PARTIAL DIFFERENTIAL EQUATIONS

I Introduction

An equation containing partial derivatives of a function of two or more independent

variables is called a partial differential equation (PDE).

∂u ∂u
e.g. y2 + =u where u(x, y) is the unknown function.
∂x ∂y

For convenience we denote

∂u ∂2u ∂2u
ux = , uxx = , uxy = , etc.
∂x ∂x2 ∂x∂y

so that the above PDE can be written y 2 ux + uy = u. The order of the PDE is the

order of the highest partial derivative. A PDE whose unknown function and its

partial derivatives appear linearly in the equation is said to be linear. For example,

A(x, y)uxx + B(x, y)uxy + C(x, y)uyy + D(x, y)ux + E(x, y)uy + F (x, y)u = G(x, y)

which is the general form of second-order linear PDE. If each term in the PDE contains

either the dependent variable or one of its derivatives, i.e., G(x, y) = 0, then it is said

to be homogeneous, otherwise it is nonhomogeneous.

2

Examples
1. ut = c2 uxx 1 dimensional heat conduction equation

2. utt = c2 uxx 1 dimensional wave equation

3. uxx + uyy = 0 2 dimensional Laplace equation

4. uxx + uyy = f (x, y) 2 dimensional Poisson equation

1
5. uxx + uyy = utt − λ2 u 2 dimensional Klein-Gordon equation
c2
6. utt = c2 (uxx + uyy + uzz ) 3 dimensional wave equation

In general, the solution of PDE presents a much more difficult problem than

the solution of ODE and except for certain special types of linear PDE, no general

method of solution is available. It is remarkable and fortunate that a large number

of the important equations in practice are not only linear, but also of second order,

for which solutions are relatively easy to find.

A solution of a PDE in some region R of the space of the independent variables

is a function that has all the partial derivatives appearing in the equation in some

domain containing R, and satisfies the equation everywhere in R.

In general, the totality of solutions of a PDE is very large.

Example. The functions

(i) u = x3 − 3xy 2 (ii) u = sin x cosh y

³y ´
(iii) u = ln(x2 + y 2 ) (iv) u = tan−1
x
are all solutions of the 2 dimensional Laplace equation uxx + uyy = 0, altough they

are entirely different from each other.

The Heat Equation (Diffusion equation) 3

Theorem

(i) If u1 and u2 are two solutions of a linear homogeneous PDE in some region,

then u = k1 u1 + k2 u2 is also a solution of the PDE in that region, where k1 and

k2 are constants.

(ii) u ≡ 0 is always a solution of a linear homogeneous PDE.

Proof Trivial.

II The Heat Equation (Diffusion equation)

Consider a thin metal rod with cross-section area A. Assume that the lateral surface

is wrapped with insulation such that heat conducts along the x-direction only.

∆x

x x0
L

Figure 1

Let u(x, t) denote the temperature at position x, at time t.

Newton’s law of cooling (Fourier’s law) =⇒ the quantity of heat (flux) flowing across

x0 per unit time is proportional to the temperature gradient at x0 .

Therefore Qx0 = −kAux (x0 , t) where k is the thermal conductivity of the metal.

The quantity of heat loss along a small segment [x0 , x0 + 4x] per unit time is given

by
H = Qx0 − Qx0 +4x = kA [ux (x0 + 4x, t) − ux (x0 , t)] .
4

The average change in temperature 4u in the time interval 4t is proportional to the

amount of heat loss and inversely proportional to the mass 4m of the element.
H4t
Therefore 4u = , where s is the specific heat of the metal. This gives
s4m

4u kA [ux (x0 + 4x, t) − ux (x0 , t)]

=
4t sρA4x
k [ux (x0 + 4x, t) − ux (x0 , t)] k
= −→ uxx (x0 , t) as 4x → 0.
sρ 4x sρ

4u
On the other hand, 4x → 0 =⇒ 4t → 0 =⇒ → ut (x0 , t), therefore
4t

k
ut (x, t) = uxx (x0 , t) =⇒ ut = c2 uxx
sρ
k
where c2 = is called the thermal diffusivity of the material (note that k, s, ρ are
sρ
all properties of the material and are all positive).

Note If heat is added or removed from the rod, then the equation becomes ut =

c2 uxx +f(x, t) which is nonhomogeneous and f(x, t) is the heat source or sink density.

Suppose that the ends x = 0 and x = L of the rod are kept at temperatures T0 and

T1 , and the initial temperature distribution of the rod is f (x), then the boundary

conditions (B.C.) are u(0, t) = T0 and u(L, t) = T1 , while the initial condition (I.C.)

is u(x, 0) = f(x).

The problem becomes an initial boundary value problem (IBVP):

ut = c2 uxx ; 0 < x < L, 0 < t < ∞,




 u(0, t) = T0 ,
B.C. 0 < t < ∞,


 u(L, t) = T1 ,

I.C. u(x, 0) = f(x), 0 < x < L.

