7.

6: Method of Eigenfunction Expansions

We have seen that the use of eigenfunction expansions is another technique for finding solutions of differential equations. In this
section we will show how we can use eigenfunction expansions to find the solutions to nonhomogeneous partial differential
equations. In particular, we will apply this technique to solving nonhomogeneous versions of the heat and wave equations.

Nonhomogeneous Heat Equation

In this section we solve the one dimensional heat equation with a source using an eigenfunction expansion. Consider the problem
ut = kuxx + Q(x, t), 0 < x < L, t >0

u(0, t) = 0, u(L, t) = 0, t >0

u(x, 0) = f (x), 0 <x <L (7.6.1)

The homogeneous version of this problem is given by

vt = kvxx , 0 < x < L, t > 0,

v(0, t) = 0, v(L, t) = 0. (7.6.2)

We know that a separation of variables leads to the eigenvalue problem

′′
ϕ + λϕ = 0, ϕ(0) = 0, ϕ(L) = 0.

The eigenfunctions and eigenvalues are given by

nπx nπ 2

ϕn (x) = sin , λn = ( ) , n = 1, 2, 3, … .
L L

We can use these eigenfunctions to obtain a solution of the nonhomogeneous problem (7.6.1). We begin by assuming the solution
is given by the eigenfunction expansion
∞

u(x, t) = ∑ an (t)ϕn (x). (7.6.3)

n=1

In general, we assume that v(x, t) and ϕ (x) satisfy the same boundary conditions and that v(x, t) and v (x, t) are continuous
n x

functions.Note that the difference between this eigenfunction expansion and that in Section 4.3 is that the expansion coefficients
are functions of time.
In order to carry out the full process, we will also need to expand the initial profile, f (x), and the source term, Q(x, t), in the basis
of eigenfunctions. Thus, we assume the forms

f (x) = u(x, 0)
∞

= ∑ an (0)ϕn (x), (7.6.4)

n=1

Q(x, t) = ∑ qn (t)ϕn (x). (7.6.5)

n=1

Recalling from Chapter 4, the generalized Fourier coefficients are given by

L
⟨f , ϕn ⟩ 1
an (0) = = ∫ f (x)ϕn (x)dx, (7.6.6)
2 2
∥ ϕn ∥ ∥ ϕn ∥ 0

L
⟨Q, ϕn ⟩ 1
qn (t) = = ∫ Q(x, t)ϕn (x)dx. (7.6.7)
2 2
∥ ϕn ∥ ∥ ϕn ∥ 0

The next step is to insert the expansions (7.6.3) and (7.6.5) into the nonhomogeneous heat equation (7.6.1). We first note that

7.6.1 https://math.libretexts.org/@go/page/90962
 ∞

ut (x, t) = ∑ ȧn (t)ϕn (x),

n=1

uxx (x, t) = − ∑ an (t)λn ϕn (x). (7.6.8)

n=1

Inserting these expansions into the heat equation (7.6.1), we have

ut = kuxx + Q(x, t)
∞ ∞ ∞

∑ ȧn (t)ϕn (x) = −k ∑ an (t)λn ϕn (x) + ∑ qn (t)ϕn (x). (7.6.9)

n=1 n=1 n=1

Collecting like terms, we have

∞

∑ [ ȧn (t) + kλn an (t) − qn (t)] ϕn (x) = 0, ∀x ∈ [0, L].

n=1

Due to the linear independence of the eigenfunctions, we can conclude that

ȧn (t) + kλn an (t) = qn (t), n = 1, 2, 3, … .

This is a linear first order ordinary differential equation for the unknown expansion coefficients.
We further note that the initial condition can be used to specify the initial condition for this first order ODE. In particular,
∞

f (x) = ∑ an (0)ϕn (x).

n=1

The coefficients can be found as generalized Fourier coefficients in an expansion of f (x) in the basis . These are given by
ϕn (x)

Equation (7.6.6).
Recall from Appendix B that the solution of a first order ordinary differential equation of the form
′
y (t) + a(t)y(t) = p(t)

is found using the integrating factor

t

μ(t) = exp ∫ a(τ )dτ .

