24th October, 2011.

By Samsung, SUM Sung Fung.

MATH4052 - Partial Differential Equations

Tutorial Notes #07
Samsung, SUM Sung Fung <delamath@ust.hk>

Techniques to Solve PDEs (II)

Separation of Variables
Before the midterm, we have discussed the wave and diffusion equations on a
compact domain [a, b] , with DBCs or NBCs in end points, and the only method we
knew was to make periodic (repeated) extensions and solve the extended system on
the whole line. The figure below shows an example of such tedious extension:
ext

a b x
odd even odd even odd even odd even odd

Lets try to solve those problems on another way, which is much simpler!

Observe that linearity of a system is a very nice property: suppose u and v are
solutions of a linear system, then any element in span{u , v} {au + bv : a, b } is
also a solution of the same system. In this course, our system is a PDE system and we
will play with this property frequently from now on.

Conversely, suppose we have a linear system of PDEs, denoted as L (u ) = 0 , then any

solution v to such problem can be divided into to solutions v1 and v2 , with
L (v1 ) = 0 and L (v2 ) = 0 . We can further subdivide v1 and v2 into v11 , v12 and
v21 , v22 ; and so on into v111 , v112 , v121 , v122 , .

Now, we assume that upon such (infinite) process, we can reduce v to countably
many components, say 1 , 2 , 3 , , and all those n have an extremely nice form:

n ( x, t ) = X n ( x) Tn (t )

We can try to solve n one by one, and v can then be recovered by v = cnn ,
n

where cn are some constants to be determined from ICs or BCs.

Note: The above assumption has to be verified later.

MATH4052 Tutorial Notes #07 P.1 of 7

 24th October, 2011.
By Samsung, SUM Sung Fung.

Boundary Conditions

Suppose is a connected domain (usually in n ) and L is a linear differential

operator of functions in , our scope of linear PDE system is restricted to the
systems of the following forms:

L(u ) = 0 x , t > 0

ICs x , t = 0
zero BCs x , t > 0

The zero boundary conditions refer to the following types:

(i) Dirichlet Boundary Condition(DBC)

u = 0

(ii) Neumann Boundary Condition(NBC)

u
=0
n

Reduction to these zero BCs from non-zero BCs can be achieved, by methods similar
to those stated in previous tutorial notes.

Explicit Examples: Homogeneous Wave Equation

To solve the wave equation

utt c 2u xx = 0 x (0, l ), t > 0

u ( x, 0) = ( x) x (0, l )

ut ( x, 0) = ( x) x (0, l )
zero BCs x = 0, l t > 0

by separation of variables, we first assume v( x, t ) = X ( x)T (t ) , and plug this into

the wave equation:

vtt c 2 vxx = X ( x)T ''(t ) c 2 X ''( x)T (t ) = 0

Suppose we exclude the trivial case v 0 , i.e. assume X ( x), T (t ) 0 . We have

T ''(t ) X ''( x)
=
c 2T (t ) X ( x)

Note that the LHS depends only on t and the RHS only on x ; if we fix x and
vary t , as the RHS is fixed, the LHS is also fixed in spite of variation in t .

MATH4052 Tutorial Notes #07 P.2 of 7

 24th October, 2011.
By Samsung, SUM Sung Fung.

Similarly, if we fix t , the RHS is unchanged with respect to x . The only

possibility is that both sides equal to the same constant, say . The minus sign
is introduced by convention.

Solving the LHS, for 0 :

T (t ) = Aect + Be ct
T ''(t )
= or equivalently
c 2T (t )
( )
T (t ) = C sin ct + D cos ct ( )
The two forms above are equivalent by eiz = cos z + i sin z .

For = 0 ,
T ''(t ) = 0 T (t ) = A + Bt

Solving The Spatial Part

To find the eigenvalue and the eigenfunction X ( x) , we need to make use

of the zero BCs, a few examples are given as follows.

1. Zero DBC at both ends

Suppose u (0, t ) = u (l , t ) = 0 , we have X (0) = X (l ) = 0 . Thus what we need to

solve is the system of ODE in x :

X ''( x) = X ( x) x (0, l )

X (0) = X (l ) = 0
From the first equation, we get
x
X ( x) = Ae x
+ Be

X '' = X or equivalently
( )
X ( x) = C sin x + D cos x ( )
For convenience, lets choose the second form. Plugging it into the second
equation yields
X (0) = D = 0
( ) (
X (l ) = C sin l + D cos l = 0 )
(
We have D = 0 and C sin l = 0 . )
Since C = 0 is the trivial case, the non-trivial solution requires sin l = 0 , ( )
which is equivalent to l = n , n = 1, 2, 3, (Why n 0 is ignored?)