The Heat Equation (Diffusion equation) 5

Separation of Variables

As the heat equation is linear and homogeneous, the approach is to seek solu-

tions of the differential equation and boundary conditions and then superposing them

to satisfy the initial conditions. First we solve the PDE ut = c2 uxx by a method

known as seperation of variables. Assume u(x, t) = X(x)T (t). Then

ut = XT 0 and uxx = X 00 T,
T0 X 00
hence ut = c2 uxx =⇒ XT 0 = c2 X 00 T =⇒ = .
c2 T X

As the L.H.S. is a function of t only and the R.H.S. is a function of x only,

T0 X 00
= = −λ, where λ is some constant.
c2 T X

Hence 


 T 0 + λc2 T = 0 =⇒
2t
T = e−λc


 X 00 + λX = 0

and there are three cases for the solution of X:

(i) λ = 0 =⇒ X 00 = 0 =⇒ X = α + βx.

(ii) λ < 0 (let λ = −k2 ) =⇒ X 00 − k2 X = 0 =⇒ X = αekx + βe−kx .

(iii) λ > 0 (let λ = k2 ) =⇒ X 00 + k2 X = 0 =⇒ X = α cos kx + β sin kx.

Therefore,




 α + βx if λ = 0,



u= 2 2
e−k c t (αekx + βe−kx ) if λ = −k2 < 0,






 e−k2 c2 t (α cos kx + β sin kx) if λ = k2 > 0.
6

Note that α, β and λ (or k) are arbitrary in the above solution, whose values

will be determined by the boundary and initial conditions, which depend on the type

of such conditions. Consider the following homogeneous boundary conditions:




 a1 u(0, t) + a2 ux (0, t) = 0, (a1 , a2 ) 6= (0, 0)
B.C. ; t > 0,


 a3 u(L, t) + a4 ux (L, t) = 0, (a3 , a4 ) 6= (0, 0)

and the initial condition: I.C. u(x, 0) = f (x), 0 ≤ x ≤ L.

A PDE together with the initial and boundary conditions is called an IBV P

(Initial Boundary Value Problem).

We now discuss how to find the solution to the IBVP for the heat equation

with different cases of (a1 , a2 , a3 , a4 ).

Case I ut = c2 uxx ; 0 < x < L, 0 < t < ∞.

  


 u(0, t) = 0,  (a1 , a2 ) = (1, 0) 
B.C. 0 < t < ∞, i.e., 
  

 u(L, t) = 0, (a3 , a4 ) = (1, 0)

I.C. u(x, 0) = f(x), 0 < x < L.

(i) λ = 0 =⇒ u = α + βx

u(0, t) = u(L, t) = 0 =⇒ α = β = 0 =⇒ u ≡ 0 (trivial solution).

2 c2 t
(ii) λ < 0 =⇒ u = e−k (αekx + βe−kx ) (λ = −k2 ),

u(0, t) = 0 =⇒ α + β = 0, u(L, t) = 0 =⇒ αekx + βe−kx = 0.

Therefore α = β = 0 =⇒ u ≡ 0 (trivial solution)

2 c2 t
(iii) λ > 0 =⇒ u = e−k (α cos kx + β sin kx) (λ = k2 ),

u(0, t) = 0 =⇒ α = 0, u(L, t) = 0 =⇒ β sin kL = 0.

The Heat Equation (Diffusion equation) 7

nπ
To get non-trivial solution, we must keep β 6= 0, hence sin kL = 0 =⇒ k = =⇒
L
³ nπ ´2
λ = k2 = , n = 1, 2, · · · . Thus for each n = 1, 2, · · · ,
L

2 c2 t nπx
un = β n e−(nπ/L) sin
L

is a solution of the PDE which satisfies the B.C.

³ nπ ´2
Note. λn = are called the eigenvalues and the corresponding un the
L
eigenfunctions of the IBVP problem. Hence an eigenfunction is a solution to the

PDE which satisfies the boundary conditions, but not the initial condition in general.

In order to satisfy the initial condition, we add all un together (known as the

principle of superposition):

∞
X ∞
X 2 c2 t nπx
u(x, t) = un = β n e−(nπ/L) sin
n=1 n=1
L

and then determine β n so that the initial condition is satisfied.

I.C. u(x, 0) = f(x) =⇒

∞
X nπx
f (x) = β n sin , 0 < x < L.
n=1
L

ZL
2 nπx
Using Fourier sine series expansion, we get β n = f(x) sin dx.
L L
0
Therefore the solution to the IBVP is given by

∞
X ZL
−(nπ/L)2 c2 t nπx 2 nπx
u(x, t) = β ne sin with β n = f(x) sin dx.
n=1
L L L
0

(Note : lim u(x, t) = 0 when B.C. are u(0, t) = u(L, t) = 0.)

t→∞
8

Case II ut = c2 uxx; 0 < x < L, 0 < t < ∞.  



 ux (0, t) = 0,  (a1 , a2 ) = (0, 1) 
B.C. 0 < t < ∞, i.e., 
  

 ux (L, t) = 0, (a3 , a4 ) = (0, 1)

I.C. u(x, 0) = f(x), 0 ≤ x ≤ L.

