THE METHOD OF SEPARATION OF VARIABLES

To solve the BVPs that we have encountered so far, we will use separation of
variables on the homogeneous part of the BVP. This separation of variables leads
to problems for ordinary dierential equations (some with endpoints conditions).
The ODE problems are much easier to solve. (At this time you need to refresh
your knowledge about linear dierential equations that you have learned in the
rst course of dierential equations: MAP2302).
To understand and appreciate the technique of separation of variables, we are
going to solve in details some BVPs dealing with heat, wave , and Laplace equations. The nonhomogeneous terms of the problems below are cooked up so as to
understand the situation. For more general nonhomogeneous terms, we have to
wait until after Fourier series are studied.
The separation of variables is based on the following obvious observation,
Observation. Let x and y be independent variables varying in intervals I, J R.
Suppose that f and g are functions of x and y, respectively. If
f (x) = g(y)

x I,

y J,

then both functions are equal to the same constant. That is, there is a constant
C R such that
f (x) = g(y) = C
x I, y J.

Example 1.
Consider the BVP modeling heat propagation in a rod of length L with both
ends kept at constant temperature 0 degrees and the initial temperature of the rod
is given by the function
3x
7x
f (x) = 100 sin
50 sin
.
L
L
If u(x, t) denotes the temperature at time t of point x, then u satises the BVP

0 < x < L, t > 0 ;

ut (x, t) = kuxx (x, t)
u(0, t) = 0, u(L, t) = 0
t>0;
(1)

u(x, 0) = f (x)
0<x<L.
where k is the thermal diusivity of the rod. Now we describe the method of
separation for this BVP.
Step 1. Separate the BVP into its homogeneous part (HP) and nonhomogeneous
part (NHP). In this problem the HP is:
{
ut (x, t) = kuxx (x, t)
0 < x < L, t > 0 ;
(2)
u(0, t) = 0, u(L, t) = 0
t>0.
and the NHP is
(3)

u(x, 0) = f (x)

Date: January 28, 2014.

1

0<x<L.

THE METHOD OF SEPARATION OF VARIABLES

Note that the homogeneous part HP consists of all equations that are satised by
the zero function. So in our BVP, u(x, t) 0 solves the rst three equations but
not the fourth. Hence, the rst three equations form HP and the fourth is NHP.
Step 2. Separate HP into ODE problems. Assume that HP has a nontrivial 1
solution of the form
u(x, t) = X(x)T (t)
with X a function of x alone and T a function of t alone. Such a function has
separated variables. We rewrite the HP for such a function u. It becomes
{
X(x)T (t) = kX (x)T (t)
0 < x < L, t > 0 ;
(4)
X(0)T (t) = 0, X(L)T (t) = 0
t>0.
At all points where X(x)T (t) = 0 the PDE can be written as
T (t)
X (x)
=
.
kT (t)
X(x)
The functions T /kT and X /X depend on the independent variables t and x and
they are equal. Thus the functions are equal to a constant:
T (t)
X (x)
=
= ,
kT (t)
X(x)
with a constant (called the separation constant). The minus sign in front of is
just for latter convenience. It follows that X and T solve the ODEs
X (x) + X(x) = 0,

0 < x < L,

and T (t) + kT (t) = 0,

t > 02 .

The boundary conditions in (4) imply that X(0) = 0 and X(L) = 0 (since T 0).
We have therefore split HP into two ODE problems. The X-problem:
{
X + X = 0
0<x<L;
(5)
X(0) = X(L) = 0 ,
and the T -problem:
(6)

T + kT = 0 ,

t>0.

The X-problem, which consists of a second order ode and two endpoints conditions,
is an example of a Sturm-Liouville problem. The values of for which the SLproblem has nontrivial solutions are called the eigenvalues of the problem and the
corresponding nontrivial solutions are the eigenfunctions.
Step 3. Find the eigenvalues and eigenfunctions of the SL-problem. The characteristic equation of the ODE in the SL-problem is

m2 + = 0 with roots m = .
Depending on the sign of , we distinguish three cases:
Case : < 0. We write = 2 with > 0. The characteristic roots are
m = . The general solution of the ODE in the X-problem is
X(x) = Aex + Bex ,
1By nontrivial solutions we mean the solutions that are not identically zero.
2To be more rigorous, we have these odes in the regions where XT = 0 but then it follows

from uniqueness that these equations are valid on the intervals where X and T are defined.

