiple of a period is again another period. The smallest such T is
called the fundamental period of the given function f. A special case is the
constant functions. Every constant function is clearly a periodic function,
with an arbitrary period. It, however, has no fundamental period, because its
period can be an arbitrarily small real number. The Fourier series
representation defined above is unique for each function with a fixed period
T = 2L. However, since a periodic function has infinitely many (non-
fundamental) periods, it can have many different Fourier series by using
different values of L in the definition above. The difference, however, is
really in a technical sense. After simplification they would look the same.
© 2008, 2012 Zachary S Tseng E-2 - 2

Therefore, technically at least, a Fourier series of a periodic function
depends both on the function as well as its chosen period.
Note 5: The definite integrals in the Euler-Fourier formulas can be found be

integrating over any interval of length 2L. However, from −L to L is the
convention, and is often the most convenient interval to use.
Note 6: Since the Fourier coefficients are calculated by definite integrals,

which are insensitive to the value of the function at finitely many points.
Consequently, piecewise continuous functions of the same period that differ
from each other at finitely many points (notably, at isolated discontinuities)
per period will have the same Fourier series.
Note 7: The constant term in the Fourier series, which has expression
L L
a0 1 1 1
= ⋅ ∫ f ( x) cos(0) dx =
2 L −∫L
f ( x) dx ,
2 2 L −L
is just the average or mean value of f (x) on the interval [−L, L]. Since f is
periodic, this average value is the same for every period of f. Therefore, the
constant term in a Fourier series represents the average value of the function
f over its entire domain.
© 2008, 2012 Zachary S Tseng E-2 - 3

Example: Find a Fourier series for f (x) = x, −2 < x < 2, f (x + 4) = f (x).
First note that T = 2L = 4, hence L = 2.

The constant term is one half of:
2 2
mπ x 1 x2
L
1 1 1
a0 = ∫ f ( x) cos dx = ∫−2 x dx = 2 2 = ( 2 − 2) = 0
L −L L 2 −2
2
The rest of the cosine coefficients, for n = 1, 2, 3, …, are
2
nπ x nπx
L
1 1
an = ∫ f ( x) cos dx = ∫ x cos dx
L −L L 2 −2
2
1  2 x nπx 
2 2
nπx 2
=
2  nπ
sin −
2 − 2 nπ ∫−2sin 2 dx 

1  2x 
2
nπx 4 nπx
=  sin + 2 2 cos 
2  nπ 2 nπ 2 −2 

1  4   4 
=   0 + 2 2 cos( nπ )  −  0 + 2 2 cos( − nπ )   = 0
2  nπ   nπ 
Hence, there is no nonzero cosine coefficient for this function. That is,
its Fourier series contains no cosine terms at all. (We shall see the
significance of this fact a little later.)
© 2008, 2012 Zachary S Tseng E-2 - 4

The sine coefficients, for n = 1, 2, 3, …, are
2
nπ x nπ x
L
1 1
bn = ∫ f ( x) sin dx = ∫ x sin dx
L −L L 2 −2
2
1  − 2 x nπ x 
2 2
nπ x −2
2 −2 nπ −∫2
= cos − cos dx
2  nπ 2 

2
1  − 2x nπ x 4 nπ x 
=  cos + 2 2 sin 
2  nπ 2 nπ 2 −2 

1  − 4   4 
=   cos( nπ ) − 0  −  cos( − nπ ) − 0  
2   nπ   nπ 
−2
= (cos(nπ ) + cos(nπ ) ) = − 4 cos( nπ )
nπ nπ
 4
 nπ , n = odd (−1) n +1 4
= =
−4
 , n = even nπ .
 nπ
4 ∞
(−1) n + 1 nπ x
Therefore, f ( x ) = ∑
π n =1 n
sin
2
.
© 2008, 2012 Zachary S Tseng E-2 - 5

Figure: the graph of the partial sum of the first 30 terms of the
Fourier series
4 ∞
(−1) n + 1 nπ x
f ( x) = ∑ sin .
π n =1 n 2
Compare it against the graph of the actual function the series

represents the function f (x) = x, −2 < x < 2, f (x + 4) = f (x), seen
earlier.
© 2008, 2012 Zachary S Tseng E-2 - 6

Example: Find a Fourier series for f (x) = x, 0 < x < 4, f (x + 4) = f (x). How
will it be different from the series above?
4 4
1 1 x2 1
a0 = ∫ x dx = = (8 − 0) = 4
20 2 2 0
2
For n = 1, 2, 3, … :
4
1 nπx
an = ∫ x cos dx
20 2
1  2x 
4
nπx 4 nπx
=  sin + 2 2 cos 
2  nπ 2 nπ 2 0 

1  4   4 
=   0 + 2 2 cos( 2nπ )  −  0 + 2 2 cos(0)   = 0
2  nπ   nπ 
4
1 nπx
bn = ∫ x sin dx
20 2
1  − 2x 
4
nπx 4 nπx
=  cos + 2 2 sin 
2  nπ 2 nπ 2 0 

1  − 8   −4
=   cos( 2nπ ) − 0  − (0 − 0 ) =
2   nπ   nπ
Consequently,
a0 ∞  nπ x nπ x  −4 ∞ 1 nπx
f ( x) = + ∑  an cos + bn sin  = 2+ ∑ sin
2 n =1  L L  π n =1 n 2 .
© 2008, 2012 Zachary S Tseng E-2 - 7

Example: Find a Fourier series for f (x) = | x |, −2 < x < 2, f (x + 4) = f (x).
8 ∞
1 (2n − 1)π x
Answer : f ( x) = 1 −
π 2 ∑ (2n − 1)
n =1
2
cos
2
Example: Find a Fourier series for

 − 2, −1 ≤ x < 0
f ( x) =  , f (x + 2) = f (x).
 2, 0 ≤ x <1
∞
8 1
Answer : f ( x) =
π
∑ (2n − 1) sin((2n − 1)π x)
n =1
© 2008, 2012 Zachary S Tseng E-2 - 8

Comment: Just because a Fourier series could have infinitely many (nonzero)
terms does not mean that it will always have that many terms. If a periodic
function f can be expressed by finitely many terms normally found in a
Fourier series, then the expression must be the Fourier series of f. (This is
analogous to the fact that the Maclaurin series of any polynomial function is
just the polynomial itself, which is a sum of finitely many powers of x.)
Example: The Fourier series (period 2π) representing f (x) = 5 + cos(4x) −

sin(5x) is just f (x) = 5 + cos(4x) − sin(5x).
Example: The Fourier series (period 2π) representing f (x) = 6 cos(x) sin(x) is
not exactly itself as given, since the product cos(x) sin(x) is not a term in a
Fourier series representation. However, we can use the double-angle
formula of sine to obtain the result: 6 cos(x) sin(x) = 3 sin(2x).
Consequently, the Fourier series is f (x) = 3 sin(2x).
© 2008, 2012 Zachary S Tseng E-2 - 9

The Fourier Convergence Theorem
Here is a theorem that states a sufficient condition for the convergence of a

given Fourier series. It also tells us to what value does the Fourier series
converge to at each point on the real line.
Theorem: Suppose f and f ′ are piecewise continuous on the interval

−L ≤ x ≤ L. Further, suppose that f is defined elsewhere so that it is periodic
with period 2L. Then f has a Fourier series as stated previously whose
coefficients are given by the Euler-Fourier formulas. The Fourier series
converge to f (x) at all points where f is continuous, and to
 lim f ( x) + lim f ( x) / 2
 x → c − x → c+ 
at every point c where f is discontinuous.
Comment: As seen before, the fact that f is piecewise continuous guarantees

that the Fourier coefficients can be found. The condition that f ′ is also
piecewise continuous is a sufficient condition to guarantee that the series
thusly found will be convergent everywhere on the real line. As well, recall
that, suppose f is continuous at c, then by definition f (c) equals both one-
sided limits of f (x) as x approaches c. Therefore, the second part of the
theorem could be even more succinctly stated as that the Fourier series
representing f will always converge to
 lim f ( x) + lim f ( x) / 2
 x → c − x → c+ 
at every point c (and not just at discontinuities of f ).
A consequence of this theorem is that the Fourier series of f will “fill in” any
removable discontinuity the original function might have. A Fourier series
will not have any removable-type discontinuity.
© 2008, 2012 Zachary S Tseng E-2 - 10

Example: Let us revisit the earlier calculation of the Fourier series
representing f (x) = x, −2 < x < 2, f (x + 4) = f (x).
The Fourier series, as we have found, is
4 ∞
(−1) n + 1 nπ x
f ( x) = ∑ sin .
π n =1 n 2
The following figures are the graphs of various finite n-th partial sums of the
series above.
n=3
n = 10
© 2008, 2012 Zachary S Tseng E-2 - 11

n = 20
n = 30
n = 50
© 2008, 2012 Zachary S Tseng E-2 - 12

Note that superimposed sinusoidal curves take on the general shape of the
piecewise continuous periodic function f (x) almost immediately. As well,
for the parts of the curve where f (x) is continuous (where the Fourier
Convergence Theorem predicts a perfect match) the composite curve of the
Fourier series converges rapidly to that of f (x), as predicted. The
convergence is not as rapidly near the jump discontinuities. Indeed, for all
but the lowest partial sums of the Fourier series, the curve seems to
“overshoot” that of f (x) near each jump discontinuity by a noticeable margin.
Further more, this discrepancy does not fade away for any finitely larger n.
That is, the convergence of a Fourier series, while predictable, is not uniform.
(That is a small price we pay for approximating a piecewise continuous
periodic function by sinusoidal curves. It can be done, but the Fourier series
does not converge uniformly to the actual function.)
This behavior is known as the Gibbs Phenomenon. It further states that the
partial sums of a Fourier series will overshoot a jump discontinuity by an
amount approximately equal to 9% of the jump. That is, near each jump
discontinuity, the overshoot amounts to about
0.09 lim+ f ( x) − lim− f ( x) ,

x→c x→c
for large n. Further, this overshoot does not go away for any finitely large n.
© 2008, 2012 Zachary S Tseng E-2 - 13

Question: Sketch the graph of the Fourier series of
f (x) = x, −2 < x < 2, f (x + 4) = f (x).
We have seen a few graphs of its partial sums. But what will the graph of
the actual Fourier series look like?
Example: Sketch the graph of the Fourier series of

f (x) = | x |, −2 < x < 2, f (x + 4) = f (x).
Example: Sketch the graph of the Fourier series of

 − 2, −1 ≤ x < 0
f ( x) =  , f (x + 2) = f (x).
 2, 0 ≤ x <1
© 2008, 2012 Zachary S Tseng E-2 - 14

Even and Odd Functions
Recall that an even function is any function f such that
f (−x) = f (x), for all x in its domain.
Examples: cos(x), sec(x), any constant function, x2, x4, x6, … , x −2, x −4, …
An odd function is any function f such that
f (−x) = −f (x), for all x in its domain.
Examples: sin(x), tan(x), csc(x), cot(x), x, x3, x5, … , x −1, x −3, …
Most functions, however, are neither even nor odd. There is one function
that is both even and odd. (What is it?)
Arithmetic Combinations of Even and Odd Functions
The table below summaries the result of performing the common arithmetic
operations on a pair of even and/or odd functions:
Even and Even Odd and Odd Even and Odd

+ / − Even Odd Neither
× / ÷ Even Even Odd
The result above can be extended to arbitrarily many terms. For example, a
sum of three or more even functions will again be even. (Care needs to be
taken in the cases where 3 or more odd functions forming a product/quotient.
For example, a product of 3 odd functions will be odd, but a product of 4
odd functions is even.)
© 2008, 2012 Zachary S Tseng E-2 - 15

Calculus Properties of Even and Odd Functions
Suppose f is an even function, continuous on −L ≤ x ≤ L, then
L L
∫
−L
f ( x ) dx = 2 ∫ f ( x) dx .
0
Suppose f is an odd function, continuous on −L ≤ x ≤ L, then
∫
−L
f ( x ) dx = 0 .
© 2008, 2012 Zachary S Tseng E-2 - 16