Note that as u = XT =⇒ ux = X 0 T, the B.C. become X 0 (0)T (t) = 0 and

X 0 (L)T (t) = 0, t > 0, which lead to X 0 (0) = X 0 (L) = 0 if we want nontrivial

solutions. Now again,

λ = 0 =⇒ u = α + βx. Hence

ux (0, t) = 0 =⇒ X 0 (0) = 0 =⇒ β = 0 =⇒ u ≡ α.

λ < 0 =⇒ X = αekx + βe−kx and so X 0 (x) = αkekx − βke−kx .

2 c2 t
Thus u = ek (αekx + βe−kx ) and

X 0 (0) = X 0 (L) = 0 =⇒ α = β = 0 =⇒ X(x) = 0 =⇒ u ≡ 0.

λ > 0 =⇒ X = α cos kx + β sin kx and X 0 = −αk sin kx + βk cos kx =⇒

2 c2 t
u = e−k (α cos kx + β sin kx).

Hence X 0 (0) = 0 =⇒ βk = 0 =⇒ β = 0 =⇒ X 0 (L) = −αL sin kL = 0. To

nπ
get nontrivial solutions we must take k = , n = 1, 2, · · · . Therefore X =
L
nπx
α cos , n = 1, 2, · · · . It follows that
L

nπx −(nπ/L)2 c2 t
u0 (x, t) ≡ α0 , un (x, t) = αn cos e , n = 1, 2, · · · .
L

Applying the principle of superposition,

∞
X ∞
X 2 c2 t nπx
u(x, t) = un = α0 + αn e−(nπ/L) cos .
n=0 n=1
L
The Heat Equation (Diffusion equation) 9

The initial condition u(x, 0) = f(x) then yields

∞
X nπx
u(x, 0) = f(x) = α0 + αn cos , 0 ≤ x ≤ L,
n=1
L

and so by half-range cosine series we obtain

ZL ZL
1 2 nπx
α0 = f (x)dx, αn = f (x) cos dx n = 1, 2, · · · .
L L L
0 0

Thus the solution to the IBVP is

∞
X 2 c2 t nπx
u(x, t) = α0 + αn e−(nπ/L) cos .
n=1
L

with above α0 and αn .

ZL
1
Note : lim u(x, t) = α0 = f(x)dx which is the average initial temperature.
t→∞ L
0

Case III ut = c2 uxx ; 0 < x < 1, 0 < t < ∞.

  


 u(0, t) = 0,  (a1 , a2 ) = (1, 0) 
B.C. 0 < t < ∞,  
  

 u(1, t) + ux (1, t) = 0; (a3 , a4 ) = (1, 1)

I.C. u(x, 0) = f(x), 0 ≤ x ≤ 1.

u = X(x)T (t), u(0, t) = 0 =⇒ X(0) = 0,

ux = X 0 (x)T (t), u(1, t) + ux (1, t) = 0 =⇒ X(1) + X 0 (1) = 0.

λ = 0 =⇒ X = α + βx

X(0) = 0 =⇒ α = 0 =⇒ X(x) = βx =⇒ X 0 (x) = β

X(1) + X 0 (1) = 0 =⇒ β + β = 0 =⇒ β = 0

=⇒ X ≡ 0 =⇒ u ≡ 0 (trivial solution).
10

λ < 0 =⇒ X = αekx + βe−kx =⇒ X 0 = αkekx − βke−kx

X(0) = 0 =⇒ α + β = 0 =⇒ α = −β.

X(1) + X 0 (1) = 0

=⇒ (αek + βe−k ) + k(αek − βe−k ) = β(−(1 + k)ek + (1 − k)e−k ) = 0

=⇒ β = 0, α = −β = 0 =⇒ X = 0 =⇒ u ≡ 0 (trivial solution).

2 c2 t
λ > 0 =⇒ X = α cos kx + β sin kx and u = e−k (α cos kx + β sin kx).

Hence X(0) = 0 =⇒ α = 0 =⇒ X = β sin kx =⇒ X 0 = βk cos kx,

X(1) + X 0 (1) = 0 =⇒ β sin k + kβ cos k = 0 =⇒ k = − tan k (if β 6= 0).