THE METHOD OF SEPARATION OF VARIABLES

with A and B constants. We need to nd A and B so that X satises the endpoints

conditions:
X(0) = 0

A+B =0
X(L) = 0

AeL + BeL = 0
The above linear system for A and B has the unique solution A = B = 0. The
reason is the following. From the rst equation we have B = A and then the
second equation becomes
A(eL eL ) = 0 .
Now L > 0 and > 0 imply that eL eL > 0. Hence the equation for A holds
only when A = 0.
With A = 0 and B = 0, the function X is zero X(x) = 0 for 0 x L. We
have proved that if < 0, the only solution for the X-problem is X 0. This
means that < 0 cannot be an eigenvalue of the SL-problem.
Case : = 0. The characteristic roots are m = 0 (with multiplicity 2). The
general solution of the ODE of the X-problem is
X(x) = A + Bx ,
with A, B constants. The endpoints conditions X(0) = X(L) = 0 imply that A
and B satisfy the system
A = 0 and

A + BL = 0

whose only solution is A = 0, B = 0. The corresponding solution of the X-problem

is again X 0. Therefore, = 0 is not an eigenvalue.
Case : > 0. We write = 2 with > 0. The characteristic roots are
m = i. The general solution of the ODE in the X-problem is
X(x) = A cos (x) + B sin (x) ,
with A and B constants. The endpoint conditions implies
A cos 0 + B sin 0 = 0,

A cos(L) + B sin(L) = 0 .

The rst equation gives A = 0 and the second reduces to

B sin(L) = 0 .
Note that if B = 0, then again X 0 is the trivial solution. Thus in order to have
a nontrivial solution X, we must have
sin(L) = 0 .
Therefore, must be of the form
n
, with n = 1, 2, 3, . . .
=
L
In this case
nx
X(x) = B sin
L
is a nontrivial solution of the X-problem. This solution is obtained when
( n )2
=
,
L
which is therefore an eigenvalue of the SL-problem.

THE METHOD OF SEPARATION OF VARIABLES

Summary of the above discussion. The SL-problem has innitely many eigenvalues. For each n Z+ we have:
( n )2
nx
,
eigenfunction : Xn (x) = sin
eigenvalue : n =
.
L
L
Note that any nonzero multiple of Xn is also an eigenfunction. We need only write
a generator of the eigenspace.
Step 4. For each eigenvalue, solve the corresponding T -problem. For = n , the
T -problem is:
( n )2
T (t) + k
T (t) = 0 .
L
The general solution is
T (t) = Cek(

n
L

with C constant.

Step 5. Use steps 3 and 4 to write the solutions with separated variables for the
Homogeneous problem HP: For each n Z+ , we have:
eigenvalue: n = (n/L)2 ;
eigenfunction: Xn (x) = sin(nx/L);
T -solution: Tn (t) = exp(k(n/L)2 t);
a corresponding solution of HP with separated variable is:
n 2
nx
un (x, t) = Tn (t)Xn (x) = ek( L ) t sin
.
L
Step 6. Use the principle of superposition to obtain more solutions of HP. Any
linear combination of the solutions un obtained in step 5 is again a solution of
problem (2). That is, for c1 , , cN constants the function
u(x, t) = c1 u1 (x, t) + + cN uN (x, t) =

cj ek( L ) t sin

j=1

jx
L

is also a solution of (2). In fact, under appropriate conditions, we can form a linear
combination of innitely many solutions un and obtain again a solution of (2) (see
Principle of Superposition 1).
Step 7. Find the coecients cj so that the solution of HP found in step 6 satises
also the nonhomogeneous condition (3). This means, we need to nd c1 , c2 ,
so that
3x
7x
u(x, 0) = f (x) = 100 sin
50 sin
.
L
L
We have,
2x
N x
3x
7x
x
+ c2 sin
+ + cN sin
= 100 sin
50 sin
L
L
L
L
L
We take all coecients cj to be zero except c3 = 100 and c7 = 50.
u(x, 0) = c1 sin

Conclusion: The solution of the original BVP (1) is:

u(x, t) = 100ek(3/L) t sin
2

2
3x
7x
50ek(7/L) t sin
.
L
L

THE METHOD OF SEPARATION OF VARIABLES

The Exponential and the Trigonometric Functions

Recall that if x R, then
eix = cos x + i sin x and eix = cos x i sin x .
The trigonometric functions cos and sin can be expressed as:
cos x =

eix + eix
2

and

sin x =

eix eix
.
2i

For x, y R, we have ex+iy = ex (cos y + i sin y).