The Fourier Cosine Series
Suppose f is an even periodic function of period 2L, then its Fourier series
contains only cosine (include, possibly, the constant term) terms. It will not
have any sine term. That is, its Fourier series is of the form
a0 ∞ nπ x
f ( x) = + ∑ an cos
2 n =1 L .
Conversely, any periodic function whose Fourier series has the form of a
cosine series as shown must be an even periodic function. Computationally,
this means that the Fourier coefficients of an even periodic function are
given by
mπ x mπ x
L L
1 2
am = ∫ f ( x) cos dx = ∫ f ( x) cos dx ,
L −L L L0 L
m = 0, 1, 2, 3, …
bn = 0, n = 1, 2, 3, …
Notice that the integrand in the definite integral used to find the cosine
coefficients a’s is an even function (it is a product of two even functions, f (x)
and cos x). Therefore, we can use the symmetric property of even functions
to simplify the integral.
© 2008, 2012 Zachary S Tseng E-2 - 17

The Fourier Sine Series
If f is an odd periodic function of period 2L, then its Fourier series contains
only sine terms. It will not have any cosine term. That is, its Fourier series
is of the form
∞
nπ x
f ( x) = ∑ bn sin
n =1 L
Conversely, any periodic function whose Fourier series has the form of a
sine series as shown must be an odd periodic function. Therefore, the
Fourier coefficients of an odd periodic function are given by
am = 0, m = 0, 1, 2, 3, …
nπ x
L
2
bn =
L ∫0
f ( x) sin
L
dx , n = 1, 2, 3, …
Example: We have calculated earlier that the function f (x) = x, −2 < x < 2,
f (x + 4) = f (x), has as its Fourier series consists of purely sine terms:
4 ∞ (−1) n + 1 nπ x
f ( x) = ∑ sin .
π n =1 n 2
We now see that this sine series signifies that the function is odd periodic.
It is perhaps not very obvious, but the integrand in the integral for the
Fourier sine coefficients is another even function. It is a product of two odd
functions, f (x) and sin x, which makes it even. Therefore, we can again take
advantage of the symmetric property of even functions to simplify the
integral.
© 2008, 2012 Zachary S Tseng E-2 - 18

The Cosine and Sine Series Extensions
If f and f ′ are piecewise continuous functions defined on the interval

0 ≤ t ≤ L, then f can be extended into an even periodic function, F, of period
2L, such that f (x) = F(x) on the interval [0, L], and whose Fourier series is,
therefore, a cosine series. Similarly, f can be extended into an odd periodic
function of period 2L, such that f (x) = F(x) on the interval (0, L), and whose
Fourier series is, therefore, a sine series. The process that such extensions
are obtained is often called cosine /sine series half-range expansions.
Here is an outline of how this can be done. Start with a function that is
defined only on an interval of finite length, from 0 to L. First expand the
function to be defined on the interval from −L to L such that the function is
an even or an odd function as required. Then define the function to be
periodic with a period of T = 2L by requiring F(x + 2L) = F(x). This process
is actually much easier than it sounds. Mathematically, the process can be
achieved rather simply, as described below.
Even (cosine series) extension of f (x)
Given f (x) defined on [0, L]. Its even extension of period 2L is:
 f ( x), 0 ≤ x ≤ L
F ( x) =  , F(x + 2L) = F(x).
 f (− x), − L < x < 0
a0 ∞ nπ x
Where F ( x) = + ∑ an cos , such that
2 n =1 L
mπ x
L
2
am =
L ∫
0
f ( x) cos
L
dx , m = 0, 1, 2, 3, …
bn = 0, n = 1, 2, 3, …
© 2008, 2012 Zachary S Tseng E-2 - 19

Odd (sine series) extension of f (x)
Given f (x) defined on (0, L). Its odd extension of period 2L is:
 f ( x), 0< x<L


F ( x) =  0 , x = 0, L , F(x + 2L) = F(x).
− f (− x), − L < x < 0

∞
nπ x
Where F ( x ) = ∑ bn sin , such that
n =1 L
am = 0, m = 0, 1, 2, 3, …
nπ x
L
2
bn =
L ∫
0
f ( x) sin
L
dx , n = 1, 2, 3, …
Example: Let f (x) = x, 0 ≤ x < 2. Find its cosine and sine series
extensions of period 4.
8 ∞
1 (2n − 1)π x
Answers: Cosine series: f ( x) = 1 − ∑
π 2 n =1 (2n − 1) 2
cos
2
4 (−1) n + 1
∞
nπ x
Sine series: f ( x) = ∑ sin
π n =1 n 2
© 2008, 2012 Zachary S Tseng E-2 - 20

Back to the Heat Conduction Problem
Previously, we had found the general solution of the initial-boundary value

problem given by the one-dimensional heat conduction equation modeling a
bar that has both of its ends kept at 0 degree. The general solution is
∞
nπ x
u ( x, t ) = ∑ C n e −α 2 n 2π 2 t / L2
sin
n =1 L .
Setting t = 0 and applying the initial condition u(x, 0) = f (x), we get
∞
nπ x
u ( x,0) = ∑ Cn sin = f ( x) .
n =1 L
We now know that the above equation says that the initial condition needs to
be an odd periodic function of period 2L. Since the initial condition could
be an arbitrary function, it usually means that we would need to “force the
issue” and expand it into an odd periodic function of period 2L. That is
∞
nπ x
f ( x ) = ∑ bn sin
n =1 L .
Compare the two expressions, we see that

∞
nπ x ∞
nπ x
u ( x,0) = ∑ Cn sin = f ( x) = ∑ bn sin
n =1 L n=1 L .
Therefore, the particular solution is found by setting all the coefficients

Cn = bn, where bn’s are the Fourier sine coefficients of (or the odd periodic
extension of) the initial condition f (x):
nπ x
L
2
Cn = bn = ∫ f ( x) sin dx .
L0 L
© 2008, 2012 Zachary S Tseng E-2 - 21

Example: Solve the heat conduction problem
8 uxx = ut , 0 < x < 5, t > 0,

u(0, t) = 0, and u(5, t) = 0,
u(x, 0) = 2sin(πx) − 4sin(2πx) + sin(5πx).
2
Since the standard form of the heat conduction equation is α uxx = ut,
2
we see that α = 8; and we also note that L = 5. Therefore, the general
solution is
∞
nπ x
u ( x, t ) = ∑ Cn e −α 2 n 2π 2 t / L2
sin
n =1 L
∞
nπ x
= ∑ Cn e −8 n 2π 2 t / 25
sin
n =1 5
The initial condition, f (x), is already an odd periodic function (notice

that it is a Fourier sine series) of the correct period T = 2L = 10.
Therefore, no additional calculation is needed, and all we need to do is
to extract the correct Fourier sine coefficients from f (x). To wit
C5 = b5 = 2,
C10 = b10 = −4,
C25 = b25 = 1,
Cn = bn = 0, for all other n, n ≠ 5, 10, or 25.
Hence,
2
)π 2 t / 25 2
)π 2 t / 25
u ( x, t ) = 2e−8(5 sin(πx) − 4e−8(10 sin(2πx)
2
)π 2 t / 25
+ e−8( 25 sin(5πx)
© 2008, 2012 Zachary S Tseng E-2 - 22

What will the particular solution be if the initial condition is u(x, 0) = x
instead? That is, solve the following heat conduction problem:
8 uxx = ut , 0 < x < 5, t > 0,

u(0, t) = 0, and u(5, t) = 0,
u(x, 0) = x.
The general solution is still

∞
nπ x
u ( x, t ) = ∑ Cn e−8n π
2 2
t / 25
sin
n =1 5 .
The initial condition is an odd function, but it is not a periodic

function. Therefore, it needs to be expanded into its odd periodic
extension of period 10 (T = 2L). Its coefficients are, for n = 1, 2, 3, …
5
nπ x nπ x
L
2 2
bn =
L ∫0
f ( x ) sin
L
dx = ∫ x sin
50 5
dx
2  − 5 x nπ x 
5 5
nπ x −5
5 0 nπ ∫0
= cos − cos dx
5  nπ 5 

5
2  − 5x nπ x 25 nπ x 
=  cos + 2 2 sin 
5  nπ 5 nπ 5 0


2   − 25  
=   cos( nπ ) − 0  − (0 − 0 )
5   nπ  
− 10
= cos( nπ )
nπ
 10
 nπ , n = odd (−1) n+110
= =
− 10 nπ
 , n = even
 nπ
© 2008, 2012 Zachary S Tseng E-2 - 23

The resulting sine series is (representing the function f (x) = x, −5 < x
< 5, f (x + 10) = f (x)):
10 (−1) n +1
∞
nπ x
f ( x) = ∑ sin .
π n =1 n 5
The particular solution can then be found by setting each coefficient,

Cn, to be the corresponding Fourier sine coefficient of the series above,
(−1) n+110
Cn = bn = . Therefore, the particular solution is
nπ
10 ∞
(−1)n +1 −8n 2π 2t / 25 nπ x
u( x, t ) =
π
∑n =1 n
e sin
5 .
© 2008, 2012 Zachary S Tseng E-2 - 24

Exercises E-2.1:
1 – 8 Find the Fourier series representation of each periodic function.

Determine the values to which each series converge to at x = 0.
1. f (x) = 5, −1 < x < 1, f (x + 2) = f (x).
2
2. f (x) = (cos x + sin x) , −π < x < π, f (x + 2π) = f (x).
3. f (x) = 6 − 3x, 0 ≤ x < 2, f (x + 2) = f (x).
2
4. f (x) = x , 0 ≤ x < π, f (x + π) = f (x).
2
5. f (x) = x , −π < x < π, f (x + 2π) = f (x).
 0, − 2 < x < 0

6. f ( x) = 4, 0 ≤ x ≤ 1 , f (x + 4) = f (x).
 0, 1 < x < 2

7. f (x) = │sin x│, −π < x < π, f (x + 2π) = f (x).
8. f (x) = δ(x − c), −π < x < π, 0 < c < π, f (x + 2π) = f (x).
9 – 12 Expand each function into its cosine series and sine series
representations of the indicated period. Determine the values to which each
series converge to at x = 0, x =2, and x = −2.
9. f (x) = 3 − x, T = 6. 10. f (x) = e, T = 2π.
 x, 0 ≤ x < 2
11. f (x) = sin x, T = 2π. 12. f ( x) =  , T = 6.
 2, 2 ≤ x < 3
13. Solve the heat conduction problem
2 uxx = ut , 0 < x < 9, t > 0,

u(0, t) = 0, and u(9, t) = 0,
u(x, 0) = 25sin(πx / 3) + 45sin(4πx / 3) − 12sin(3πx).
14. Solve the heat conduction problem of the given initial conditions.
9 uxx = ut , 0 < x < 12, t > 0,

u(0, t) = 0, and u(12, t) = 0,
(a) u(x, 0) = 3sin(πx) − sin(7πx / 6) − 6sin(2πx),

(b) u(x, 0) = 4,
(c) u(x, 0) = 4 − x.
© 2008, 2012 Zachary S Tseng E-2 - 25

Answers E-2.1:
1. f (x) = 5, f (0) = 5.
2. f (x) = 1 + sin(2x), f (0) = 1.
∞
6 1
3. f ( x) = 3 + ∑ sin( nπx) , f (0) = 3.
π n =1 n
π2 1 ∞
π  π2
4. f ( x) = + ∑  2 cos(2nx ) − sin( 2nx ) , f ( 0) = .
3 n =1  n n  2
π2 (−1) n ∞
5. f ( x) = + 4 ∑ 2 cos( nx ) , f (0) = 0.
3 n =1 n
4 ∞
1  nπ   nπx  1   nπ    nπx 
6. f ( x) = 1 +
π
∑  n sin   cos  + 1 − cos
2   2  n  2
  sin 
   2 
 ,
n =1 
f (0) = 2.
2 4 ∞ 1
7. f ( x) = − ∑ cos( 2nx ) , f (0) = 0.
π π n =1 (2n − 1)(2n + 1)
1 1 ∞
8. f ( x) = + ∑ cos( n( x − c )) , f (0) = 0.
2π π n =1
3 ∞ 12  nπx 
9. Cosine series: f ( x ) = + ∑ 2 2
cos ,
2 n =1 ( 2n − 1) π  3 
∞
9  nπx 
Sine series: f ( x ) = ∑ sin  ;
n = 1 nπ  3 
The cosine series converges to 3, 1, and 1; the sine series converges to 0, 1,
and −1, respectively, at x = 0, x =2, and x = −2.
10. Cosine series: f (x) = e,
∞
4e
Sine series: f ( x ) = ∑ sin( nx) ;
n =1 ( 2 n − 1)π
The cosine series converges to e at all 3 points. The sine series converges
to 0, e, and − e, respectively, at x = 0, x =2, and x = −2.
2 4 ∞ 1
11. Cosine series: f ( x ) = − ∑ cos(2nx ) ,
π π n =1 ( 2n − 1)(2n + 1)
Sine series: f (x) = sin x;
Both series converge to 0 at x = 0 and to sin(2) at x = 2. At x = −2, the
cosine series converges to sin(2), the sine series to − sin(2).
© 2008, 2012 Zachary S Tseng E-2 - 26