It can be shown that tan k = −k has infinitely many roots 0 < k1 < k2 < · · · < ∞,
2 2
hence X(x) = β sin kx =⇒ un = β n e−kn c t sin kn x, n = 1, .2, · · · . It follows that

∞
X 2 2
u(x, t) = β n e−kn c t sin kn x
n=1
∞
X
u(x, 0) = f(x) =⇒ f(x) = β n sin kn x, 0 < x < 1.
n=1

Z1
To find β n , use the fact that sin kn x sin km xdx = 0, m 6= n, to obtain
0

Z1 Z1 ÃX
∞
!
f(x) sin km xdx = β n sin kn x sin km xdx
0 0 n=1

∞
X Z1 Z1
= βn sin kn x sin km xdx = β m sin2 km xdx.
n=1 0 0

R1
f(x) sin km xdx
0
Hence β m = , m = 1, 2, · · · .
R1 2
sin km xdx
0

Other cases for (a1 , a2 , a3 , a4 ) can be dealt with in the similar way.
The Heat Equation (Diffusion equation) 11

conduction equation ut = c2 uxx ; 0 < x < L, 0 < t < ∞ with

Summary: Heat


 a1 u(0, t) + a2 ux (0, t) = 0, (a1 , a2 ) 6= (0, 0)
B.C. ; t > 0,


 a3 u(L, t) + a4 ux (L, t) = 0, (a3 , a4 ) 6= (0, 0)

I.C. u(x, 0) = f(x), 0 < x < L.

By means of separation of variables, i.e. let u(x, t) = X(x)T (t), we get

T 0 (t) + λc2 T (t) = 0  (*)



 a1 X(0) + a2 X 0 (0) = 0,
and X 00 (x)+λX(x) = 0 with (**)


 a3 X(L) + a4 X 0 (L) = 0.

(**) is referred to as the Sturm-Liouville Problem (S-L) associated with the PDE

and B.C. To solve the S-L problem, those λ which gives non-trivial solution of X are

called the eigenvalues and the corresponding solutions X(x) are the eigenfunctions

of the S-L problem. The solution to the IBVP has the form

∞
X 2
u(x, t) = β n e−λn c t Xn (x)
n=1

where λn and Xn are the eigenvalues and eigenfunctions of the S-L problem.

The following theorem about the eigenfunctions of the S-L problem is useful:

Sturm-Liouville Theorem (simplified)

If φ1 (x) and φ2 (x) are two eigenfunctions of the Sturm-Liouville problem




 a1 y(0) + a2 y 0 (0) = 0
y 00 + λy = 0, 0 < x < L, with


 a3 y(L) + a4 y 0 (L) = 0

corresponding to eigenvalues λ1 and λ2 respectively and if λ1 6= λ2 , then φ1 (x) and

ZL
φ2 (x) are orthogonal in the sense that φ1 (x)φ2 (x)dx = 0.
0
12

III Non-homogeneous boundary Conditions

1. ut = c2 uxx ; 0 < x < L, 0 < t < ∞ with

B.C. u(0, t) = T1 , u(L, t) = T2 , 0 < t < ∞, (T1 , T2 are constant)

I.C. u(x, 0) = f(x), 0 < x < L.

As the boundary conditions are not homogeneous, the method of separation

of variables cannot be used. Let u(x, t) = v(x) + w(x, t) where v(x) is time-

independent representing the “steady state” solution and w(x, t) represents the

“transient” solution. Then

ut = wt and uxx = vxx + wxx .

PDE becomes wt = c2 (vxx + wxx ).




u(0, t) = v(0) + w(0, t) = T1 
B.C. becomes t > 0,

u(L, t) = v(L) + w(L, t) = T2 


I.C. becomes u(x, 0) = v(x) + w(x, 0) = f(x), 0 < x < L.

x
Take v(x) = (T2 − T1 ) + T1 . Then
L

vxx = 0 with v(0) = T1 and v(L) = T2 .

The problem then becomes

wt = c2 wxx ; 0 < x < L, 0 < t < ∞.

B.C. w(0, t) = 0, w(L, t) = 0, 0 < t < ∞,

I.C. w(x, 0) = f (x) − v(x), 0 < x < L.

which has homogeneous boundary conditions and so can be solved as before.

Non-homogeneous boundary Conditions 13

2. ut = c2 uxx ; 0 < x < L, 0 < t < ∞ with

B.C. u(0, t) = T1 (t), u(L, t) = T2 (t), 0 < t < ∞,

I.C. u(x, 0) = f(x), 0 < x < L.

Here T1 (t) and T2 (t) are functions of t. Hence we consider

u(x, t) = v(x, t) + w(x, t).

Then ut = vt + wt and uxx = vxx + wxx .

PDE becomes (vt + wt ) = c2 (vxx + wxx ).



u(0, t) = v(0, t) + w(0, t) = T1 (t) 

B.C. becomes t > 0,

u(L, t) = v(L, t) + w(L, t) = T2 (t) 


I.C. becomes u(x, 0) = v(x, 0) + w(x, 0) = f(x), 0 < x < L.

Therefore w(0, t) = T1 (t) − v(0, t), w(L, t) = T2 (t) − v(L, t),

w(x, 0) = f (x) − v(x, 0).