The hyperbolic trigonometric functions cosh (cosine hyperbolic) and sinh (sine
hyperbolic) are dened by
cosh x =

ex + ex
2

and

sinh x =

ex ex
.
2

We also have
ex = cosh x + sinh x and ex = cosh x sinh x .
The following identities of cosh and sinh are left for you to verify as an exercise.
cosh(x)
sinh(x)

1
0

Figure 1. Graphs of cosh and sinh

cosh x = sinh x ; sinh x = cosh x;

cosh x = cosh x ; sinh x = sinh x.
cosh(x + y) = cosh x cosh y + sinh x sinh y;
sinh(x + y) = sinh x cosh y + cosh x sinh y;
cosh(x y) = cosh x cosh y sinh x sinh y;
sinh(x y) = sinh x cosh y cosh x sinh y;
cosh2 x sinh2 y = 1;

THE METHOD OF SEPARATION OF VARIABLES

Example 2.
The following BVP models wave propagation with damping in a string of length
L.
(7)

utt (x, t) + 2aut (x, t) = c2 uxx (x, t)

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

u(x, 0) = f (x), ut (x, 0) = g(x)

0 < x < L, t > 0 ;

t>0;
0<x<L.

The waves speed is c and the damping constant is a. We assume that 0 a < c/L.
This problem models the vibrations in a string of length L with end points attached.
The initial position and initial velocity are given by the functions f (x) and g(x).
Take for instance,
x
6x
x
13x
(8)
f (x) = 20 sin
sin
, and g(x) = 3 sin
+ sin
.
L
L
L
L
To solve this problem, we proceed with steps of Example 1.
Step 1. Separate the BVP into its homogeneous part (HP) and nonhomogeneous
part (NHP).
Homogeneous part (HP):
{
utt (x, t) + 2aut (x, t) = c2 uxx (x, t)
0 < x < L, t > 0 ;
(9)
u(0, t) = 0, u(L, t) = 0
t>0.
Nonhomogeneous part (NHP)
(10)

u(x, 0) = f (x), ut (x, 0) = g(x) 0 < x < L .

Step 2. Separate HP into ODE problems. Assume that u(x, t) = X(x)T (t) is a
nontrivial solution of (HP) We rewrite the HP for such a function u. It becomes
{
X(x)T (t) + 2aX(x)T (t) = c2 X (x)T (t)
0 < x < L, t > 0 ;
(11)
X(0)T (t) = 0, X(L)T (t) = 0
t>0.
By separating the variables, the PDE can be written as
T (t)
2aT (t)
X (x)
+
=
= (constant).
c2 T (t)
c2 T (t)
X(x)
It follows that X and T solve the ODEs
X (x) + X(x) = 0 ,

and T (t) + 2aT (t) + c2 T (t) = 0 .

The boundary conditions imply that X(0) = 0 and X(L) = 0. Problem (11) splits
then into two ODE problems. The X-problem (SL-problem):
{
X + X = 0
0<x<L;
(12)
X(0) = X(L) = 0 ,
and the T -problem:
(13)

T + 2aT + c2 T = 0 ,

t>0.

Step 3. Find the eigenvalues and eigenfunctions of the SL-problem. The Xproblem was solved in Example 1 and we found that it has innitely many eigenvalues. For each n Z+ we have:
( n )2
,
eigenvalue :
n = n2 =
L
nx
.
eigenfunction : Xn (x) = sin n x = sin
L

THE METHOD OF SEPARATION OF VARIABLES

Step 4. For each eigenvalue, solve the corresponding T -problem. For = n , the
T -problem is:
T (t) + 2aT (t) + c2 n2 T (t) = 0 .
The characteristic equation of this ODE is
m2 + 2am + (cn )2 = 0
with roots
m = a

a2 (cn )2 .

Note that since a < c/L, then a2 (cn )2 < 0 for every n Z+ . We can then
write a2 (cn )2 = n2 with n > 0. Thus, the characteristic roots are complex
conjugate and can be written as
m = a in .
Two independent solutions of the T -problem are:
Tn1 (t) = eat cos n t and Tn2 (t) = eat sin n t .
Step 5. Use steps 3 and 4 to write the solutions with separated variables for the
Homogeneous problem (11). For each n Z+ , we have:
eigenvalue: n = n2 = (n/L)2 ;
eigenfunction: Xn (x) = sin(n x) = sin(nx/L);
T -solutions: Tn1 (t) = eat cos n t and Tn2 (t) = eat sin n t
Corresponding solutions of HP with separated variable are:
u1n (x, t) = Tn1 (t)Xn (x) = eat cos(n t) sin(n x) ;
u2n (x, t) = Tn2 (t)Xn (x) = eat sin(n t) sin(n x) .
Step 6. Use the principle of superposition to obtain more solutions of HP. Any
linear combination of the solutions u1n and u2n obtained in step 5 is again a solution
of problem (9). That is, for A1 , , AN and B1 , , BN constants the function
u(x, t)

Aj u1j (x, t) + Bj u2j (x, t)

j=1

eat [Aj cos(j t) + Bj sin(j t)] sin(j x) .

j=1

is also a solution of problem (9).