4 ∞ 6   2nπ    nπx 
12. Cosine series: f ( x ) = + ∑ 2 2  cos  − 1 cos ,
3 n =1 n π   3    3 
∞
 6  2nπ  4( −1)   nπx 
n
Sine series: f ( x) = ∑  2 2 sin  − sin  ;

n =1  n π  3  nπ   3 
Both series converge to 0 at x = 0 and to 2 at x = 2. At x = −2, the cosine
series converges to 2, the sine series to − 2.
−2π 2t / 9 πx 2 4πx 2
13. u( x, t ) = 25e sin( ) + 45e −32π t / 9 sin( ) −12e −18π t sin(3πx)

3 3
−9π 2t − 49π 2 t / 4 7πx −36π 2t
14. (a) u ( x, t ) = 3 e sin( π x) − e sin( ) − 6 e sin(2πx) ;
6
16 ∞ 1 −9n 2π 2 t / 144 nπ x
(b) u( x, t ) = ∑ e sin .
π n =1 2n − 1 12
© 2008, 2012 Zachary S Tseng E-2 - 27

Part III
One-dimensional Heat Conduction Equation revisited; temperature

distribution of a bar with insulated ends; nonhomogeneous boundary
conditions; temperature distribution of a bar with ends kept at arbitrary
temperatures; steady-state solution
Previously, we have learned that the general solution of a partial differential

equation is dependent of boundary conditions. The same equation will have
different general solutions under different sets of boundary conditions. We
shall witness this fact, by examining additional examples of heat conduction
problems with new sets of boundary conditions.
Keep in mind that, throughout this section, we will be solving the same
partial differential equation, the homogeneous one-dimensional heat
conduction equation:
α2 uxx = ut
where u(x, t) is the temperature distribution function of a thin bar, which has
2
length L, and the positive constant α is the thermo diffusivity constant of
the bar. The equation will now be paired up with new sets of boundary
conditions.
© 2008 Zachary S Tseng E-3 - 1

Bar with both ends insulated
Now let us consider the situation where, instead of them being kept at
constant 0 degree temperature, the two ends of the bar are also sealed with
perfect insulation so that no heat could escape to the outside environment
(recall that the side of the bar is always perfectly insulated in the one-
dimensional assumption), or vice versa. The new boundary conditions are
ux(0, t) = 0 and ux(L, t) = 0*, reflecting the fact that there will be no heat
transferring, spatially, across the points x = 0 and x = L. (Hence, this is a
Neumann type problem.) The heat conduction problem becomes the initial-
boundary value problem below.
(Heat conduction eq.) α2 uxx = ut , 0 < x < L, t > 0,
(Boundary conditions) ux(0, t) = 0, and ux(L, t) = 0,
(Initial condition) u(x, 0) = f (x).
The first step is the separation of variables. The equation is the same as
before. Therefore, it will separate into the exact same two ordinary
differential equations as in the first heat conduction problem seen earlier.
The new boundary conditions separate into
ux(0, t) = 0 → X ′(0)T(t) = 0 → X ′(0) = 0 or T(t) = 0

ux(L, t) = 0 → X ′(L)T(t) = 0 → X ′(L) = 0 or T(t) = 0
As before, we cannot choose T(t) = 0. Else we could only get the trivial
solution u(x,t) = 0, rather than the general solution. Hence, the new
boundary conditions should be X ′(0) = 0 and X ′(L) = 0.
Again, we end up with a system of two simultaneous ordinary differential

equations. Plus a set of two boundary conditions that goes with the spatial
independent variable x:
*
The conditions say that the instantaneous rate of change with respect to x, the spatial variable (i.e., the rate
of point-to-point heat transfer), is zero at each end. They do not suggest that the temperature is constant
(that is, there is no change in temperature through time, which would require ut = 0) at each end.

X ″ + λX = 0, X ′(0) = 0 and X ′(L) = 0,
T ′ + α2 λ T = 0 .
The second step is to solve the eigenvalue problem
X″ + λX = 0, X ′(0) = 0 and X ′(L) = 0.
The result is summarized below.
Case 1: If λ < 0: No such λ exists.
Case 2: If λ = 0: Zero is an eigenvalue. Its eigenfunction is the constant

function X0 = 1 (or any other nonzero constant).
Case 3: If λ > 0: The positive eigenvalues λ are
n2 π 2
λ= 2 , n = 1, 2, 3, …
L
The corresponding eigenfunctions that satisfy the said boundary conditions

are
nπ x
X n = cos
L , n = 1, 2, 3, …
The third step is to substitute the positive eigenvalues found above into the
equation of t and solve:
n 2π 2
T′+α 2
T =0.
L2

Notice that this is exactly the same equation as in the first (both ends kept at
0 degree) heat conduction problem, due to the fact that both problems have
the same set of eigenvalues (but with different eigenfunctions).
As a result, the solutions of the second equation are just the ones we have
gotten the last time
Tn (t ) = Cn e −α n π 2 t / L2
2 2
, n = 1, 2, 3, …
There is this extra eigenvalue of λ = 0 that also needs to be accounted for. It

has as an eigenfunction the constant X0(x) = 1. Put λ = 0 into the second
equation and we get T ′ = 0, which has only constant solutions T0(t) = C0.
Thus, we get the (arbitrary) constant function u0(x, t) = X0(x)T0(t) = C0 as a
solution. Therefore, the solutions of the one-dimensional heat conduction
equation, with the boundary conditions ux(0, t) = 0 and ux(L, t) = 0, are in the
form
u 0( x , t ) = C 0,
−α 2 n 2π 2 t / L2 nπ x
u n ( x, t ) = X n (t ) Tn (t ) = Cn e cos
L ,
n = 1, 2, 3, …
The general solution is their linear combination. Hence, for a bar with both
ends insulated, the heat conduction problem has general solution:
∞
nπ x
u ( x, t ) = C 0 + ∑ C n e −α 2 n 2π 2 t / L2
cos
n =1 L .

Now set t = 0 and equate it with the initial condition u(x, 0) = f (x):
∞
nπ x
u ( x,0) = C0 + ∑ Cn cos = f ( x) .
n =1 L
We see that the requirement is that the initial temperature distribution f (x)
must be a Fourier cosine series. That is, it needs to be an even periodic
function of period 2L. If f (x) is not already an even periodic function, then
we will need to expand it into one and use the resulting even periodic
extension of f (x) in its place in the above equation. Once this is done, the
coefficients C’s in the particular solution are just the corresponding Fourier
cosine coefficients of the initial condition f (x). (Except for the constant term,
where the relation C0 = a0 / 2 holds, instead.)
The explicit formula for Cn is, therefore,
nπ x
L
2
C n = an =
L ∫
0
f ( x ) cos
L
dx , n = 1, 2, 3, …
C0 = a 0 / 2

3 uxx = ut , 0 < x < 8, t > 0,

ux(0, t) = 0, and ux(8, t) = 0,
u(x, 0) = 9 − 3 cos(πx/4) − 6 cos(2πx).
2
First note that α = 3 and L = 8, and the fact that the boundary
conditions indicating this is a bar with both ends perfectly insulated.
Substitute them into the formula we have just derived to obtain the
general solution for this problem:
∞
nπ x
u ( x, t ) = C0 + ∑ Cn e −3n 2π 2 t / 64
cos
n =1 8 .
Check the initial condition f (x), and we see that it is already in the
require form of a Fourier cosine series of period 16. Therefore, there
is no need to find its even periodic extension. Instead, we just need to
extract the correct Fourier cosine coefficients from f (x):
C0 = a0 / 2 = 9,
C2 = a2 = −3,
C16 = a16 = −6,
Cn = an = 0, for all other n, n ≠ 0, 2, or 16.
Note that C0 is actually a0 / 2, due to the way we write the constant

term of the Fourier series. But that shouldn’t present any more
difficulty. Since when you see a Fourier series, its constant term is
already expressed in the form a0 / 2. Therefore, you could just copy it
down directly to be the C0 term without thinking.
Finally, the particular solution is
2
)π 2 t / 64 πx 2
)π 2 t / 64
u ( x, t ) = 9 − 3e−3( 2 cos( ) − 6e−3(16 cos(2π x)
4

Bar with two ends kept at arbitrary temperatures:
An example of nonhomogeneous boundary conditions
In both of the heat conduction initial-boundary value problems we have seen,

the boundary conditions are homogeneous − they are all zeros. Now let us
look at an example of heat conduction problem with simple
nonhomogeneous boundary conditions. The general set-up is the same as
the first example (where the both ends of the bar were kept at constant 0
degree, but were not insulated), except now the ends are kept at arbitrary
(but constant) temperatures of T1 degrees at the left end, and T2 degrees at
the right end. The initial condition, as usual, is arbitrary. The heat
conduction problem is therefore given by the initial-boundary value problem:
α2 uxx = ut , 0 < x < L, t > 0,
u(0, t) = T1, and u(L, t) = T2,
u(x, 0) = f (x).
The boundary conditions is now nonhomogeneous (unless T1 and T2 are both

0, then the problem becomes identical to the earlier example), because at
least one of the boundary values are nonzero.
The nonhomogeneous boundary conditions are rather easy to work with,

more so than we might have reasonably expected. First, let us be introduced
to the concept of the steady-state solution. It is the part of the solution u(x, t)
that is independent of the time variable t. Therefore, it is a function of the
spatial variable alone. We can thusly rewrite the solution u(x, t) as a sum of
2 parts, a time-independent part and a time-dependent part:
u(x, t) = v(x) + w(x, t).
Where v(x) is the steady-state solution, which is independent of t, and w(x, t)

is called the transient solution, which does vary with t.

The Steady-State Solution
The steady-state solution, v(x), of a heat conduction problem is the part of

the temperature distribution function that is independent of time t. It
represents the equilibrium temperature distribution. To find it, we note the
fact that it is a function of x alone, yet it has to satisfy the heat conduction
equation. Since vxx = v″ and vt = 0, substituting them into the heat
conduction equation we get
α2 vxx = 0.
Divide both sides by α2 and integrate twice with respect to x, we find that
v(x) must be in the form of a degree 1 polynomial:
v(x) = Ax + B.
Then, rewrite the boundary conditions in terms of v: u(0, t) = v(0) = T1, and
u(L, t) = v(L) = T2. Apply those 2 conditions to find that:
v(0) = T1 = A(0) + B = B → B = T1
v(L) = T2 = AL + B = AL + T1 → A = (T2 − T1) / L
Therefore,
T2 − T1
v( x ) = x + T1 .
L
Thing to remember: The steady-state solution is a time-independent

function. It is obtained by setting the partial derivative(s) with respect to t in
the heat equation (or, later on, the wave equation) to constant zero, and then
solving the equation for a function that depends only on the spatial variable x.

Comment: Another way to understand the behavior of v(x) is to think from
the perspective of separation of variables. You could think of the steady-
state solution as, during the separation of variables, the solution you would
have obtained if T(t) = 1, the constant function 1. Therefore, the solution is
independent of time, or time-invariant. Hence, u(x, t) = X(x)T(t) = X(x) =
v(x). We can, in addition, readily see the substitutions required for rewriting
the boundary conditions prior to solving for the steady-state solution:
u(0, t) = X(0) = v(0) = T1, and u(L, t) = X(L) = v(L) = T2.
That is, just rename the function u as v, ignore the time variable t, and put
whatever x-coordinate specified directly into v(x).