Again, in order to obtain homogeneous boundary condition, put vxx = 0 such

that v(0, t) = T1 (t) and v(L, t) = T2 (t), giving

x
v(x, t) = T1 (t) + (T2 (t) − T1 (t))
L

x 0
and vt (x, t) = T 01 (t) + (T2 (t) − T10 (t)).
L
h x 0 i
2 2
0 0
Hence, wt = c wxx − vt = c wxx − T 1 (t) + (T 2 (t) − T 1 (t))
L

with B.C. w(0, t) = 0, w(L, t) = 0, 0 < t < ∞,

I.C. w(x, 0) = f(x) − v(x, 0), 0 < x < L,

where the boundary conditions are homogeneous, but the equation is now non-

homogeneous.
14

IV Non-homogeneous Heat Equation (Eigenfunction expansion)

ut = c2 uxx +g(x, t); 0 < x < L, 0 < t < ∞.



 a1 u(0, t) + a2 ux (0, t) = 0
B.C. ; t > 0,


 a3 u(L, t) + a4 ux (L, t) = 0

I.C. u(x, 0) = f(x), 0 < x < L.

Consider the corresponding homogeneous problem ut = c2 uxx with the same

∞
P 2
B.C. and I.C. as above. Its solution is of the form β n e−λn c t Xn (x), where λn and
n=1

Xn (x) are eigenvalues and eigenfunctions of the associated S-L problem respectively.
∞
P
We now seek a solution of the form u(x, t) = Tn (t)Xn (x) for the non-homogeneous
n=1
2
equation by choosing appropriate Tn (t) (note that Tn (t) = β n e−λn c t when g = 0, but

it has to be adjusted when g 6= 0).

∞
P ∞
P
Let g(x, t) = gn (t)Xn (x) and substitute u(x, t) = Tn (t)Xn (x) into the
n=1 n=1

non-homogeneous equation we get

∞
X ∞
X ∞
X
2
Tn0 (t)Xn (x) =c Tn (t)Xn00 (x) + gn (t)Xn (x).
n=1 n=1 n=1

which is satisfied if

Tn0 (t)Xn (x) = c2 Tn (t)Xn00 (x) + gn (t)Xn (x), n = 1, 2, · · · .

However, since Xn satisfies the S-L problem, i.e. Xn00 (x) + λXn (x) = 0, the above

equations become

Tn0 (t)Xn (x) = −c2 λn Tn (t)Xn (x) + gn (t)Xn (x)

=⇒ Tn0 (t) + c2 λn Tn (t) = gn (t).

Non-homogeneous Heat Equation (Eigenfunction expansion) 15

The last equation is a first-order linear ODE and its solution is given by
 
 Zt 
−λn c2 t λn c2 p
Tn (t) = e T (0) + gn (p)e dp .
 n 
0

It remains to determine Tn (0) and gn (t), and to satisfy the I.C. To do so,
RL
note that Xm (x) and Xn (x) are orthogonal for m 6= n (i.e., Xm (x)Xn (x)dx = 0) by
0

the Sturm-Liouville theorem as they are solutions of the S-L problem. Hence

ZL ZL "X
∞
#
g(x, t)Xm (x)dx = gn (t)Xn (x) Xm (x)dx
0 0 n=1

∞
X ZL ZL
2
= gn (t) Xm (x)Xn (x)dx = gm (x) Xm (x)dx.
n=1 0 0

Therefore.
RL
g(x, t)Xn (x)dx
0
gn (t) = .
RL
Xn2 (x)dx
0

Furthermore, the I.C. =⇒

∞
X
u(x, 0) = Tn (0)Xn (x) = f (x).
n=1

By similar manipulation as above we obtain

RL
f(x)Xn (x)dx
0
Tn (0) = .
RL
Xn2 (x)dx
0

Finally, the solution of the original non-homogeneous PDE is given by

∞
X
u(x, t) = Tn (t)Xn (x).
n=1
16

Example: PDE : ut = c2 uxx + sin 3πx, 0 < x < 1, t > 0.

B.C. u(0, t) = u(1, t) = 0, t > 0,

I.C. u(x, 0) = sin πx, 0 < x < 1.

First we look for λn and Xn (x), which are the eigenvalues and eigenfunctions

of the homogeneous equation

ut = c2 uxx , 0 < x < 1, t > 0 with B.C. u(0, t) = u(1, t) = 0, t > 0.

Hence λn = n2 π2 and Xn (x) = sin nπx, n = 1, 2, 3, · · · .

Next, calculate
RL R1 
g(x, t)Xn (x)dx sin 3πx sin nπxdx 

 1, n = 3
0 0
gn (t) = = =
RL R1 

Xn2 (x)dx 2
sin nπxdx  0, n 6= 3
0 0

and 
RL R1 
f (x)Xn (x)dx sin πx sin nπxdx 
 1, n = 1,
0 0
Tn (0) = = =
RL R1 

Xn2 (x)dx 2
sin nπxdx  0, n 6= 1.
0 0

From these we obtain

 
 Zt 
−λn c2 t λn c2 p
Tn (t) = e T (0) + gn (p)e dp
 n 
0

as follows:
2 2 c2 t
n = 1, g1 (t) = 0, T1 (0) = 1 =⇒ T1 (t) = e−λ1 c t = e−π ,

n = 2, g2 (t) = 0, T2 (0) = 0 =⇒ T2 (t) = 0,

n = 3, g3 (t) = 1, T3 (0) = 0 =⇒
Zt Zt " 2 2
#t
c2 t c2 p −9π 2 c2 t 9π 2 c2 p −9π 2 c2 t e9π c p
T3 (t) = e−λ3 eλ3 dp = e e dp = e
9π 2 c2
0 0 0
−9π 2 c2 t 9π 2 c2 t
e
(e − 1) 1 2 2
= 2 2
= 2 2 (1 − e−9π c t ),
9π c 9π c
n ≥ 4 =⇒ gn (t) = Tn (0) = 0 =⇒ Tn (t) = 0.
More Heat Conduction Equations 17