Step 7. Find the coecients Aj and Bj so that the solution of HP found in step
6 satises also the nonhomogeneous conditions (10). This means, we need to nd
A1 , B1 , A2 , B2 , so that
x
6x
u(x, 0) = f (x) = 20 sin
(14)
sin
;
L
L
13x
x
(15)
+ sin
.
ut (x, 0) = g(x) = 3 sin
L
L
We need to write ut (x, 0). We have the following
ut (x, t) =

j=1

eat [(Bj j aAj ) cos(j t) (Aj j + aBj ) sin(j t)] sin(j x)

THE METHOD OF SEPARATION OF VARIABLES

The initial position and initial conditions (14) and (15) became
u(x, 0) =

Aj sin

j=1
N

jx
x
6x
= 20 sin
sin
L
L
L

(Bj j aAj ) sin

ut (x, 0) =

j=1

jx
x
13x
= 3 sin
+ sin
.
L
L
L

By equating the coecients, the rst equation gives

A1 = 20, A6 = 1, Aj = 0 if j = 1 and j = 6 .
The second equation gives
B1 1 aA1 = 3, B13 13 aA13 = 1, Bj j aAj = 0 if j = 1, 13 .
Thus,
B1 =

1
a
3 + 20a
, B13 =
, B6 =
, and Bj = 0, j = 1, 6, 13.
1
13
6

Conclusion: The solution of the original BVP (7) is:

[
]
3 + 20a
at
u(x, t) = e
20 cos(1 t) +
sin(1 t) sin(1 x)+
1
]
[
+eat cos(6 t) a6 sin(6 t) sin(6 x) + 113 eat sin(13 t) sin(13 x) .
Example 3.
Consider the BVP given in polar coordinates by

1 < r < 2 , 0 2 ;
u(r, ) = 0 ,
u(1, ) = cos ,
0 2 ;
(16)

ur (2, ) = sin 2; ,
0 2 .
This BVP can be interpreted as that for the steady-state temperature in an annular

u=0

u=cos

ur=sin 2

Figure 2. BVP with mixed boundary conditions

region where on the outer boundary the heat ux is prescribed while on the inner
boundary the temperature is prescribed. Recall that u = 0 in polar coordinates
means
1
1
urr + ur + 2 u = 0
r
r

THE METHOD OF SEPARATION OF VARIABLES

Remark The polar coordinates (r, ) and (r, + 2) represent the same point in
the plane. So any function u dened in the plane is 2-periodic in : u(r, ) =
u(r, + 2). Hence, although it is not explicitly listed in the BVP, it is understood
that u(r, 0) = u(r, 2) and also u (r, 0) = u (r, 2).
Now, we proceed as in the previous problems.
Step 1. Separate the BVP into its homogeneous part (HP) and nonhomogeneous
part (NHP).
Homogeneous part (HP):
1
1
(17)
urr (r, ) + ur (r, ) + 2 u = 0 1 < r < 2, 0 2 .
r
r
Nonhomogeneous part (NHP)
(18)

u(1, ) = cos , ur (2, ) = sin 2,

0 2 .

Step 2. Separate HP into ODE problems. Assume that u(r, ) = R(r)() is a

nontrivial solution of (HP). We rewrite HP for such a function u. It becomes
1
1
(19)
R (r)() + R (r)() + 2 R(r) () = 0
r
r
By separating the variables, the PDE can be written as
r2 R (r) rR (r)
()
=

=
()
R(r)
R(r)
It follows that and R solve the ODEs
() + () = 0 ,

(constant).

and r2 R (r) + rR (r) R(r) = 0 .

Remember that u and also u must be 2-periodic in . Thus the functions ()

and () must be 2-periodic:
(0) = (2),

(0) = (2)

We have then the following ODE problems. The -problem (periodic SL-problem):
{
+ = 0 0 2 ;
(20)
(0) = (2) and (0) = (2) .
and the R-problem:
(21)

r2 R (r) + rR (r) R(r) = 0 ,

1<r<2.

Step 3. Find the eigenvalues and eigenfunctions of the periodic SL-problem. The
characteristic equation of the ODE in the SL-problem is

m2 + = 0 with roots m = .
Depending on the sign of , we distinguish three cases:
Case : < 0. We write = 2 with > 0. The characteristic roots are
m = . The general solution of the ODE in the -problem is
() = A cosh(x) + B sinh(x) ,
with A and B constants. We need to nd A and B so that satises the periodic
conditions:
(0) = (2)

A = A cosh(2) + B sinh(2)
(0) = (2)

B = A sinh(2) + B cosh(2)

10

THE METHOD OF SEPARATION OF VARIABLES

The above linear system for A and B can be written as

(cosh(2) 1)A + sinh(2)B
sinh(2)A + (cosh(2) 1)B

=0
=0

The determinant of this linear system is

(cosh(2) 1)2 sinh2 (2) = 2(1 cosh(2)) = 0
(because > 0 and cosh(x) = 1 only when x = 0). Since the determinant of the
linear system is not zero, the unique solution is A = B = 0. Hence 0. Thus
< 0 is not an eigenvalue.
Case : = 0. The characteristic roots are m = 0 (with multiplicity 2). The
general solution of the ODE of the -problem is
() = A + B ,
with A, B constants. The periodic conditions are
A = A + 2B

and

B = B.