Multiplying the ODE by the integrating factor, one has

t t
d
[y(t) exp ∫ a(τ )dτ ] = p(t) exp ∫ a(τ )dτ .
dt

After integrating, the solution can be found providing the integral is doable.
For the current problem, we have

ȧn (t) + kλn an (t) = qn (t), n = 1, 2, 3, … .

Then, the integrating factor is

t
kλn t
μ(t) = exp ∫ kλn dτ = e .

Multiplying the differential equation by the integrating factor, we find

kλn t kλn t
[ ȧn (t) + kλn an (t)] e = qn (t)e

d
kλn t kλn t
(an (t)e ) = qn (t)e . (7.6.10)
dt

Integrating, we have

7.6.2 https://math.libretexts.org/@go/page/90962
 t
kλn t kλn τ
an (t)e − an (0) = ∫ qn (τ )e dτ ,
0

or
t
−kλn t −kλn (t−τ)
an (t) = an (0)e +∫ qn (τ )e dτ .
0

Using these coefficients, we can write out the general solution.

∞

u(x, t) = ∑ an (t)ϕn (x)

n=1

∞ t
−kλn t −kλn (t−τ)
= ∑ [an (0)e +∫ qn (τ )e dτ ] ϕn (x). (7.6.11)

n=1 0

We will apply this theory to a more specific problem which not only has a heat source but also has nonhomogeneous boundary
conditions.

 Example 7.6.1

Solve the following nonhomogeneous heat problem using eigenfunction expansions:

ut − uxx = x + t sin 3πx, 0 ≤ x ≤ 1, t >0

u(0, t) = 2, u(L, t) = t, t >0

u(x, 0) = 3 sin 2πx + 2(1 − x), 0 ≤x ≤1 (7.6.12)

Solution
This problem has the same nonhomogeneous boundary conditions as those in Example 7.3.2. Recall that we can define

u(x, t) = v(x, t) + 2 + (t − 2)x

to obtain a new problem for v(x, t). The new problem is

vt − vxx = t sin 3πx, 0 ≤ x ≤ 1, t > 0,

v(0, t) = 0, v(L, t) = 0, t > 0,

v(x, 0) = 3 sin 2πx, 0 ≤ x ≤ 1. (7.6.13)

We can now apply the method of eigenfunction expansions to find v(x, t) . The eigenfunctions satisfy the homogeneous
problem
′′
ϕn + λn ϕn = 0, ϕn (0) = 0, ϕn (1) = 0.

The solutions are

2
nπx nπ
ϕn (x) = sin , λn = ( ) , n = 1, 2, 3, …
L L

Now, let
∞

v(x, t) = ∑ an (t) sin nπx.

n=1

Inserting v(x, t) into the PDE, we have

∞
2 2
∑ [ ȧn (t) + n π an (t)] sin nπx = t sin 3πx.

n=1

Due to the linear independence of the eigenfunctions, we can equate the coefficients of the sin nπx terms. This gives

7.6.3 https://math.libretexts.org/@go/page/90962
 2 2
ȧn (t) + n π an (t) = 0, n ≠ 3,

2
ȧ3 (t) + 9 π a3 (t) = t, n = 3. (7.6.14)

This is a system of first order ordinary differential equations. The first set of equations are separable and are easily solved. For
n ≠ 3 , we seek solutions of

d
2 2
an = −n π an (t).
dt

These are given by

2 2
−n π t
an (t) = an (0)e , n ≠ 3.

In the case n = 3 , we seek solutions of

d
2
a3 + 9 π a3 (t) = t.
dt

The integrating factor for this first order equation is given by

2
9π t
μ(t) = e .