Finally, we obtain the solution as

u(x, t) = T1 (t)X1 (x) + T3 (t)X3 (x)

2 c2 t 1 2 2
= e−π sin πx + 2 2
(1 − e−9π c t ) sin 3πx.
9π c

V More Heat Conduction Equations

ut = c2 uxx − αu; 0 < x < 1, 0 < t < ∞.

B.C. u(0, t) = u(1, t) = 0; t > 0,

I.C. u(x, 0) = f(x), 0 ≤ x ≤ 1,

where −αu represents heat flow across the lateral boundary.

As diffusion along the x direction is represented by c2 uxx , the term −αu would

affect the solution in t only.

Let u(x, t) = φ(t)w(x, t). Then

ut = φwt + φt w and uxx = φwxx .

Therefore PDE =⇒ φwt + φt w = c2 φwxx − αφw
µ ¶
2 φt
=⇒ wt = c wxx − α + w.
φ
Let φt /φ = −α. Then φt + αφ = 0 =⇒ φ = e−αt and the PDE becomes

wt = c2 wxx

with B.C. w(0, t) = w(1, t) = 0

I.C. w(x, 0) = f (x).

Thus w(x, t) may be obtained as previously and u(x, t) = e−αt w(x, t).
18

VI The Wave Equation

Consider a perfectly flexible string of uniform density ρ stretched to a uniform tension

T between two end points x = 0 and x = L.

Suppose that the string is distorted and then at a certain instant, t = 0 say,

it is released and allowed to vibrate, the problem is to determine the vibration of the

string or to find its deflection u(x, t) at any point x and at each instant time t > 0.

β
α T2
u B
A
T1

x
O x0 x0 + ∆x L

Figure 2

Assumptions : (i) the string offers no resistance to bending so that the tension

is tangential to the string at each point.

(ii) gravitational force on the string is negligible compared to the

tension on the string.

(iii) the motion of the string is only in a vertical plane.

Let T1 and T2 be the tensions at the points A and B respectively. As there is

no motion in the horizontal direction, the horizontal components of the tension must

be constant, i.e. T1 cos α = T2 cos β = T0 (constant). Vertically, the resultant of these

The Wave Equation 19

forces leads to the vertical vibration of the string and

Newton’s 2nd law =⇒ T2 sin β − T1 sin α = ρδxutt

(ρδx is the mass of the string between A and B).

It follows that

T2 sin β T1 sin α ρδx

tan β − tan α = − = utt .
T2 cos β T1 cos α T0

As tan β = ux (B, t) = ux (x + δx, t) and tan α = ux (A, t) = ux (x, t), we get

1 ρ
[ux (x + δx, t) − ux (x, t)] = utt .
δx T0

ρ
Letting δx → 0 =⇒ uxx = utt =⇒ utt = c2 uxx (c2 = T0 /ρ), which is called
T0
the wave equation.

IBVP for Wave equation

PDE: utt = c2 uxx ; 0 < x < L, t > 0.

B.C. u(0, t) = u(L, t) = 0, t > 0,



 u(x, 0) = f(x)
I.C. , 0 < x < L.


 ut (x, 0) = g(x)

By the method of separation of variables, take u(x, t) = X(x)T (t). Then

utt = c2 uxx =⇒ X(x)T 00 (t) = c2 X 00 (x)T (t) =⇒

T 00 (t) X 00 (x)
= = −λ (constant) =⇒
c2 T (t) X(x) 

X 00 (x) + λX(x) = 0  
(i) associated Sturm-Liouville problem


X(0) = X(L) = 0 

(ii) T 00 (t) + c2 λT (t) = 0.