We get B = 0 and A arbitrary. Therefore = 0 is an eigenvalue with eigenfunction

1.
Case : > 0. We write = 2 with > 0. The characteristic roots are
m = i. The general solution of the ODE in the -problem is
() = A cos () + B sin () ,
with A and B constants. The periodic conditions give
A = A cos(2) + B sin(2),

B = A sin(2) + B cos(2) .

We rewrite this system in standard form

(1 cos(2))A sin(2)B
sin(2)A + (1 cos(2))B

=0
=0

The determinant of this system is 2(1 cos(2)). In order to have nontrivial

solutions, it is necessary to have the determinant equal to 0. That is,
cos(2) = 1 .
For this to happen, we need = 1, 2, 3, . When = n N, any values of A
and B satisfy the system. Hence, for = n2 , problem (20) has nontrivial solutions:
cos(n) and sin(n).
Summary: The eigenvalues and eigenfunctions of the -problem are
Eigenvalues
0 = 0
n = n2 , (n Z+ )

Eigenfunctions
0 () = 1
1n () = cos(n) and 2n () = sin(n)

Step 4. For each eigenvalue, solve the corresponding R-problem. Note that the
ode for R is a Cauchy-Euler equation. We can nd solutions of the form rm . The
characteristic equation is
m(m 1) + m = 0

m2 = 0 .

For = 0, we have m = 0 with multiplicity 2 and the general solution of the ode is
R(r) = A + B ln r .

THE METHOD OF SEPARATION OF VARIABLES

11

For = n = n2 , the characteristic roots are m = n and the general solution of

the ode is
B
R(r) = Arn + n
r
Step 5. Use steps 3 and 4 to write the solutions with separated variables for the
Homogeneous problem (19).
For = 0, we have
u0 (r, ) = (A0 + B0 ln r)
2
+
For = n , with n Z , we have
(
)
(
)
Bn1
Bn2
1
1 n
2
2 n
un (r, ) = An r + n cos(n), un (r, ) = An r + n sin(n)
r
r
Step 6. Any linear combination of the solutions obtained in step 5:
u(x, t) = u0 (r, ) +

u1n (r, ) + u2n (r, )

n=1

is also a solution of problem (19).

Step 7. Now we seek the coecients A1j , A2j , Bj1 and Bj2 so that the solution u of
step 6 satises the nonhomogeneous conditions u(1, ) = cos and ur (2, ) = sin 2.
The rst condition gives (check this claim)
A0 = 0, A2n + Bn2 = 0, n = 1, , N
A11 + B11 = 1, A1n + Bn1 = 0, n = 2, , N
The second condition gives
B0
Bn1
= 0, A1n 2n1 n+1
= 0, n = 1, , N
2
2
2
B
Bn2
2(A22 2 32 ) = 1, A2n 2n1 n+1
= 0, n = 1, 3, 4, , N
2
2
It gives
A0 = B0 = 0, A1n = Bn1 = 0 for n = 1, A2j = Bj2 = 0 for j = 2
and

1
4
2
2
, B11 = , A22 =
, B22 =
5
5
17
17
The solution of the original problem (16) is
(
)
(
)
1
4
2
1
u(r, ) =
r+
cos +
r2 2 sin(2)
5
r
17
r
A11 =

Example 4
Consider the BVP

ut (x, t) = uxx (x, t) ,

u(0, t) = 0 ,
(22)
u(, t) = ux (, t) ,

u(x, 0) = f (x)

0 < x < , t > 0 ;

t>0;
t>0;
0<x<.

This problem models heat conduction in a rod in which one end (at x = 0) is kept
at 0 degrees and at the other end (x = ) the heat ux is controlled. The initial
temperature is f (x).