The solution of bar with two ends kept at arbitrary temperatures
Once the steady-state solution has been found, we can set it aside for the
time being and proceed to find the transient part of solution, w(x, t). First we
will need to rewrite the given initial-boundary value problem slightly. Keep
in mind that the initial and boundary conditions as originally given were
meant for the temperature distribution function u(x, t) = v(x) + w(x, t). Since
we have already found v(x), we shall now subtract out the contribution of v(x)
from the initial and boundary values. The results will be the conditions that
the transient solution w(x, t) alone must satisfy.
Change in the boundary conditions:
u(0, t) = T1 = v(0) + w(0, t) → w(0, t) = T1 − v(0) = 0

u(L, t) = T2 = v(L) + w(L, t) → w(L, t) = T2 − v(L) = 0
Note: Recall that u(0, t) = v(0) = T1, and u(L, t) = v(L) = T2.
Change in the initial condition:
u(x, 0) = f (x) = v(x) + w(x, 0) → w(x, 0) = f (x) − v(x)
Consequently, the transient solution is a function of both x and t that must

satisfy the new initial-boundary value problem:
α2 wxx = wt , 0 < x < L, t > 0,
w(0, t) = 0, and w(L, t) = 0,
w(x, 0) = f (x) − v(x).

Surprise! Notice that the new problem just described is precisely the same
initial-boundary value problem associated with the heat conduction of a bar
with both ends kept at 0 degree. Therefore, the transient solution w(x, t) of
the current problem is just the general solution of the previous heat
conduction problem (with homogeneous boundary conditions), that of a bar
with 2 ends kept constantly at 0 degree:
∞
nπ x
w( x, t ) = ∑ Cn e−α n π 2 t / L2
2 2
sin
n =1 L .
Where the coefficients Cn are equal to the corresponding Fourier sine
coefficients bn of the (newly rewritten) initial condition w(x, 0) = f (x) − v(x).
(Or those of w(x, 0)’s odd periodic extension, of period 2L, if it is not already
an odd periodic function of the correct period.) Explicitly, they are given by
nπ x
L
2
Cn = bn =
L ∫ ( f ( x) − v( x) )sin
0
L
dx , n = 1, 2, 3, …
Finally, combining the steady-state and transient solutions together, the

general solution of the temperature distribution of a bar whose ends are kept
at T1 degrees at the left, and T2 degrees at the right, becomes
u ( x, t ) = v( x) + w( x, t )
 T2 − T1  ∞ nπ x
x + T1  + ∑ Cn e−α n π t / L sin
2 2 2 2
= .
 L  n =1 L

8 uxx = ut , 0 < x < 5, t > 0,

u(0, t) = 10, u(5, t) = 90,
u(x, 0) = 16x + 10 + 2sin(πx) − 4sin(2πx) + sin(6πx).
2
First we note that α = 8 and L = 5. Since T1 = 10 and T2 = 90, the
steady-state solution is v(x) = (90 − 10)x / 5 + 10 = 16x + 10. We then
subtract v(0) = 10 from u(0, t), v(5) = 90 from u(5, t), and v(x) = 16x +
10 from u(x, 0) to obtain a new set of initial-boundary values that the
transient solution w(x,t) alone must satisfy:
w(0, t) = 0, and w(5, t) = 0,

w(x, 0) = 2sin(πx) − 4sin(2πx) + sin(6πx).
2
Base on α = 8 and L = 5, we write down the general solution:
∞
nπ x
w( x, t ) = ∑ Cn e−8n π
2 2
t / 25
sin
n =1 5
The new initial condition, f (x) − v(x), is already an odd periodic

function of the period T = 2L = 10. Therefore, just extract the correct
Fourier sine coefficients from it:
C5 = b5 = 2,
C10 = b10 = −4,
C30 = b30 = 1,
Cn = bn = 0, for all other n, n ≠ 5, 10, or 30.
Add together the steady-state and transient solutions, we have
2
)π 2 t / 25
u ( x, t ) = v( x) + w( x, t ) = 16 x + 10 + 2e−8(5 sin(π x)
−8(10 2 )π 2 t / 25 −8 ( 30 2 )π 2 t / 25
− 4e sin(2π x) + e sin(6π x)

Back to the Steady-State Solution
A thing to remember: the steady-state solution of the one-dimensional

homogeneous heat conduction equation is always in the form v(x) = Ax + B.
Since it is independent of t, the effects of boundary conditions on v(x) are
also simplified as: u(x0, t) = v(x0), and ux(x0, t) = v′(x0).
Nonhomogeneous heat conduction equations (that is, the equations

themselves contain forcing function terms; not to be confused with the
homogeneous equation with accompanying nonhomogeneous boundary
conditions that we have just seen), however, could have different forms of
v(x).
Fact: The steady-state temperature distribution satisfies the property:
lim u ( x, t ) = v( x) .
t →∞
This relation is true only for the solution of heat conduction equation
(modeling diffusion-like processes that are thermodynamically irreversible).
Physically speaking, v(x) describes the eventual state of maximum entropy
as dictated by the second law of Thermodynamics.
Caution: The above relation is not true, in general, for solutions of the wave
equation in the next section. This difference is due to the fact that the wave
equation models wave-like motions which are thermodynamically reversible
processes.

Further examples of steady-state solutions of the heat conduction equation:
Find v(x), given each set of boundary conditions below.
1. u(0, t) = 50, ux(6, t) = 0
We are looking for a function of the form v(x) = Ax + B that satisfies

the given boundary conditions. Its derivative is then v′(x) = A. The
two boundary conditions can be rewritten to be u(0, t) = v(0) = 50, and
ux(6, t) = v′(6) = 0. Hence,
v(0) = 50 = A(0) + B = B → B = 50
v′(6) = 0 = A → A=0
Therefore, v(x) = 0x + 50 = 50.
2. u(0, t) − 4ux(0, t) = 0, ux(10, t) = 25
The two boundary conditions can be rewritten to be

v(0) − 4 v′(0) = 0, and v′(10) = 25.
Hence,
v(0) − 4 v′(0) = 0 = (A(0) + B) − 4A = −4A + B
v′(10) = 25 = A → A = 25
Substitute A = 25 into the first equation: 0 = −4A + B = −100 + B

→ B = 100
Therefore, v(x) = 25x + 100.

3. u(0, t) = 35, u(4, t) + 3ux(4, t) = 0
Rewriting the boundary conditions:

v(0) = 35, and v(4) + 3 v′(4) = 0.
Hence,
v(0) = 35 = A(0) + B = B → B = 35
v(4) + 3 v′(4) = 0 = (A(4) + B) + 3A = 7A + B
0 = 7A + 35
A = −5
Therefore, v(x) = −5x + 35
Comment: Notice how in each example we have seen, the steady-state

solution is uniquely determined by the boundary conditions alone. In
general this is true – that the steady-state solution is almost always
independent of the initial condition. The lone exception is the insulated-
ends problem (with boundary conditions ux(0, t) = 0 = ux(L, t) ). In this
special case, the boundary conditions only tell us that the steady-state
solution should be a constant, which turns out to be the constant term of the
general solution. As we have seen, that constant is indeed dependent on the
initial condition – it is just the constant term of the initial condition, as the
latter is expanded into a Fourier cosine series of period 2L. In all other
cases the boundary conditions alone determine the steady-state solution
(therefore, the limiting temperature) of a problem.

Summary: Solving Second Order Linear Partial Differential
Equations
The Method of Separation of Variables:
0. (If the boundary conditions are nonhomogeneous) Solve for the

steady-state solution, v(x), which is a function of x only that satisfies
both the PDE and the boundary conditions. Afterwards rewrite the
problem’s boundary and initial conditions to subtract out the
contribution from the steady-state solution. Therefore, the problem is
now transformed into one with homogeneous boundary conditions.
1. Separate the PDE into ODEs of one independent variable each.

Rewrite the boundary conditions so they associate with only one of
the variables.
2. One of the ODEs is a part of a two-point boundary value problem.

Solve this problem for its eigenvalues and eigenfunctions.
3. Solve the other ordinary differential equation.
4. Multiply the results from steps (2) and (3), and sum up all the
products to find the general solution respect to the given
homogeneous boundary conditions. Add to it the steady-state solution
(from step 0, if applicable) to find the overall general solution.
5. Expand the initial condition into a suitable (e.g. a sine or a cosine

series, depending on whichever type the eigenfunctions are) Fourier
series. Then compare it against u(x, 0) to find the coefficients for the
particular solution.

Summary of Heat Conduction Problems
Here is a list of heat conduction problems and their solutions. All solutions
obey the homogeneous one-dimensional heat conduction equation
α2 uxx = ut .
They only differ in boundary conditions (which are given below for each
problem). The initial condition is always arbitrary, u(x, 0) = f (x).
1. Bar with both ends kept at 0 degree (boundary conditions: u(0, t) = 0,

u(L, t) = 0)
∞
nπ x
u ( x, t ) = ∑ Cn e −α 2 n 2π 2 t / L2
sin
n =1 L .
Expand f (x) to be a Fourier sine series, then Cn = bn.
Steady-state solution is v(x) = 0.
2. Bar with both ends perfectly insulated (boundary conditions:

ux(0, t) = 0, ux(L, t) = 0)
∞
nπ x
u ( x, t ) = C 0 + ∑ C n e −α 2 n 2π 2 t / L2
cos
n =1 L .
Expand f (x) to be a Fourier cosine series, then C0 = a0 / 2, and Cn = an,
n = 1, 2, 3, …
Steady-state solution is v(x) = C0.
3. Bar with T1 degrees at the left end, and T2 degrees at the right end
(boundary conditions: u(0, t) = T1, u(L, t) = T2)
 T2 − T1  ∞ nπ x
u ( x, t ) =  x + T1  + ∑ Cn e −α 2 n 2π 2 t / L2
sin
 L  n =1 L .
Expand (f (x) − v(x)) to be a Fourier sine series, then Cn = bn.
T2 − T1
Steady-state solution is v( x) = x + T1 .
L

Exercises E-3.1:
1 – 7 Find the steady-state solution v(x) of the heat conduction equation,

given each set of boundary conditions below.
1. u(0, t) = 200, u(10, t) = 100
2. u(0, t) = 100, ux(10, t) = 50
3. ux(0, t) = 8, u(10, t) = 100
4. ux(0, t) = 30, ux(10, t) = 10
5. u(0, t) + ux(0, t) = 10, u(10, t) = 100
6. u(0, t) + ux(0, t) = 0, u(10, t) − ux(10, t) = 200
7. u(0, t) − 10ux(0, t) = 30, u(10, t) − 5ux(10, t) = 0
2 uxx = ut , 0 < x < 6, t > 0,

ux(0, t) = 0, and ux(6, t) = 0,
(a) u(x, 0) = π + 3cos(πx) − 4cos(3πx / 2) − cos(3πx),

(b) u(x, 0) = 4,
(c) u(x, 0) = x2,
(d) u(x, 0) = 0.
9. For each particular solution found in #8, find lim

t →∞
u ( x, t ) .
2 uxx = ut , 0 < x < 6, t > 0,

u(0, t) = 40, and u(6, t) = 10,
(a) u(x, 0) = −5x + 40 + 5sin(2πx) − 2sin(5πx / 2),

(b) u(x, 0) = 0.
11. For each particular solution found in #10, find lim

t →∞
u (4, t ) .

12. Consider the heat conduction problem
9 uxx = ut , 0 < x < 4, t > 0,

u(0, t) = 32, and u(4, t) = 32,
u(x, 0) = f (x).
(a) Find its general solution.