Multiplying the differential equation by the integrating factor, we have

d 2 2
9π t 9π t
(a3 (t)e ) = te .
dt

Integrating, we obtain the solution

t
2 2 2
−9 π t −9 π t 9π τ
a3 (t) = a3 (0)e +e ∫ τe dτ ,
0

−9 π
2
t −9 π
2
t
1 9π
2
τ
1 9π
2
τ
= a3 (0)e +e [ τe − e ] ,
2 2 2
9π (9 π )
0

−9 π
2
t
1 1 −9 π
2
τ
= a3 (0)e + t− [1 − e ]. (7.6.15)
2 2
9π (9 π 2 )

Up to this point, we have the solution

u(x, t) = v(x, t) + w(x, t)
∞

= ∑ an (t) sin nπx + 2 + (t − 2)x, (7.6.16)

n=1

where
2 2
−n π t
an (t) = an (0)e , n ≠3

2 1 1 2
−9 π t −9 π τ
a3 (t) = a3 (0)e + t− [1 − e ] (7.6.17)
2 2
9π (9 π ) 2

We still need to find an (0), n = 1, 2, 3, … .

The initial values of the expansion coefficients are found using the initial condition
∞

v(x, 0) = 3 sin 2πx = ∑ an (0) sin nπx.

n=1

It is clear that we have an (0) =0 for n ≠ 2 and a 2 (0) =3 . Thus, the series for v(x, t) has two nonvanishing coefficients,
2
−4 π t
a2 (t) = 3 e

1 1 2
−9 π τ
a3 (t) = t− [1 − e ]. (7.6.18)
2 2
9π (9 π )
2

7.6.4 https://math.libretexts.org/@go/page/90962
 Therefore, the final solution is given by
2
2 −9 π τ
9 π t − (1 − e )
2
−4 π t
u(x, t) = 2 + (t − 2)x + 3 e sin 2πx + sin 3πx.
4
81π

Forced Vibrating Membrane

We now consider the forced vibrating membrane. A two-dimensional membrane is stretched over some domain D . We assume
Dirichlet conditions on the boundary, u = 0 on ∂D . The forced membrane can be modeled as
2 2
utt = c ∇ u + Q(r, t), r ∈ D, t > 0,

u(r, t) = 0, r ∈ ∂D, t > 0,

u(r, 0) = f (r), ut (r, 0) = g(r), r ∈ D. (7.6.19)

The method of eigenfunction expansions relies on the use of eigenfunctions, ϕ (r) , for α ∈ J ⊂ Z α
2
a set of indices typically of
the form (i, j) in some lattice grid of integers. The eigenfunctions satisfy the eigenvalue equation
2
∇ ϕα (r) = −λα ϕα (r), ϕα (r) = 0, on ∂D.

We assume that the solution and forcing function can be expanded in the basis of eigenfunctions,

u(r, t) = ∑ aα (t)ϕα (r),

α∈J

Q(r, t) = ∑ qα (t)ϕα (r). (7.6.20)

α∈J

Inserting this form into the forced wave equation (7.6.19), we have
2 2
utt = c ∇ u + Q(r, t)

2
∑ äα (t)ϕα (r) = −c ∑ λα aα (t)ϕα (r) + ∑ qα (t)ϕα (r)

α∈J α∈J α∈J

2
0 = ∑ [ äα (t) + c λα aα (t) − qα (t)] ϕα (r). (7.6.21)

α∈J

The linear independence of the eigenfunctions then gives the ordinary differential equation
2
äα (t) + c λα aα (t) = qα (t).

We can solve this equation with initial conditions a α (0) and ȧ α (0) found from

f (r) = u(r, 0) = ∑ aα (0)ϕα (r),

α∈J

g(r) = ut (r, 0) = ∑ ȧα (0)ϕα (r). (7.6.22)

α∈J

 Example 7.6.2

Periodic Forcing, Q(r, t) = G(r) cos ωt .