12

THE METHOD OF SEPARATION OF VARIABLES

Step 1. The homogeneous part is

{
ut (x, t) = uxx (x, t)
(23)
u(0, t) = 0, u(, t) = ux (, t)

0 < x < , t > 0 ;

t>0.

and the nonhomogeneous part is

(24)

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

Step 2. Separation of variables for the homogeneous part. Suppose u(x, t) =

X(x)T (t) solves the homogeneous problem (23). Then
{
X(x)T (t) = X (x)T (t)
0 < x < , t > 0 ;
(25)
X(0)T (t) = 0, X()T (t) = X ()T (t)
t>0.
The separation of variables leads to the ODE problems for X and T . The Xproblem:
{
X + X = 0
0<x<;
(26)
X(0) = 0 X() = X () ,
and the T -problem:
(27)

T + T = 0 ,

t>0.

Step 3. Eigenvalues and eigenfunctions of the X-problem. The characteristic

equation of the ODE in the SL-problem is

m2 + = 0 with roots m = .
Depending on the sign of , we distinguish three cases:
Case : < 0. Set = 2 with > 0. The characteristic roots are m = .
The general solution of the ODE in the X-problem is
X(x) = Aex + Bex ,
with A and B constants. The endpoint conditions implies
A + B = 0,
Ae + Be = [Ae Be ] .
In order to get a nontrivial solution must satisfy the equation
e e = [e + e ]
or equivalently
1+
1
This equation has a unique positive solution 0 . Furthermore 0 (0, 1) (check
this claim as an exercise). Thus 0 = 02 is the unique negative eigenvalue with
the corresponding eigenfunction
e2 =

X0 (x) =

e0 x e0 x
= sinh(0 x)
2

Case : = 0. I leave it as an exercise for you to show that 0 cannot be an

eigenvalue.

THE METHOD OF SEPARATION OF VARIABLES

13

Case : > 0. We write = 2 with > 0. The characteristic roots are

m = i. The general solution of the ODE in the X-problem is
X(x) = A cos (x) + B sin (x) ,
with A and B constants. The endpoint conditions implies
A cos 0 + B sin 0 = 0,
A cos() + B sin() = A sin() + B cos() .
The rst equation gives A = 0 and the second reduces to
B sin() = B cos() .
Thus in order to have a nontrivial solution X, the parameter > 0 we must satisfy
sin() = cos() tan() = .

(28)

This equation has innitely many positive solutions. For each k Z+ , it has a

Figure 3. Solutions of equation tan() =

unique solution k between k and k + (1/2). To see why, consider the function
f () = sin() cos() .
We have f (k) = k(1) and f (k + (1/2)) = (1)k . These values have opposite
signs. It follows from the intermediate value theorem that there is at least on
number k (k, k + (1/2)) at which f (k ) = 0. To show that there is only one
solution between k and k + (1/2), we can look at the derivative of f :
k

f () = ( 1) cos() + sin() .
Note that cos() and sin() are either both positive or both negative in the
interval (k, k + (1/2)). Hence, f does not change sign in the interval and f is
either increasing or either decreasing in the interval. Therefore, it cannot have
two zeroes. I will leave it as an exercise to verify that f does not have zeroes

14

THE METHOD OF SEPARATION OF VARIABLES

between k + (1/2) and k + 1. We can also consider the second form of the equation
tan() = . The solutions are given by the intersection of the graphs of the
functions y = tan() and y = (See gure).
Therefore, to each solution j > 0 of the above equation tan() = corresponds
a nontrivial solution Xj (x) = sin(j x) of the Sturm-Liouville problem (26). A
numerical estimate of the rst 5 values j and j = j2 is given in the following
table.
n
n
n = n2

1
1.2901
1.6644

2
2.3731
5.6314

3
3.4092
11.6225

4
5
4.4293 5.4422
19.6189 29.6171

Step 4. For = n = n2 , the T -problem is:

T (t) + n2 T (t) = 0 .
The general solution is
T (t) = Cen t
2

with C constant.

For = 0 = 02 . The T -problem is:

T (t) 02 T (t) = 0 .
The general solution is
2

T (t) = Ce0 t

with C constant.

Step 5. For each n Z+ , a solution with separated variable of the homogeneous

part (23) is
un (x, t) = Tn (t)Xn (x) = en t sin(n x) .
2

and for n = 0 we have

2

u0 (x, t) = T0 (t)X0 (x) = e0 t sinh(0 x) .

Step 6. Any linear combination of the solutions un is again a solution of problem

(23). That is, for c0 , c1 , , constants the function
u(x, t) = c0 e0 t sinh(0 x) + c1 e1.6644 t sin(1.2901x) + c2 e5.6314 t sin(2.3731x) +
2

solves (23).
Step 7. If we know how to express f (x) in terms of the eigenfunctions sin(n x)
and sinh(0 x), then we would match the coecients. To complete this step, we
need to study Fourier series.