MATH4052 Tutorial Notes #07 P.3 of 7

 24th October, 2011.
By Samsung, SUM Sung Fung.
2
n
So the eigenvalues are n = , n = 1, 2, 3, and the corresponding
l
n
eigenfunctions are X n = C sin x .
l

2. Zero NBC at both ends

Suppose u x (0, t ) = u x (l , t ) = 0 , we have X '(0) = X '(l ) = 0 . Thus what we need

to solve is the system of ODE in x :
X ''( x) = X ( x) x (0, l )

X '(0) = X '(l ) = 0
From the first equation, we get
x + Be
X ( x) = Ae x
X '' = X or equivalently
( )
X ( x) = C sin x + D cos x ( )
Again choose the second form. Plugging it into the second equation yields

( )
X '(0) = C cos 0 D sin 0 = C = 0 ( )
X '(l ) = C cos ( l ) D sin ( l ) = 0

We have C = 0 and D sin l = 0 . ( )

Since D = 0 is the trivial case, the non-trivial solution requires
( )
sin l = 0 , which is equivalent to l = n , n = 0,1, 2, 3, (Why
n < 0 ignored, but NOT for n = 0 ?)
2
n
So the eigenvalues are n = , n = 0,1, 2, 3, and the corresponding
l
n
eigenfunctions are X n = D cos x .
l

Finding The Coefficients

It is NOT an end after finding n and the corresponding X n , because we have

not find the unique solution to our problem yet.
The general solution to our problem is

u ( x, t ) = Tn (t ) X n ( x)
n

And we need to determine all the unknown coefficients from the given ICs.

MATH4052 Tutorial Notes #07 P.4 of 7

 24th October, 2011.
By Samsung, SUM Sung Fung.
For example the wave equation with zero NBCs:

n c n c n
u ( x, t ) = A + Bt + Cn sin t + Dn cos t cos x
n =1 l l l

Here the constant for X n is also absorbed into A , B , Cn and Dn .

If the ICs are u ( x, 0) = ( x) and ut ( x, 0) = ( x) , we have

n
u ( x, 0) = ( x) = A + Dn cos x
n =1 l

n c n
ut ( x, 0) = ( x) = B + Cn cos x
n =1 l l
We can find out the coefficients by the orthogonality of eigenfunctions:
U U

l n m l
0 sin l x sin l x dx = 2 mn
l n
m l
0 cos
x cos
l
x dx = mn
l 2
l n m
0 sin l x cos l x dx = 0
1 m = n
Where mn = is called the Kronecker delta.
0 m n
The coefficients are:
1 l
l 0
A= ( x)dx
1 l
B = ( x)dx
l 0
2 l m x
Cm =
m c 0 ( x) cos
l
dx

2 l m x
Dm = ( x) cos dx
l 0 l
1 1
Sometimes for unified expressions, one can define A = D0 and B = C0 as
2 2
stated in the text book.

MATH4052 Tutorial Notes #07 P.5 of 7

 24th October, 2011.
By Samsung, SUM Sung Fung.

Supplementary Exercise

(i) Solve the following problem by separation of variables

utt c 2u xx = 0 x (0,1), t > 0

u ( x, 0) = 0
ut ( x, 0) = x
u (0, t ) = 0 t>0

u x (1, t ) = 0 t>0

Solution: Substitute v( x, t ) = X ( x)T (t ) to the PDE yields

T '' X ''
= =
c 2T X
The spatial part:

= 0 : X = A + Bx

(
> 0 : X = A sin x + B cos x ) ( )

(
< 0 : X = A sinh x + B cosh x
) ( )
Plugging into the BCs:
= 0 : X (0) = A = 0 , X '(1) = B = 0 trvial solution
> 0 : X (0) = B = 0 , X '(1) = A cos = 0
< 0 : X (0) = B = 0 , X '(1) = A cosh = 0 trvial solution

For the nontrivial case > 0

1
B = 0 , A 0 cos = 0 = n + n = 0,1, 2,
2
1
Corresponding eigenfunction: X = A sin n + x
2
The temporal part:
1 1
( ) ( )
T = A sin ct + B cos ct = A sin n + ct + B cos n + ct
2 2

Hence the general solution with the specific BCs is

1 1 1
u ( x, t ) = An sin n + ct + Bn cos n + ct sin n + x 2
n =0 2 2 2

To obtain all Bm s from the first IC, put t = 0 in equation 2 , multiply both sides
1
by sin m + x and integrate along the interval (0,1) , Am s are found from
2
such similar way by differentiating 2 w.r.t t and making use of the second IC:

MATH4052 Tutorial Notes #07 P.6 of 7

 24th October, 2011.
By Samsung, SUM Sung Fung.

m
2 1 1 ( 1) 2
Am =
( m + 1/ 2 ) c 0
x sin m + x dx =
2 3
(
c m+ 1 3
2 )
Bm = 0

(ii) Find the general solution of X ( x) in both cases on P.3-4 for = 0 . In the case
of zero NBC, show that it can be reduced into the form of cosine function with
n = 0 , as stated in P.4.

Solution: For = 0 , the spatial part becomes X '' = 0 , so the general solution is
X ( x) = Ax + B

DBC at both sides:

0 = X (0) = A 0 + B
A = B = 0 (trivial solution)
0 = X (l ) = A l + B

NBC at both sides:

0 = X '(0) = A
A=0
0 = X '(l ) = A

By nontrivial requirement, we need B 0 , and the corresponding eigenfunction

becomes
0
X ( x) = B0 = B0 cos ( 0 ) = B0 cos x
l

MATH4052 Tutorial Notes #07 P.7 of 7