Solution
It is enough to specify Q(r, t) in order to solve for the time dependence of the expansion coefficients. A simple example is the
case of periodic forcing, Q(r, t) = h(r) cos ωt . In this case, we expand Q in the basis of eigenfunctions,

Q(r, t) = ∑ qα (t)ϕα (r),

α∈J

G(r) cos ωt = ∑ γα cos ωtϕα (r). (7.6.23)

α∈J

7.6.5 https://math.libretexts.org/@go/page/90962
Inserting these expressions into the forced wave equation (7.6.19) , we obtain a system of differential equations for the
expansion coefficients,
2
äα (t) + c λα aα (t) = γα cos ωt.

In order to solve this equation we borrow the methods from a course on ordinary differential equations for solving
nonhomogeneous equations. In particular we can use the Method of Undetermined Coefficients as reviewed in Section B.3.1.
The solution of these equations are of the form

aα (t) = aαh (t) + aαp (t),

where a αh (t) satisfies the homogeneous equation,

2
äαh (t) + c λα aαh (t) = 0, (7.6.24)

and aαp (t) is a particular solution of the nonhomogeneous equation,

2
äαp (t) + c λα aαp (t) = γα cos ωt. (7.6.25)

The solution of the homogeneous problem (7.6.24) is easily founds as

aαh (t) = c1α cos(ω0α t) + c2α sin(ω0α t),

−−
where ω 0α = c √λα .
The particular solution is found by making the guess a αp (t) = Aα cos ωt . Inserting this guess into Equation (ceqn2), we have
2 2
[−ω + c λα ] Aα cos ωt = γα cos ωt

Solving for A , we obtain

α

γα
2 2
Aα = , ω ≠ c λα .
2 2
−ω + c λα

Then, the general solution is given by

γα
aα (t) = c1α cos(ω0α t) + c2α sin(ω0α t) + cos ωt,
2 2
−ω + c λα

−−
where ω 0α = c √λα and ω 2 2
≠ c λα

In the case where ω = c λ , we have a resonant solution. This is discussed in Section FO on forced oscillations. In this case
2 2
α

the Method of Undetermined Coefficients fails and we need the Modified Method of Undetermined Coefficients. This is
because the driving term, γ cos ωt, is a solution of the homogeneous problem. So, we make a different guess for the particular
α

solution. We let

aαp (t) = t (Aα cos ωt + Bα sin ωt) .

Then, the needed derivatives are

aαp (t) = ωt (−Aα sin ωt + Bα cos ωt) + Aα cos ωt + Bα sin ωt

2
aαp (t) = −ω t (Aα cos ωt + Bα sin ωt) − 2ωAα sin ωt + 2ωBα cos ωt

2
= −ω aαp (t) − 2ωAα sin ωt + 2ωBα cos ωt (7.6.26)

Inserting this guess into Equation (ceqn 2 ) and noting that ω 2

= c λα
2
, we have

−2ωAα sin ωt + 2ωBα cos ωt = γα cos ωt.

Therefore, A α =0 and
γα
Bα = .
2ω

So, the particular solution becomes

γα
aαp (t) = t sin ωt.
2ω

7.6.6 https://math.libretexts.org/@go/page/90962
 The full general solution is then
γα
aα (t) = c1α cos(ωt) + c2α sin(ωt) + t sin ωt,
2ω

−−
where ω = c√λ . α

We see from this result that the solution tends to grow as t gets large. This is what is called a resonance. Essentially, one is
driving the system at its natural frequency for one of the frequencies in the system. A typical plot of such a solution in Figure
7.6.1.

Figure 7.6.1 : Plot of a solution showing resonance.

This page titled 7.6: Method of Eigenfunction Expansions is shared under a CC BY-NC-SA 3.0 license and was authored, remixed, and/or curated
by Russell Herman via source content that was edited to the style and standards of the LibreTexts platform; a detailed edit history is available
upon request.

7.6.7 https://math.libretexts.org/@go/page/90962