20

n2 π 2 nπx
(i) =⇒ λn = ; Xn (x) = sin , n = 1, 2, · · ·
L2 L
nπc nπc
(ii) =⇒ Tn (t) = αn cos t + β n sin t.
L L

Therefore the solution is given by

∞ ³
P nπc nπc ´ nπx
u(x, t) = αn cos t + β n sin t sin ,
n=1 L L L
ZL
2 nπx
u(x, 0) = f (x) =⇒ αn = f(x) sin dx,
L L
0
ZL
2 nπx
ut (x, 0) = g(x) =⇒ βn = g(x) sin dx.
nπc L
0

nπc nπc h nπc i p

Note that αn cos t +β n sin t = γ n cos (t + φn ) (where γ n = α2n + β 2n
L L L
L αn
and φn = arctan ). Hence the solution may also be expressed as
nπc βn
∞
X h nπc i nπx
u(x, t) = γ n cos (t + φn ) sin .
n=1
L L

VII Classification of 2nd order linear PDE in 2 independent variables

Consider the 2nd order linear PDE

A(x, y)uxx + B(x, y)uxy + C(x, y)uyy + D(x, y)ux + E(x, y)ux + F (x, y)u = G(x, y) (∗)

If B 2 − 4AC > 0 for (x, y) ∈ Ω ⊂ R2 , (∗) is said to be hyperbolic in Ω.

If B 2 − 4AC < 0 for (x, y) ∈ Ω ⊂ R2 , (∗) is said to be elliptic in Ω.

If B 2 − 4AC = 0 for (x, y) ∈ Ω ⊂ R2 , (∗) is said to be parabolic in Ω.

Example utt = c2 uxx =⇒ c2 uxx − utt = 0. Let y = t. Then A = c2 , B = 0,

C = −1. Hence B 2 − 4AC = 4c2 > 0. Thus the wave equation is hyperbolic.

For ut = c2 uxx , A = c2 , B = C = 0, so B 2 − 4AC = 0. Hence the heat equation is

parabolic.
Laplace Equations 21

For uxx + uyy = 0, A = 1, B = 0, C = 1, so B 2 − 4AC = −4 < 0. Hence the

Laplace equation is elliptic.

VIII Laplace Equations

Consider ∇2 u = uxx + uyy = 0.

Note that solutions of Laplace equation are called harmonic functions. Particular

solutions are :
(i) u(x, y) = a + bx + cy, (ii) u(x, y) = xy,

(iii) u(x, y) = x2 − y 2 , (iv) u(x, y) = ex cos y,

x
(v) u(x, y) = .
x2 + y2
2-dimensional polar coordinates: x = r cos θ, y = r sin θ =⇒

1 1
∇2 u = urr + ur + 2 uθθ = 0.
r r

3-dimensional cylindrical coordinates: x = ρ cos φ, y = ρ sin φ, z = z =⇒

1 1
∇2 u = uxx + uyy + uzz = uρρ + uρ + 2 uφφ + uzz = 0.
r ρ

3-dimensional spherical coordinates:

x = r sin θ cos φ, y = r sin θ sin φ, z = r cos θ =⇒

2 1 cot θ 1
∇2 u = urr + ur + 2 uθθ + 2 uθ + 2 2 uφφ = 0.
r r r r sin θ

Simple solutions of Laplace equation

1. Consider the 2-D Laplace equation with cylindrical symmetry (e.g. steady state

temperature distribution in a infinitely long homogeneous cylindrical metal

pipe).
22

1 1
∇2 u = urr + ur + 2 uθθ = 0.
r r

Cylindrical symmetry =⇒ uθ = 0 so that uθθ = 0. Therefore

· ¸
2 d2 u 1 du d2 u du d du
∇u = + = 0 =⇒ r 2 + =0 =⇒ r =0
dr2 r dr dr dr dr dr
du dr
=⇒ r = k (constant) =⇒ du = k
dr r
=⇒ u = A ln r + B, r 6= 0.

This gives the solution to ∇2 u = 0 with cylindrical symmetry.

2. Consider the 3-D Laplace equation with spherical symmetry, i.e. uθ = uφ = 0.

d2 u 2 du d2 u du
∇2 u = 0 =⇒ + =0 =⇒ r2 + 2r =0
dr2 r dr dr2 dr

· ¸
d 2 du du
=⇒ r = 0 =⇒ r2 = k (constant)
dr dr dr
A
=⇒ u = + B, r 6= 0.
r

Laplace equation in a rectangle region

∇2 u = 0 in Ω = {(x, y) : 0 < x < a, 0 < y < b}

u(0, y) = u(a, y) = 0; 0 < y < b,




u(x, 0) = 0 
; 0 < x < a.


u(x, b) = f(x) 

Solution By the method of separation of variables, let u(x, y) = X(x)Y (y)

X 00 (x) Y 00 (y)
=⇒ =− = −λ (constant)
X(x) Y (y)

Therefore X 00 (x) + λX(x) = 0 (*)

and Y 00 (y) − λY (y) = 0 (**)

Laplace Equations 23

Boundary conditions =⇒ X(0) = X(a) = 0 and Y (0) = 0.