(b) Write an explicit formula to determine the coefficients.
(c) Base on (a), find lim
t →∞
u ( x, t ) .
13. Consider the heat conduction equation below, subject to each of the 4
sets of boundary conditions.
9 uxx = ut , 0 < x < 10, t > 0,
(i) u(0, t) = 0, and u(10, t) = 0,

(ii) ux(0, t) = 0, and ux(10, t) = 0,
(iii) u(0, t) = 0, and u(10, t) = 100
(iv) u(0, t) = 100, and u(10, t) = 50.
Given the common initial condition u(x, 0) = 300, determine the temperature
at the midpoint of the bar (at x = 5) after a very long time has elapsed.
Which set of boundary conditions will give the highest temperature at that
point?

Answers E-3.1:
1. v(x) = −10x + 200 2. v(x) = 50x + 100 3. v(x) = 8x + 20

4. v(x) does not exist. 5. v(x) = 10x 6. v(x) = 25x − 25
7. v(x) = −2x + 10
u ( x, t ) = π + 3e −2π t cos(πx) − 4e −9π t / 2 cos(3πx / 2) − e −18π t cos(3πx) ,
2 2 2
8. (a)
(b) u(x, t) = 4,
(d) u(x, t) = 0.
9. (a) lim u ( x, t ) = π ; (b) lim u ( x, t ) = 4 ; (c) lim u ( x, t ) = 25 / 3 ;
t →∞ t →∞ t →∞
(d) lim u ( x, t ) = 0 .
t →∞
−8π t
sin(2πx) − 2e −25π
2 2
10. (a) u ( x, t ) = −5 x + 40 + 5e
t/2
sin(5πx / 2) .
11. (a) and (b) lim u (4, t ) = 20 .
t →∞
∞
nπ x
12. (a) u ( x, t ) = 32 + ∑ Cn e
−9 n π 2 2
t / 16
sin
n =1 4
nπ x
4
1
(b) C n = ∫ ( f ( x ) − 32 ) sin dx , n = 1, 2, 3, …
2 0
4
(c) lim u ( x, t ) = 32
t →∞
13. Boundary conditions (ii) will give the highest temperature.

At x = 5, the temperature is lim u (5, t ) = v(5) = 300.
t →∞

Part IV
One-dimensional undamped wave equation; D’Alembert solution of the

wave equation; damped wave equation and the general wave equation; two-
dimensional Laplace equation
The second type of second order linear partial differential equations in 2

independent variables is the one-dimensional wave equation. Together with
the heat conduction equation, they are sometimes referred to as the
“evolution equations” because their solutions “evolve”, or change, with
passing time. The simplest instance of the one-dimensional wave equation
problem can be illustrated by the equation that describes the standing wave
exhibited by the motion of a piece of undamped vibrating elastic string.
© 2008, 2012 Zachary S Tseng E-4 - 1

Undamped One-Dimensional Wave Equation:
Vibrations of an Elastic String
Consider a piece of thin flexible string of length L, of negligible weight.

Suppose the two ends of the string are firmly secured (“clamped”) at some
supports so they will not move. Assume the set-up has no damping. Then,
the vertical displacement of the string, 0 < x < L, and at any time t > 0, is
given by the displacement function u(x, t). It satisfies the homogeneous one-
dimensional undamped wave equation:
a2 uxx = utt
Where the constant coefficient a2 is given by the formula a2 = T / ρ, such that

a = horizontal propagation speed (also known as phase velocity) of the wave
motion, T = force of tension exerted on the string, ρ = mass density (mass
per unit length). It is subjected to the homogeneous boundary conditions
u(0, t) = 0, and u(L, t) = 0, t > 0.
The two boundary conditions reflect that the two ends of the string are
clamped in fixed positions. Therefore, they are held motionless at all time.
The equation comes with 2 initial conditions, due to the fact that it contains
the second partial derivative of time, utt. The two initial conditions are the
initial (vertical) displacement u(x, 0), and the initial (vertical) velocity
ut(x, 0)*, both are arbitrary functions of x alone. (Note that the string is
merely the medium for the wave, it does not itself move horizontally, it only
vibrates, vertically, in place. The resulting undulation, or the wave-like
“shape” of the string, is what moves horizontally.)
*
Velocity = rate of change of displacement with respect to time. The other first partial derivative ux
represents the slope of the string at a point x and time t.
© 2008, 2012 Zachary S Tseng E-4 - 2

Hence, what we have is the following initial-boundary value problem:
(Wave equation) a2 uxx = utt , 0 < x < L, t > 0,
(Boundary conditions) u(0, t) = 0, and u(L, t) = 0,
(Initial conditions) u(x, 0) = f (x), and ut(x, 0) = g(x).
We first let u(x, t) = X(x)T(t) and separate the wave equation into two
ordinary differential equations. Substituting uxx = X ″ T and utt = X T ″ into
the wave equation, it becomes
a2 X ″ T = X T ″.
© 2008, 2012 Zachary S Tseng E-4 - 3

2
Dividing both sides by a X T :
X ′′ T ′′
=
X a2 T
As for the heat conduction equation, it is customary to consider the constant

a2 as a function of t and group it with the rest of t-terms. Insert the constant
of separation and break apart the equation:
X ′′ T ′′
= =−λ
X a2 T
X ′′
=−λ → X ″ = −λX → X ″ + λX = 0,
X
T ′′
=−λ → T ″ = −a2 λ T → T ″ + a2 λ T = 0.
a2 T
The boundary conditions also separate:
u(0, t) = 0 → X(0)T(t) = 0 → X(0) = 0 or T(t) = 0

u(L, t) = 0 → X(L)T(t) = 0 → X(L) = 0 or T(t) = 0
As usual, in order to obtain nontrivial solutions, we need to choose

X(0) = 0 and X(L) = 0 as the new boundary conditions. The result,
after separation of variables, is the following simultaneous system of
ordinary differential equations, with a set of boundary conditions:
X ″ + λX = 0, X(0) = 0 and X(L) = 0,
T ″ + a2 λ T = 0.
© 2008, 2012 Zachary S Tseng E-4 - 4

The next step is to solve the eigenvalue problem
X ″ + λX = 0, X(0) = 0, X(L) = 0.
We have already solved this eigenvalue problem, recall. The solutions are
n2 π 2
Eigenvalues: λ= 2 , n = 1, 2, 3, …
L
nπ x
Eigenfunctions: X n = sin , n = 1, 2, 3, …
L
Next, substitute the eigenvalues found above into the second equation to find
T(t). After putting eigenvalues λ into it, the equation of T becomes
n 2π 2
T ′′ + a
2
T =0.
L2
It is a second order homogeneous linear equation with constant coefficients.

It’s characteristic have a pair of purely imaginary complex conjugate roots:
anπ
r=± i.
L
Thus, the solutions are simple harmonic:
anπ t anπ t
Tn (t ) = An cos + Bn sin
L L , n = 1, 2, 3, …
Multiplying each pair of Xn and Tn together and sum them up, we find the
general solution of the one-dimensional wave equation, with both ends fixed,
to be
© 2008, 2012 Zachary S Tseng E-4 - 5

∞
 anπ t anπ t  nπ x
u ( x, t ) = ∑  An cos + Bn sin  sin
n =1  L L  L .
There are two sets of (infinitely many) arbitrary coefficients. We can solve
for them using the two initial conditions.
Set t = 0 and apply the first initial condition, the initial (vertical)
displacement of the string u(x, 0) = f (x), we have
∞
nπ x
u ( x,0) = ∑ ( An cos(0) + Bn sin(0) ) sin
n =1 L
∞
nπ x
= ∑ An sin = f ( x)
n =1 L
Therefore, we see that the initial displacement f (x) needs to be a Fourier sine
series. Since f (x) can be an arbitrary function, this usually means that we
need to expand it into its odd periodic extension (of period 2L). The
coefficients An are then found by the relation An = bn, where bn are the
corresponding Fourier sine coefficients of f (x). That is
nπ x
L
2
An = bn = ∫ f ( x ) sin dx .
L0 L
Notice that the entire sequence of the coefficients An are determined exactly
by the initial displacement. They are completely independent of the other
sequence of coefficients Bn, which are determined solely by the second
initial condition, the initial (vertical) velocity of the string. To find Bn, we
differentiate u(x, t) with respect to t and apply the initial velocity,
ut(x, 0) = g(x).
© 2008, 2012 Zachary S Tseng E-4 - 6

∞
 anπ anπ t anπ anπ t  nπ x
ut ( x, t ) = ∑  − An sin + Bn cos  sin
n =1  L L L L  L
Set t = 0 and equate it with g(x):
∞
anπ nπ x
ut ( x,0) = ∑ Bn sin = g ( x) .
n =1 L L
We see that g(x) needs also be a Fourier sine series. Expand it into its odd
periodic extension (period 2L), if necessary. Once g(x) is written into a sine
series, the previous equation becomes
∞
anπ nπ x ∞
nπ x
ut ( x,0) = ∑ Bn sin = g ( x) = ∑ bn sin
n =1 L L n =1 L
Compare the coefficients of the like sine terms, we see

anπ nπ x
L
2
Bn = bn = ∫ g ( x )sin dx .
L L0 L
Therefore,
nπ x
L
L 2
Bn =
anπ
bn =
anπ ∫ g ( x) sin
0
L
dx .
As we have seen, half of the particular solution is determined by the initial

displacement, the other half by the initial velocity. The two halves are
determined independent of each other. Hence, if the initial displacement
f (x) = 0, then all An = 0 and u(x, t) contains no sine-terms of t. If the initial
velocity g(x) = 0, then all Bn = 0 and u(x, t) contains no cosine-terms of t.
© 2008, 2012 Zachary S Tseng E-4 - 7

Let us take a closer look and summarize the result for these 2 easy special
cases, when either f (x) or g(x) is zero.
Special case I: Nonzero initial displacement, zero initial velocity: f (x) ≠ 0,

g(x) = 0.
Since g(x) = 0, then Bn = 0 for all n.
nπ x
L
2
An = ∫ f ( x ) sin dx , n = 1, 2, 3, …
L0 L
Therefore,
∞
anπ t nπ x
u ( x, t ) = ∑ An cos sin
n =1 L L .
© 2008, 2012 Zachary S Tseng E-4 - 8

The D’Alembert Solution
In 1746, Jean D’Alembert† produced an alternate form of solution to the

wave equation. His solution takes on an especially simple form in the above
case of zero initial velocity.
Use the product formula sin(A) cos(B) = [sin(A − B) + sin(A + B)] / 2, the
solution above can be rewritten as
1 ∞  nπ ( x − at ) nπ ( x + at ) 
u ( x, t ) = ∑ An  sin + sin 
2 n =1  L L 
Therefore, the solution of the undamped one-dimensional wave equation

with zero initial velocity can be alternatively expressed as
u(x, t) = [F(x − at) + F(x + at)] / 2.
In which F(x) is the odd periodic extension (period 2L) of the initial
displacement f (x).
An interesting aspect of the D’Alembert solution is that it readily shows that

the starting waveform given by the initial displacement would keep its
general shape, but it would also split exactly into two halves. The two
halves of the wave form travel in the opposite directions at the same finite
speed of propagation a. This can be seen by the fact that the two halves of
the wave form, in terms of x, are being translated/moved in the opposite
direction, to the right and left, in the form of phase shifts, at the rate of
distance a units per unit time. Hence the value a is also known as the
wave’s phase velocity.
†
Jean le Rond d”Alembert (1717 – 1783) was a French mathematician and physicist. He is perhaps best
known to calculus students as the inventor of the Ratio Test for convergence.
© 2008, 2012 Zachary S Tseng E-4 - 9

Furthermore, once the “wave front” has passed over a point on the string, the
displacement at that point will be restored to its previous state before the
arrival of the wave. In physics, this aspect of a clearly-defined, echo-less,
wave motion of a one-dimensional wave is called the Huygens’ Principle.
(The principle also holds for solutions of a three-dimensional wave equation.
But it is not true for two-dimensional waves.)
Special case II: Zero initial displacement, nonzero initial velocity: f (x) = 0,
g(x) ≠ 0.
Since f (x) = 0, then An = 0 for all n.
nπ x
L
2
Bn =
anπ ∫
0
g ( x) sin
L
dx , n = 1, 2, 3, …
Therefore,
∞
anπ t nπ x
u ( x, t ) = ∑ Bn sin sin
n =1 L L .
© 2008, 2012 Zachary S Tseng E-4 - 10