THE METHOD OF SEPARATION OF VARIABLES

15

Example 5
Consider the BVP
(29)

utt (x, y, t) = uxx (x, y, t) + uyy (x, y, t) ,

u(x, y, 0) = 0 ,
ut (x, y, 0) = sin x sin(7y/2) ,

u(x, 0, t) = u(x, 2, t) = 0 ,

u(0, y, t) = u(, y, t) = 0 ,

0 < x < , 0 < y < 2 t > 0 ;

0 < x < , 0 < y < 2 ;
0 < x < , 0 < y < 2 ;
0<x< t>0;
0 < y < 2 , t > 0.

This problem models transversal vibrations of a rectangular membrane with xed

boundary. In this example, we have three independent variables x, y, and t.
Step 1. The homogeneous part is
(30)

utt (x, y, t) = uxx (x, y, t) + uyy (x, y, t) ,

u(x, y, 0) = 0 ,
u(x, 0, t) = u(x, 2, t) = 0 ,

u(0, y, t) = u(, y, t) = 0 ,

0 < x < , 0 < y < 2 t > 0 ;

0 < x < , 0 < y < 2 ;
0<x< t>0;
0 < y < 2 , t > 0.

and the nonhomogeneous part is

(31)

ut (x, y, 0) = sin x sin

7y
,
2

0 < x < , 0 < y < 2 .

Step 2. Separation of variables for the homogeneous part. Suppose u(x, y, t) =

X(x)Y (y)T (t) solves the homogeneous problem (30). Then
(32)

XY T = X Y T + XY T ,
0 < x < , 0 < y < 2 t > 0 ;

X(x)Y (y)T (0) = 0 ,

0 < x < , 0 < y < 2 ;
X(x)Y
(0)T
(t)
=
X(x)Y
(2)T
(t)
=
0
,
0<x< t>0;

X(0)Y (y)T (t) = X()Y (y)T (t) = 0 ,

0 < y < 2 , t > 0.
We separate the PDE as follows
T (t)
X (x) Y (y)
=
+
= (constant).
T (t)
X(x)
Y (y)
Then
X (x)
Y (y)
=
= (constant).
X(x)
Y (y)
If we let = , then we get
X (x)
Y (y)
T (t)
= ,
= ,
= , = +
X(x)
Y (y)
T (t)
The boundary conditions translate into
X(0) = X() = 0,

Y (0) = Y (2) = 0,

T (0) = 0

The ODE problems for X, Y and T are:

(33)

X + X = 0 ,
Y + Y = 0 ,
T + T = 0 ,

X(0) = X() = 0
Y (0) = Y (2) = 0
T (0) = 0 .

with = + . The X-problem and Y -problems are independent SL-problems.

Step 3. Eigenvalues and eigenfunctions of the X-problem and the Y -problem.

16

THE METHOD OF SEPARATION OF VARIABLES

For the X-problem we have j Z+ , the eigenvalue and eigenfunction are

j = j 2 ,

Xj (x) = sin(jx) .

For the Y -problem we have k Z+ , the eigenvalue and eigenfunction are

( )2
k
ky
k =
,
Yk (y) = sin
.
2
2
Step 4. For each pair of eigenvalues j , k , solve the T -problem. We have
jk = j + k = j 2 +

4j 2 + k 2
k2
2
=
= jk
4
4

The corresponding general solution of the ode for T is

T = A cos(jk t) + B sin(jk t).
The initial condition T (0) = 0 gives A = 0.
Step 5. For each j Z+ and each k Z+ , a solution with separated variable of
the homogeneous part (30) is
ujk (x, y, t) = Tjk (t)Xj (x)Yk (y) = sin(jk t) sin(jx) sin

ky
2

4j 2 + k 2
.
2
Step 6. Any linear combination of the solutions ujk is again a solution of problem
(30). That is, any function of the form

ky
cjk sin(jk t) sin(jx) sin
u(x, y, t) =
2
with jk =

jN, kM

solves problem (30).

Step 7. Finally, we need to nd the coecients cjk so that the above function
solves the nonhomogeneous condition. We have

ky
ut (x, y, t) =
cjk jk cos(jk t) sin(jx) sin
2
jN, kM

Hence,
ut (x, y, 0) =

cjk jk sin(jx) sin

jN, kM

ky
7y
= sin x sin
2
2

Therefore, cjk = 0 for all j, k except when j = 1 and k = 7, in which case

1,7 c1,7 = 1 c1,7 =

1
2
= .
1,7
53

The solution of the original problem (29) is

( )
2
7y
53
u(x, y, t) = sin
t sin x sin
.
2
2
53

THE METHOD OF SEPARATION OF VARIABLES

17

Exercises
Use the method of separation of variables to solve the following boundary value
problems.
Exercise 1.
ut (x, t) = 3uxx (x, t)
u(0, t) = u(2, t) = 0
x
u(x, 0) = sin sin(3x)
2

0 < x < 2 , t > 0

t>0
0 < x < 2

ut (x, t) = uxx (x, t)

ux (0, t) = ux (2, t) = 0
5x
u(x, 0) = 100 3 cos
4

0 < x < 2 , t > 0

t>0

Exercise 2.