(*) is the associated Sturm-Liouville problem with eigenvalues and eigenfunctions

as ³ nπx ´
n2 π 2
λn = , Xn = sin , n = 1, 2, · · ·
a2 a
³ nπy ´
(**) =⇒ Yn (y) = αn sinh .
a
³ nπx ´ ³ nπy ´
Therefore un (x, y) = Xn (x)Yn (y) = αn sin sinh and so
a a
∞
X ∞
X ³ nπx ´ ³ nπy ´
u(x, y) = un (x, y) = αn sin sinh .
n=1 n=1
a a

It remains to find αn . Use the nonhomogeneous boundary condition on u we get

∞
X ³ nπx ´ µ ¶
nπb
f(x) = u(x, b) = αn sin sinh .
n=1
a a

Za
2 nπx
Therefore, αn = ¡ ¢ f (x) sin dx
a sinh nπb
a
a
0
Theorem on the principle of superposition

2
 ∇ u = 0 in Ω = {(x, y) : 0 < x < a, 0 < y < b}
The Dirichlet problem


 u(0, y) = f1 (y); u(a, y) = f2 (y); 0 < y < b
with boundary conditions


 u(x, 0) = f3 (x); u(x, b) = f4 (x); 0 < x < a

has the solution u = u1 + u2 + u3 + u4 where u1 , u2 , u3 , u4 satisfy ∇2 u = 0 and

(i) u1 (0, y) = f1 (y), 0 < y < b; u1 = 0 on other boundaries,

(ii) u2 (a, y) = f2 (y), 0 < y < b; u2 = 0 on other boundaries,

(iii) u3 (x, 0) = f3 (x), 0 < x < a; u3 = 0 on other boundaries,

(iv) u4 (x, b) = f4 (x), 0 < x < a; u4 = 0 on other boundaries.

Proof Exercise.
24

Laplace equation in circular regions

∇2 u = 0 in Ω = {(x, y) : x2 + y 2 < A},

u(x, y) = f(x, y) on the of boundary of .Ω.

1 1
In polar coordinates, the equation becomes ∇2 u = urr + ur + 2 uθθ = 0,
r r
and the boundary condition becomes u(A, θ) = f(θ), 0 < θ < 2π.

Note that u(r, θ) is periodic in θ of period 2π and is bounded as r → 0.

Let u(r, θ) = R(r)Θ(θ) and substitute it into the equation we obtain

1 1 R00 R0 Θ00
R00 Θ + R0 Θ + 2 RΘ00 = 0 =⇒ r2 +r =− = −λ (constant)
r r R R Θ

which leads to (i) Θ00 + λΘ = 0, Θ(θ) is periodic with period 2π.

(ii) r2 R00 + rR0 − λR = 0, limR(r) < ∞.

r→0
For (i), λ = 0 =⇒ Θ(θ) = α + βθ and Θ(θ) is periodic =⇒ β = 0 =⇒ Θ(θ) = α.

λ < 0 (λ = −k2 ) =⇒ Θ = αekθ + βe−kθ and

Θ(θ) is periodic =⇒ α = β = 0 =⇒ Θ(θ) ≡ 0 (trivial solution).

λ > 0 (λ = k2 ) =⇒ Θ = α cos kθ + β sin kθ and

Θ(θ) is periodic with period 2π =⇒ k = n = 1, 2, · · · .

Therefore we get a sequence of solutions for Θ(θ) as

Θ0 (θ) = α0 , Θn (θ) = αn cos nθ + β n sin nθ; n = 1, 2, · · · .

For (ii), λ = 0 =⇒ r2 R00 + rR0 = 0 =⇒ rR00 + R0 = 0 =⇒ (rR0 )0 = 0

=⇒ rR0 = c1 =⇒ R = c2 + c1 ln r.
However, limR(r) < ∞ =⇒ c1 = 0 =⇒ R(r) = c2 .
r→0

λ = k 2 = n2 =⇒ r2 R00 + rR0 − n2 R = 0 (Euler equation)

Let R = rm and by substitution into Euler equation we get m = ±n.

Thus R(r) = c3 rn + c4 r−n and limR(r) < ∞ =⇒ c4 = 0 =⇒ R(r) = Rn (r) = rn .

r→0
Laplace Equations 25

It follows that un (r, θ) = Θn (θ)Rn (r) = rn (αn cos nθ + β n sin nθ); n = 0, 1, 2, · · · .

By the principle of superposition,

∞
X
u(r, θ) = α0 + rn (αn cos nθ + β n sin nθ)
n=1

where αn and β n are determined from the boundary conditions

∞
X
u(A, θ) = f (θ) =⇒ f (θ) = α0 + An (αn cos nθ + β n sin nθ),
n=1

Z 2π
1
which leads to α0 = f(θ)dθ,
2π 0
Z 2π
1
αn = f(θ) cos nθdθ (n = 1, 2, ...),
πAn 0
Z 2π
1
βn = f (θ) sin nθdθ (n = 1, 2, ...).
πAn 0

*******************************

Reference Books:
Kreyszig E., Advanced Engineering Mathematics, Wiley International Edition.

Farlow S.J., Partial Differential Equations for Scientists and Engineers, Wiley Inter-

national Edition.

Boyce W.E. and DiPrima R.C., Elementary Differential Equations and Boundary

Value Problems, Wiley International Edition.

Derrick W.R. and Grossman S.I., Elementary Differential Equations with Boundary

Value Problems, Addison Wesley.