Example: Solve the one-dimensional wave problem
9 uxx = utt , 0 < x < 5, t > 0,

u(0, t) = 0, and u(5, t) = 0,
u(x, 0) = 4sin(πx) − sin(2πx) − 3sin(5πx),
ut(x, 0) = 0.
2
First note that a = 9 (so a = 3), and L = 5.
The general solution is, therefore,
∞
 3nπ t 3nπ t  nπ x
n =1  5 5  5 .
Since g(x) = 0, it must be that all Bn = 0. We just need to find An. We

also see that u(x, 0) = f (x) is already in the form of a Fourier sine
series. Therefore, we just need to extract the corresponding Fourier
sine coefficients:
A5 = b5 = 4,
A10 = b10 = −1,
A25 =b25 = −3,
An = bn = 0, for all other n, n ≠ 5, 10, or 25.
Hence, the particular solution is
u(x, t) = 4cos(3π t) sin(π x) − cos(6π t) sin(2π x)

− 3cos(15π t) sin(5π x).
© 2008, 2012 Zachary S Tseng E-4 - 11

We can also solve the previous example using D’Alembert’s solution. The
problem has zero initial velocity and its initial displacement has already been
expanded into the required Fourier sine series, u(x, 0) = 4sin(πx) − sin(2πx) −
3sin(5πx) = F(x). Therefore, the solution can also be found by using the
formula u(x, t) = [F(x − at) + F(x + at)] / 2, where a = 3. Thus
u(x, t) = [ [ 4sin(π(x + 3t)) + 4sin(π(x − 3t)) ] − [sin(2π(x +

3t)) + sin(2π(x + 3t)) ] − [3sin(5π(x + 3t)) + 3sin(5π(x +
3t)) ] ] / 2
Indeed, you could easily verify (do this as an exercise) that the solution
obtained this way is identical to our previous answer. Just apply the addition
formula of sine function ( sin(α ± β) = sin(α)cos(β) ± cos(α)sin(β) ) to each
term in the above solution and simplify.
© 2008, 2012 Zachary S Tseng E-4 - 12

Example: Solve the one-dimensional wave problem
9 uxx = utt , 0 < x < 5, t > 0,

u(0, t) = 0, and u(5, t) = 0,
u(x, 0) = 0,
ut(x, 0) = 4.
2
As in the previous example, a = 9 (so a = 3), and L = 5.
Therefore, the general solution remains
∞
 3nπ t 3nπ t  nπ x
n =1  5 5  5 .
Now, f (x) = 0, consequently all An = 0. We just need to find Bn. The

initial velocity g(x) = 4 is a constant function. It is not an odd periodic
function. Therefore, we need to expand it into its odd periodic
extension (period T = 10), then equate it with ut(x, 0). In short:
nπ x nπ x
L 5
2 2
Bn =
anπ ∫
0
g ( x) sin
L
dx =
3nπ ∫0
4 sin
5
dx
 80
 , n = odd
=  3n 2π 2
 0 , n = even
Therefore,
∞
80 3( 2n − 1)π t ( 2n − 1)π x
u ( x, t ) = ∑ sin sin
n =1 3( 2n − 1) 2 π 2 5 5 .
© 2008, 2012 Zachary S Tseng E-4 - 13

The Structure of the Solutions of the Wave Equation
In addition to the fact that the constant a is the standing wave’s propagation
speed, several other observations can be readily made from the solution of
the wave equation that give insights to the nature of the solution.
To reduce the clutter, let us look at the form of the solution when there is no
initial velocity (when g(x) = 0). The solution is
∞
anπ t nπ x
u ( x, t ) = ∑ An cos sin
n =1 L L .
The sine terms are functions of x. They described the spatial wave patterns
(the wavy “shape” of the string that we could visually observe), called the
normal modes, or natural modes. The frequencies of those sine waves that
we could see, nπ / L, are called the spatial frequencies of the wave. They are
also known as the wave numbers. It measures the angular motion, in radians,
per unit distance that the wave travels. The “period” of each spatial (sine)
function, 2/( nπ / L) = 2L / n, is the wave length of each term. Meanwhile, the
cosine terms are functions of t, they give the vertical displacement of the
string relative to its equilibrium position (which is just the horizontal, or the
x-axis). They describe the up-and-down vibrating motion of the string at
each point of the string. These temporal frequencies (the frequencies of
functions of t; in this case, the cosines’) are the actual frequencies of
oscillating motion of vertical displacement. Since this is the undamped
wave equation, the motion of the string is simple harmonic. The frequencies
of the cosine terms, anπ / L (measured in radians per second), are called the
natural frequencies of the string. In a string instrument, they are the
frequencies of the sound that we could hear. The corresponding natural
periods (= 2π / natural frequency) are, therefore, T = 2L / an.
For n = 1, the observable spatial wave pattern is that of sin(πx / L). The wave
length is 2L, meaning the length L string carries only a half period of the
sinusoidal motion. It is the string’s first natural mode. The first natural
© 2008, 2012 Zachary S Tseng E-4 - 14

frequency of oscillation, aπ / L, is called the fundamental frequency of the
string. It is, given the set-up, the lowest frequency note the vibrating string
can produce. It is also called, in acoustics, as the first harmonic of the string.
For n = 2, the spatial wave pattern is sin(2πx / L) is the second natural mode.
Its wavelength is L, which is the length of the string itself. The second
natural frequency of oscillation, 2aπ / L, is also called the second harmonic,
or the first overtone, of the string. It is exactly twice of the string’s
fundamental frequency; hence its wavelength (= L) is only half as long.
Acoustically, it produces a tone that is exactly one octave higher than the
first harmonic. For n = 3, the third natural frequency, 3aπ / L, is also called
the third harmonic, or the second overtone. It is 3 times larger than the
fundamental frequency and, at a 3:2 ratio over the second harmonic, is
situated exactly halfway between the adjacent octaves (at the second and the
fourth harmonics). The fourth natural frequency (fourth harmonic/ third
overtone), 4aπ / L, is four times larger than the fundamental frequency and
twice of that the second natural frequency. The tone it produces is, therefore,
exactly 2 octaves and 1 octave higher than those generated by the first and
second harmonics, respectively. Together, the sequence of all positive
integer multiples of the fundamental frequency is called a harmonic series
(not to be confused with that other harmonic series that you have studied in
calculus).
The motion of the string is the combination of all its natural modes, as
indicated by the terms of the infinite series of the general solution. The
presence, and magnitude, of the nature modes are solely determined by the
(Fourier sine series expansion of) initial conditions.
Lastly, notice that the “wavelike” behavior of the solution of the undamped
wave equation, quite unlike the solution of the heat conduction equation
discussed earlier, does not decrease in amplitude/intensity with time. It
never reaches a steady state (unless the solution is trivial, u(x, t) = 0, which
occurs when f (x) = g(x) = 0). This is a consequence of the fact that the
undamped wave motion is a thermodynamically reversible process that
needs not obey the second law of Thermodynamics.
© 2008, 2012 Zachary S Tseng E-4 - 15

First natural mode (oscillates at the fundamental frequency / 1st harmonic):
Second natural mode (oscillates at the 2nd natural frequency / 2nd harmonic):
Third natural mode (oscillates at the 3rd natural frequency / 3rd harmonic):
© 2008, 2012 Zachary S Tseng E-4 - 16

Summary of Wave Equation: Vibrating String Problems
The vertical displacement of a vibrating string of length L, securely clamped

at both ends, of negligible weight and without damping, is described by the
homogeneous undamped wave equation initial-boundary value problem:
a2 uxx = utt , 0 < x < L, t > 0,

u(0, t) = 0, and u(L, t) = 0,
u(x, 0) = f (x), and ut(x, 0) = g(x).
The general solution is
∞
anπ t anπ t  nπ x
n =1  L L  L .
The particular solution can be found by the formulas:
nπ x
L
2
An = ∫ f ( x ) sin dx , and
L0 L
nπ x
L
2
Bn =
anπ ∫ g ( x) sin
0
L
dx .
The solution waveform has a constant (horizontal) propagation speed,

in both directions of the x-axis, of a. The vibrating motion has a
(vertical) velocity given by ut(x, t) at any location 0 < x < L along the
string.
© 2008, 2012 Zachary S Tseng E-4 - 17

Exercises E-4.1:
1. Solve the vibrating string problem of the given initial conditions.
4 uxx = utt , 0 < x < π, t > 0,

u(0, t) = 0, u(π, t) = 0,
(a) u(x, 0) = 12sin(2x) − 16sin(5x) + 24sin(6x),

ut(x, 0) = 0.
(b) u(x, 0) = 0,
ut(x, 0) = 6.
(c) u(x, 0) = 0,
ut(x, 0) = 12sin(2x) − 16sin(5x) + 24sin(6x).
2. Solve the vibrating string problem.
100 uxx = utt , 0 < x < 2, t > 0,

u(0, t) = 0, and u(2, t) = 0,
u(x, 0) = 32sin(πx) + e2 sin(3πx) + 25sin(6πx),
ut(x, 0) = 6sin(2πx) − 16sin(5πx / 2).
3. Solve the vibrating string problem.
25 uxx = utt , 0 < x < 1, t > 0,

u(0, t) = 0, and u(2, t) = 0,
u(x, 0) = x – x2,
ut(x, 0) = π.
4. Verify that the D’Alembert solution, u(x, t) = [F(x − at) + F(x + at)] / 2,
where F(x) is an odd periodic function of period 2L such that F(x) = f (x) on
the interval 0 < x < L, indeed satisfies the given initial-boundary value
problem by checking that it satisfies the wave equation, boundary conditions,
and initial conditions.
© 2008, 2012 Zachary S Tseng E-4 - 18

a2 uxx = utt , 0 < x < L, t > 0,
u(0, t) = 0, u(L, t) = 0,
u(x, 0) = f (x), ut(x, 0) = 0.
5. Use the method of separation of variables to solve the following wave

equation problem where the string is rigid, but not fixed in place, at both
ends (i.e., it is inflexible at the endpoints such that the slope of displacement
curve is always zero at both ends, but the two ends of the string are allowed
to freely slide in the vertical direction).
a2 uxx = utt , 0 < x < L, t > 0,

ux(0, t) = 0, ux(L, t) = 0,
u(x, 0) = f (x), ut(x, 0) = g(x).
6. What is the steady-state displacement of the string in #5? What is

lim u ( x, t ) ? Are they the same?
t →∞
© 2008, 2012 Zachary S Tseng E-4 - 19

Answers E-4.1:
1. (a) u(x, t) = 12cos(4t) sin(2x) − 16cos(10t) sin(5x) + 24cos(12t) sin(6x).

(c) u(x, t) = 3sin(4t) sin(2x) – 1.6sin(10t) sin(5x) + 2sin(12t) sin(6x).
5. The general solution is

∞
 an π t an π t  nπ x
u ( x, t ) = A0 + B0 t + ∑  An cos + B n sin  cos .
n =1  L L  L
The particular solution can be found by the formulas:
nπ x
L L L
1 2 1
L ∫0 ∫ L ∫0
A0 = f ( x ) dx , An = f ( x) cos dx , B0 = g ( x) dx , and
L0 L
nπ x
L
2
Bn =
anπ ∫ g ( x) cos
0
L
dx .
6. The steady-state displacement is the constant term of the solution, A0.