0 < x < 2

Exercise 3.
ut (x, t) = 2uxx (x, t)
u(0, t) = ux (, t) = 0
x
7x
11x
u(x, 0) = 50 sin 25 sin
+ 10 sin
2
2
2
Exercise 4.

utt (x, t) = 2uxx (x, t)

u(0, t) = u(3, t) = 0
ut (x, 0) = 0
u(x, 0) = f (x)

0 < x < , t > 0

t>0
0<x<

0 < x < 3 , t > 0

t>0
0 < x < 3
0 < x < 3

where
f (x) = sin
Exercise 5.

x 1
1
7x
sin(2x) + sin
3 2
5
3

utt (x, t) = 2uxx (x, t)

u(0, t) = u(3, t) = 0
u(x, 0) = 0
ut (x, 0) = g(x)

0 < x < 3 , t > 0

t>0
0 < x < 3
0 < x < 3

where
g(x) = sin

2x 1
5x
+ sin
3
2
3

Exercise 6.
utt (x, t) = uxx (x, t)
u(0, t) = ux (1, t) = 0
x
u(x, 0) = sin
2
7x
ut (x, 0) = sin
2

0 < x < 1, t > 0

t>0
0<x<1
0<x<1

18

THE METHOD OF SEPARATION OF VARIABLES

Exercise 7.
utt (x, t) = uxx (x, t)
ux (0, t) = ux (1, t) = 0
u(x, 0) = 0
1
5x
ut (x, 0) = 1 cos
2
2

0 < x < 1, t > 0

t>0
0<x<1
0<x<1

Exercise 8.
u(x, y) = 0
u(0, y) = u(1, y) = 0
u(x, 0) = sin(2x), u(x, 2) = sin(3x)

0 < x < 1, 0 < y < 2

0<y<2
0<x<1

Exercise 9.
u(x, y) = 0

0 < x < 1, 0 < y < 2

y
5y
u(0, y) = 1 cos
, u(1, y) = 3 cos
2
2
uy (x, 0) = uy (x, 2) = 0

0<y<2
0<x<1

Exercise 10. (In polar coordinates)

u(r, ) = 0
u(2, ) = 1 + cos(3) 2 sin(5)

0 < r < 2 , 0 2
0 2

Exercise 11. (In polar coordinates)

u(r, ) = 0
u(1, ) = sin(3)
ur (2, ) = cos(5)
Exercise 12.
ut (x, y, t) = u(x, y, t)
u(x, 0, t) = u(x, 2, t) = 0
u(0, y, t) = u(, y, t) = 0
3y
u(x, y, 0) = sin x sin
2
Exercise 13.
ut (x, y, t) = u(x, y, t)
u(x, 0, t) = u(x, 2, t) = 0
ux (0, y, t) = ux (, y, t) = 0
u(x, y, 0) = sin 5x sin y

1 < r < 2 , 0 2
0 2
0 2

0 < x < , 0 < y < 2 , t > 0

0 < x < , t > 0
0 < y < 2 , t > 0
0 < x < , 0 < y < 2

0 < x < , 0 < y < 2 , t > 0

0 < x < , t > 0
0 < y < 2 , t > 0
0 < x < , 0 < y < 2

Exercise 14.

utt (x, y, t) = 2u(x, y, t)

uy (x, 0, t) = uy (x, 2, t) = 0
u(0, y, t) = u(, y, t) = 0
5y
u(x, y, 0) = sin 3x cos
2
ut (x, y, 0) = 0

0 < x < , 0 < y < 2 , t > 0

0 < x < , t > 0
0 < y < 2 , t > 0
0 < x < , 0 < y < 2
0 < x < , 0 < y < 2

In exercises 15 to 17 nd all solutions with separated variables.

THE METHOD OF SEPARATION OF VARIABLES

Exercise 15.
ut (x, t) = 3uxx (x, t)
u(0, t) = 0
u(2, t) = ux (2, t)

0 < x < 2, t > 0

t>0
t>0

ut (x, t) = 3uxx (x, t)

ux (0, t) = 0
u(2, t) = ux (2, t)

0 < x < 2, t > 0

t>0
t>0

ut (x, t) = 3uxx (x, t)

u(0, t) = ux (0, t)
u(2, t) = ux (2, t)

0 < x < 2, t > 0

t>0
t>0

Exercise 16.

Exercise 17.

19