The limit does not exist unless u(x, t) = C is a constant function, which
happens when f (x) = C and g(x) = 0, in which case the limit is C. They are
not the same otherwise.
© 2008, 2012 Zachary S Tseng E-4 - 20

The General Wave Equation
The most general form of the one-dimensional wave equation is:
a2 uxx + F(x, t) = utt + γ ut + k u.
Where a = the propagation speed of the wave,

γ = the damping constant
k = (external) restoration factor, such as when vibrations occur
in an elastic medium.
F(x, t) = arbitrary external forcing function (If F = 0 then the
equation is homogeneous, else it is nonhomogeneous.)
© 2008, 2012 Zachary S Tseng E-4 - 21

The Telegraph Equation
The most well-known example of (a homogeneous version of) the general

wave equation is the telegraph equation. It describes the voltage u(x, t)
inside a piece of telegraph / transmission wire, whose electrical properties
per unit length are: resistance R, inductance L, capacitance C, and
conductance of leakage current G:
a2 uxx = utt + γ ut + k u.
Where a2 = 1 / LC, γ = G / C + R / L, and k = GR / CL.
© 2008, 2012 Zachary S Tseng E-4 - 22

Example: The One-Dimensional Damped Wave Equation
a2 uxx = utt + γ ut , γ ≠ 0.
Suppose boundary conditions remain as the same (both ends fixed): (0, t) = 0,
and u(L, t) = 0.
The equation can be separated as follow. First rewrite it as:
a2X ″ T = X T ″ + γ X T ′,
2
Divide both sides by a X T , and insert a constant of separation:
X ′′ T ′′ + γ T ′
= 2
= −λ .
X a T
Rewrite it into 2 equations:
X ″ = −λ X → X ″ + λ X = 0,
T ″ + γ T ′ = − a2 λ T → T ″ + γ T ′ + a2 λ T = 0.
The boundary conditions also are separated, as usual:
u(0, t) = 0 → X(0)T(t) = 0 → X(0) = 0 or T(t) = 0

u(L, t) = 0 → X (L)T(t) = 0 → X(L) = 0 or T(t) = 0
As before, setting T(t) = 0 would result in the constant zero solution

only. Therefore, we must choose the two (nontrivial) conditions in
terms of x: X(0) = 0, and X(L) = 0.
© 2008, 2012 Zachary S Tseng E-4 - 23

After separation of variables, we have the system
X ″ + λX = 0, X(0) = 0 and X(L) = 0,
T ″ + γ T ′ + α2 λ T = 0 .
The next step is to find the eigenvalues and their corresponding

eigenfunctions of the boundary value problem
X ″ + λX = 0, X(0) = 0 and X(L) = 0.
This is a familiar problem that we have encountered more than once

previously. The eigenvalues and eigenfunctions are, recall,
n2 π 2
Eigenvalues: λ= 2 , n = 1, 2, 3, …
L
nπ x
Eigenfunctions: X n = sin
L , n = 1, 2, 3, …
The equation of t, however, has different kind of solutions depending on the

roots of its characteristic equation.
© 2008, 2012 Zachary S Tseng E-4 - 24

(Optional topic) Nonhomogeneous Undamped Wave Equation
Problems of partial differential equation that contains a nonzero forcing

function (which would make the equation itself a nonhomogeneous partial
differential equation) can sometimes be solved using the same idea that we
have used to handle nonhomogeneous boundary conditions – by considering
the solution in 2 parts, a steady-state part and a transient part. This is
possible when the forcing function is independent of time t, which then
could be used to determine the steady-state solution. The transient solution
would then satisfy a certain homogeneous equation. The 2 parts are thus
solved separately and their solutions are added together to give the final
result. Let us illustrate this idea with a simple example: when the string’s
weight is no longer “negligible”.
Example: A flexible string of length L has its two ends firmly secured.
Assume there is no damping. Suppose the string has a weight density of 1
Newton per meter. That is, it is subject to, uniformly across its length, a
constant force of F(x, t) = 1 unit per unit length due to its own weight.
Let u(x, t) be the vertical displacement of the string, 0 < x < L, and at any
time t > 0. It satisfies the nonhomogeneous one-dimensional undamped
wave equation:
a2 uxx + 1 = utt.
The usual boundary conditions u(0, t) = 0, and u(L, t) = 0, apply. Plus the
initial conditions u(x, 0) = f (x) and ut(x, 0) = g(x).
Since the forcing function is independent of time t, its effect is to

impart, permanently, a displacement on the string that depends only
on the location (the effect is subject to the boundary conditions, thus
might change with x). That is, the effect is to introduce a nonzero
© 2008, 2012 Zachary S Tseng E-4 - 25

steady-state displacement, v(x). Hence, we can rewrite the solution
u(x, t) as:
u(x, t) = v(x) + w(x, t).
By setting t to be a constant and rewrite the equation and the boundary

conditions to be dependent of x only, the steady-state solution v(x)
must satisfy:
a2 v″ + 1 = 0,
v(0) = 0, v(L) = 0.
Rewrite the equation as v″ = − 1 / a2, and integrate twice, we get
−1 2
v( x) = x + C1 x + C 2 .
2a 2
Apply the boundary conditions to find C1 = L / 2a2 and C2 = 0:
−1 2 L
v( x) = x + x.
2a 2 2a 2
Comment: Thus, the sag of a wire or cable due to its own weight can be
seen as a manifestation of the steady-solution of the wave equation. The sag
is also parabolic, rather than sinusoidal, as one might have reasonably
assumed, in nature.
We can then subtract out v(x) from the equation, boundary conditions,
and the initial conditions (try this as an exercise), the transient
solution w(x, t) must satisfy:
a2 wxx = wtt , 0 < x < L, t > 0,

w(0, t) = 0, w(L, t) = 0,
w(x, 0) = f (x) − v(x), wt(x, 0) = g(x).
© 2008, 2012 Zachary S Tseng E-4 - 26

The problem is now transformed to the homogeneous problem we
have already solved. The solution is just
 ∞
anπ t anπ t  nπ x
w( x, t ) = ∑  An cos + Bn sin  sin
n =1  L L  L .
Combining the steady-state and transient solutions, the general

solution is found to be
u ( x, t ) = v( x) + w( x, t )
−1 2 L ∞
 anπ t anπ t  nπ x
= 2 x + 2 x + ∑  An cos + Bn sin  sin
2a 2a n =1  L L  L
The coefficients can be calculated and the particular solution

determined by using the formulas:
nπ x
L
2
An = ∫ ( f ( x) − v( x) )sin dx , and
L0 L
nπ x
L
2
Bn =
anπ ∫
0
g ( x) sin
L
dx .
Note: Since the velocity ut(x, t) = vt(x) + wt(x, t) = 0 + wt(x, t) = wt(x, t). The
initial velocity does not need any adjustment, as ut(x, 0) = wt(x, 0) = g(x).
Comment: We can clearly see that, even though a nonzero steady-state

solution exists, the displacement of the string still will not converge to it as
t → ∞.
© 2008, 2012 Zachary S Tseng E-4 - 27

The Laplace Equation / Potential Equation
The last type of the second order linear partial differential equation in 2
independent variables is the two-dimensional Laplace equation, also called
the potential equation. Unlike the other equations we have seen, a solution
of the Laplace equation is always a steady-state (i.e. time-independent)
solution. Indeed, the variable t is not even present in the Laplace equation.
The Laplace equation describes systems that are in a state of equilibrium
whose behavior does not change with time. Some applications of the
Laplace equation are finding the potential function of an object acted upon
by a gravitational / electric / magnetic field, finding the steady-state
temperature distribution of the (2- or 3-dimensional) heat conduction
equation, and the steady-state flow of an ideal fluid (where the flow velocity
forms a vector field that has zero curl and zero divergence).
Since the time variable is not present in the Laplace equation, any problem
of the Laplace equation will not, therefore, have any initial condition. A
Laplace equation problem has only boundary conditions.
Let u(x, y) be the potential function at a point (x, y), then it is governed by
the two-dimensional Laplace equation
uxx + uyy = 0.
Any real-valued function having continuous first and second partial

derivatives that satisfies the two-dimensional Laplace equation is called a
harmonic function.
Similarly, suppose u(x, y, z) is the potential function at a point (x, y, z), then it
is governed by the three-dimensional Laplace equation
uxx + uyy + uzz = 0.
© 2008, 2012 Zachary S Tseng E-4 - 28

Comment: The one-dimensional Laplace equation is rather dull. It is merely
uxx = 0, where u is a function of x alone. It is not a partial differential
equation, but rather a simple integration problem of u″ = 0. (What is its
solution? Where have we seen it just very recently?)
The boundary conditions that accompany a 2-dimensional Laplace equation

describe the conditions on the boundary curve that encloses the 2-
dimensional region in question. While those accompany a 3-dimensional
Laplace equation describe the conditions on the boundary surface that
encloses the 3-dimensional spatial region in question.
© 2008, 2012 Zachary S Tseng E-4 - 29

The Relationships among Laplace, Heat, and Wave Equations
(Optional topic)
Now let us take a step back and see the bigger picture: how the
homogeneous heat conduction and wave equations are structured, and how
they are related to the Laplace equation of the same spatial dimension.
Suppose u(x, y) is a function of two variables, the expression uxx + uyy is

called the Laplacian of u. It is often denoted by
∇2 u = uxx + uyy.
Similarly, for a three-variable function u(x, y, z), the 3-dimensional Laplacian

is then
∇2 u = uxx + uyy + uzz.
(As we have just noted, in the one-variable case, the Laplaian of u(x),
degenerates into ∇2 u = u″.)
The homogeneous heat conduction equations of 1-, 2-, and 3- spatial

dimension can then be expressed in terms of the Laplacians as:
α2 ∇2 u = ut,
where α2 is the thermo diffusivity constant of the conducting material.

Thus, the homogeneous heat conduction equations of 1-, 2-, and 3-
dimension are, respectively,
α2 uxx = ut
α2 (uxx + uyy) = ut
α2 (uxx + uyy + uzz) = ut
© 2008, 2012 Zachary S Tseng E-4 - 30

As well, the homogeneous wave equations of 1-, 2-, and 3- spatial dimension
can then be similarly expressed in terms of the Laplacians as:
a2 ∇2 u = utt,
where the constant a is the propagation velocity of the wave motion. Thus,
the homogeneous wave equations of 1-, 2-, and 3-dimension are,
respectively,
a2 uxx = utt
a2 (uxx + uyy) = utt
a2 (uxx + uyy + uzz) = utt‡
Now let us consider the steady-state solutions of these heat conduction and
wave equations. In each case, the steady-state solution, being independent
of time, must have all zero as its partial derivatives with respect to t.
Therefore, in every instance, the steady-state solution can be found by
setting, respectively, ut or utt to zero in the heat conduction or the wave
equations and solve the resulting equation. That is, the steady-state solution
of a heat conduction equation satisfies
α2 ∇2 u = 0,
and the steady-state solution of a wave equation satisfies

a2 ∇2 u = 0.
‡
Even the electromagnetic waves are described by this equation. It can be easily shown by vector calculus
that any electric field E and magnetic field B satisfying the Maxwell’s Equations will also satisfy the 3-
dimensional wave equation, with propagation speed a = c ≈ 299792 km/s, the speed of light in vacuum.
© 2008, 2012 Zachary S Tseng E-4 - 31

2 2
In all cases, we can divide out the (always positive) coefficient α or a from
the equations, and obtain a “universal” equation:
∇2 u = 0.
This universal equation that all the steady-state solutions of heat conduction
and wave equations have to satisfy is the Laplace / potential equation!
Consequently, the 1-, 2-, and 3-dimensional Laplace equations are,

respectively,
uxx = 0,
uxx + uyy = 0,
uxx + uyy + uzz = 0.
Therefore, the Laplace equation, among other applications, is used to solve

the steady-state solution of the other two types of equations. And all
solutions of a Laplace equation are steady-state solutions. To answer the
earlier question, we have had seen and used the one-dimensional Laplace
equation (which, with only one independent variable, x, is a very simple
ordinary differential equation, u″ = 0, and is not a PDE) when we were
trying to find the steady-state solution of the one-dimensional homogeneous
heat conduction equation earlier.
© 2008, 2012 Zachary S Tseng E-4 - 32

Laplace Equation for a rectangular region
Consider a rectangular region of length a and width b. Suppose the top,

bottom, and left sides border free-space; while beyond the right side there
lies a source of heat/gravity/magnetic flux, whose strength is given by f (y).
The potential function at any point (x, y) within this rectangular region,
u(x, y), is then described by the boundary value problem:
(2-dim. Laplace eq.) uxx + uyy = 0, 0 < x < a, 0 < y < b,
(Boundary conditions) u(x, 0) = 0, and u(x, b) = 0,

u(0, y) = 0, and u(a, y) = f (y).
The separation of variables proceeds similarly. A slight difference here is

that Y(y) is used in the place of T(t). Let u(x, y) = X(x)Y(y) and substituting
uxx = X ″ Y and uyy = X Y ″ into the wave equation, it becomes
X ″ Y + X Y ″ = 0,
X ″ Y = −X Y ″.
Dividing both sides by X Y :
X ′′ Y ′′
=−
X Y
Now that the independent variables are separated to the two sides, we can
insert the constant of separation. Unlike the previous instances, it is more
convenient to denote the constant as positive λ instead.
X ′′ Y ′′
=− =λ
X Y
© 2008, 2012 Zachary S Tseng E-4 - 33

X ′′
=λ → X ″ = λX → X ″ − λX = 0,
X
Y ′′
− =λ → Y ″ = −λ Y → Y ″ + λ Y = 0.
Y
The boundary conditions also separate:
u(x, 0) = 0 → X(x)Y(0) = 0 → X(x) = 0 or Y(0) = 0

u(x, b) = 0 → X(x)Y(b) = 0 → X(x) = 0 or Y(b) = 0
u(0, y) = 0 → X(0)Y(y) = 0 → X(0) = 0 or Y(y) = 0
u(a, y) = f (y) → X(a)Y(y) = f (y) [cannot be simplified further]
As usual, in order to obtain nontrivial solutions, we need to ignore the

constant zero function in the solution sets above, and instead choose
Y(0) = 0, Y(b) = 0, and X(0) = 0 as the new boundary conditions. The
fourth boundary condition, however, cannot be simplified this way.
So we shall leave it as-is. (Don’t worry. It will play a useful role
later.) The result, after separation of variables, is the following
simultaneous system of ordinary differential equations, with a set of
boundary conditions:
X ″ − λX = 0, X(0) = 0,
Y ″ + λ Y = 0, Y(0) = 0 and Y(b) = 0.
Plus the fourth boundary condition, u(a, y) = f (y).
The next step is to solve the eigenvalue problem. Notice that there is
another slight difference. Namely that this time it is the equation of Y that
gives rise to the two-point boundary value problem which we need to solve.
© 2008, 2012 Zachary S Tseng E-4 - 34

Y ″ + λY = 0, Y(0) = 0, Y(b) = 0.
However, except for the fact that the variable is y and the function is Y,
rather than x and X, respectively, we have already seen this problem before
(more than once, as a matter of fact; here the constant L = b). The
eigenvalues of this problem are
n2 π 2
λ =σ = 2 ,
2
n = 1, 2, 3, …
b
Their corresponding eigenfunctions are
nπ y
Yn = sin , n = 1, 2, 3, …
b
Once we have found the eigenvalues, substitute λ into the equation of x. We

have the equation, together with one boundary condition:
n 2π 2
X ′′ − X =0, X(0) = 0.
b2
n 2π 2 nπ
Its characteristic equation, r −
2
= 0 , has real roots r = ± .
b2 b
Hence, the general solution for the equation of x is
nπ − nπ
x x
X = C1 e b
+ C2 e b
.
The single boundary condition gives
X(0) = 0 = C1 + C2 → C2 = −C1.
© 2008, 2012 Zachary S Tseng E-4 - 35

Therefore, for n = 1, 2, 3, …,
 nbπ x −nπ
x
X n = Cn  e − e b  .
 
Because of the identity for the hyperbolic sine function
eθ − e −θ
sinh θ = ,
2
the previous expression is often rewritten in terms of hyperbolic sine:
nπ x
X n = K n sinh , n = 1, 2, 3, …
b
The coefficients satisfy the relation: Kn = 2Cn.
Combining the solutions of the two equations, we get the set of solutions
that satisfies the two-dimensional Laplace equation, given the specified
boundary conditions:
nπ x nπ y
u n ( x, y ) = X n ( x) Yn ( y ) = K n sinh sin ,
b b
n = 1, 2, 3, …
The general solution, as usual, is just the linear combination of all the above,
linearly independent, functions un(x, y). That is,
∞
nπ x nπ y
u ( x, y ) = ∑ K n sinh sin
n =1 b b .
© 2008, 2012 Zachary S Tseng E-4 - 36

This solution, of course, is specific to the set of boundary conditions
u(x, 0) = 0, and u(x, b) = 0,
u(0, y) = 0, and u(a, y) = f (y).
To find the particular solution, we will use the fourth boundary condition,
namely, u(a, y) = f (y).
∞
anπ nπ y
u (a, y ) = ∑ K n sinh sin = f ( y)
n =1 b b
We have seen this story before. There is nothing really new here. The
summation above is a sine series whose Fourier sine coefficients are
bn = Kn sinh(anπ / b). Therefore, the above relation says that the last
boundary condition, f (y), must either be an odd periodic function (period =
2b), or it needs to be expanded into one. Once we have f (y) as a Fourier sine
series, the coefficients Kn of the particular solution can then be computed:
anπ nπ y
b
2
K n sinh = bn = ∫ f ( y ) sin dy
b b0 b
Therefore,
nπ y
b
bn 2
Kn =
anπ
=
anπ ∫ f ( y ) sin
b
dy
.
sinh b sinh 0
b b
© 2008, 2012 Zachary S Tseng E-4 - 37

(Optional topic) Laplace Equation in Polar Coordinates
The steady-state solution of the two-dimensional heat conduction or wave

equation within a circular region (the interior of a circular disc of radius k,
that is, on the region r < k) in polar coordinates, u(r, θ), is described by the
polar version of the two-dimensional Laplace equation
1 1
u rr + u r + 2 uθθ = 0 .
r r
The boundary condition, in this set-up, specifying the condition on the

circular boundary of the disc, i.e., on the curve r = k, is given in the form
u(k, θ) = f (θ), where f is a function defined on the interval [0, 2π). Note that
there is only one set of boundary condition, prescribed on a circle. This will
cause a slight complication. Furthermore, the nature of the coordinate
system implies that u and f must be periodic functions of θ, of period 2π.
Namely, u(r, θ) = u(r, θ + 2π), and f (θ) = f (θ + 2π).
By letting u(r, θ) = R(r)Θ(θ), the equation becomes
1 1
R′′Θ + R′Θ + 2 RΘ′′ = 0 .
r r
Which can then be separated to obtain
r 2 R′′ + rR′ Θ′′

=− =λ.
R Θ
This equation above can be rewritten into two ordinary differential equations:
r2 R″ + rR′ − λR = 0,
Θ″ + λΘ = 0.
© 2008, 2012 Zachary S Tseng E-4 - 38

The eigenvalues are not found by straight forward computation. Rather,
they are found by a little deductive reasoning. Based solely on the fact that
Θ must be a periodic function of period 2π, we can conclude that λ = 0 and λ
= n2, n = 1, 2, 3…, are the eigenvalues. The corresponding eigenfunctions
are Θ0 = 1 and Θn = An cos nθ + Bn sin nθ. The equation of r is an Euler
equation (the solution of which is outside of the scope of this course).
The general solution of the Laplace equation in polar coordinates is
A0 ∞
u (r , θ ) = + ∑ ( An cos nθ + Bn sin nθ ) r n .
2 n =1
Applying the boundary condition u(k, θ) = f (θ), we see that
A0 ∞
u (k , θ ) = ( )
+ ∑ An k n cos nθ + Bn k n sin nθ = f (θ ) .
2 n =1
Since f (θ) is a periodic function of period 2π, it would already have a

suitable Fourier series representation. Namely,
a0 ∞
f (θ ) = + ∑ (an cos nθ + bn sin nθ ) .
2 n =1
Hence, A0 = a0, An = an / kn, and Bn = bn / kn, n = 1, 2, 3…
For a problem on the unit circle, whose radius k = 1, the coefficients An and
Bn are exactly identical to, respectively, the Fourier coefficients an and bn of
the boundary condition f (θ).
© 2008, 2012 Zachary S Tseng E-4 - 39

(Optional topic) Undamped Wave Equation in Polar Coordinates
The vibrating motion of an elastic membrane that is circular in shape can be

described by the two-dimensional wave equation in polar coordinates:
urr + (1 / r) ur + (1 / r2) uθθ = a −2 utt.
The solution is u(r, θ, t), a function of 3 independent variables that describes

the vertical displacement of each point (r, θ) of the membrane at any time t.
© 2008, 2012 Zachary S Tseng E-4 - 40

Pde

Uploaded by

Copyright:

Available Formats

Pde

Uploaded by

Document Information

Original Description:

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Pde

Uploaded by

Copyright:

Available Formats

Second Order Linear Partial Differential Equations

Second linear partial differential equations; Separation of Variables; 2-

We are about to study a simple type of partial differential equations (PDEs):

α2 uxx = ut (one-dimensional heat conduction equation)

a2 uxx = utt (one-dimensional wave equation)

uxx + uyy = 0 (two-dimensional Laplace/potential equation)

In this class we will develop a method known as the method of Separation of

© 2008, 2012 Zachary S Tseng E-1 - 1

Consider the generic form of a second order linear partial differential

a uxx + b uxy + c uyy + d ux + e uy + f u = g(x,y).

If b2 – 4ac > 0, then the equation is called hyperbolic. The wave

If b2 – 4ac = 0, then the equation is called parabolic. The heat

If b2 – 4ac < 0, then the equation is called elliptic. The Laplace

In general, elliptic equations describe processes in equilibrium. While the

Example: Consider the one-dimensional damped wave equation

It can be rewritten as: 9uxx − utt − 6ut = 0. It has coefficients a = 9, b = 0,

© 2008, 2012 Zachary S Tseng E-1 - 2

Consider a thin bar of length L, of uniform cross-section and constructed of

© 2008, 2012 Zachary S Tseng E-1 - 3

In addition, there is an initial condition: the initial temperature distribution

Hence, what we have is a problem given by:

(Heat conduction eq.) α2 uxx = ut , 0 < x < L, t > 0,

(Boundary conditions) u(0, t) = 0, and u(L, t) = 0,

(Initial condition) u(x, 0) = f (x).

This is an example of what is known, formally, as an initial-boundary value

© 2008, 2012 Zachary S Tseng E-1 - 4

© 2008, 2012 Zachary S Tseng E-1 - 5

Start with the one-dimensional heat conduction equation α2 uxx = ut .

Hence, the heat conduction equation α2 uxx = ut can be rewritten as

© 2008, 2012 Zachary S Tseng E-1 - 6

© 2008, 2012 Zachary S Tseng E-1 - 7

The boundary conditions can be rewritten as:

u(0, t) = 0 → X(0)T(t) = 0 → X(0) = 0 or T(t) = 0

If we choose T(t) = 0, both conditions would be satisfied. However, it

Therefore, at the end of this process, we have two ordinary differential

X ″ + λX = 0, X(0) = 0 and X(L) = 0,

© 2008, 2012 Zachary S Tseng E-1 - 8

Let u(x, t) = X(x)T(t) and rewrite the equation in terms of X and T:

Divide both sides by X ″ T ″, we have separated the variables:

Finally, rewrite it into 2 equations:

© 2008, 2012 Zachary S Tseng E-1 - 9

Let u(x, t) = X(x)T(t) and rewrite the equation in terms of X and T:

Divide both sides by X ′ T ″, and insert a constant of separation:

The boundary conditions also must be separated:

u(0, t) = 0 → X(0)T(t) = 0 → X(0) = 0 or T(t) = 0

As before, setting T(t) = 0 would result in the constant zero solution

© 2008, 2012 Zachary S Tseng E-1 - 10

(1) X ″ + λX = 0, X(0) = 0 and X(L) = 0,

The next task is to solve this system of two simultaneous ordinary

A little background first: unlike an initial value problem of a second order

© 2008, 2012 Zachary S Tseng E-1 - 11

Note: Why are λ and X(t) called, respectively, eigenvalue and

where X is restricted to functions that satisfy the boundary conditions

© 2008, 2012 Zachary S Tseng E-1 - 12

X ″ + λX = 0, X(0) = 0 and X(L) = 0,

Case 1: If λ < 0 (−4λ > 0, distinct real roots of characteristic equation):

Let us denote λ = −σ 2, where σ = − λ > 0 . The characteristic

The second boundary condition gives

Hence, either C1 = 0 = C2 , (this would be the zero, or the trivial