Method 2 : Particular Solutions by Separation of Variables

In this method, which is simple but powerful, it is assumed that a solution can be
expressed as a product of unknown functions, each of which depends on only one of
the independent variables. By repetition of this, the unknown functions can be
determined. Superposition of these solutions can then be used to find the actual
solution.

Separation of variables applied when:

⚫ PDE is linear and homogeneous.
⚫ Non-constant coefficients are allowed.
⚫ The boundary conditions are linear and homogeneous.
𝜕𝑢 𝜕𝑢
For example, α | + 𝛽𝑢|𝑥=0 = 0, γ | + 𝛿𝑢|𝑥=𝐿 = 0
𝜕𝑥 𝑥=0 𝜕𝑥 𝑥=𝐿

where α, 𝛽, γ, 𝛿 are constants.

Ex: Solve the boundary value problem

∂𝑢 ∂𝑢
=4 , 𝑢(0, 𝑦) = 8𝑒 −3𝑦
∂𝑥 ∂𝑦
by the method of separation of variables.

<Sol>: Let 𝑢 = 𝑋𝑌, where 𝑋 depends only on 𝑥 and 𝑌 depends only on 𝑦.

then 𝑋 ′ 𝑌 = 4𝑋𝑌 ′ ⇒ 𝑋 ′ /4𝑋 = 𝑌 ′ /𝑌, where 𝑋 ′ = 𝑑𝑋/𝑑𝑥 and 𝑌 ′ = 𝑑𝑌/𝑑𝑦

Since 𝑋 depends only on 𝑥 and 𝑌 depends only on 𝑦 and since 𝑥 and 𝑦 are
independent variables, each side must be a constant, say c.
⇒ 𝑋 ′ /4𝑋 = 𝑌 ′ /𝑌 = 𝑐
Then, 𝑋 ′ − 4𝑐𝑋 = 0, 𝑌 ′ − 𝑐𝑌 = 0
whose solutions are 𝑋 = 𝐴 𝑒 4𝑐𝑥 , 𝑌 = 𝐵 𝑒 𝑐𝑦
∴ 𝑢(𝑥, 𝑦) = 𝑋𝑌 = 𝐴𝐵 𝑒 𝑐(4𝑥+𝑦) = 𝐾 𝑒 𝑐(4𝑥+𝑦)

B.C.: 𝑢(0, 𝑦) = 𝐾 𝑒 𝑐𝑦 = 8𝑒 −3𝑦 ⇒ 𝐾 = 8, 𝑐 = −3

Then, 𝑢(𝑥, 𝑦) = 8𝑒 −3(4𝑥+𝑦) = 8𝑒 −12𝑥−3𝑦 is the solution.

Ex: Solve
∂𝑢 ∂2 𝑢
= 2 2 , 0 < 𝑥 < 3, 𝑡 > 0
∂𝑡 ∂𝑥
Given that 𝑢(0, 𝑡) = 𝑢(3, 𝑡) = 0 , 𝑢(𝑥, 0) = 5 sin 4𝜋𝑥 − 3 sin 8𝜋𝑥 + 2 sin 10𝜋𝑥 ,
|𝑢(𝑥, 𝑡)| < 𝑀, where the last condition states that 𝑢 is bounded.

<Sol>: Let 𝑢 = 𝑋𝑌, then 𝑋𝑇 ′ = 2𝑋 ′′ 𝑇 ⇒ 𝑋 ′′ /𝑋 = 𝑇 ′ /2𝑇

Each side must be a constant, which we call −𝜆2 . (If we use 𝜆2 , the resulting solution
obtained 𝑒 ±𝜆𝑥 does not satisfy the boundedness condition for real values of 𝜆).

Then 𝑥 ′′ + 𝜆2 𝑥 = 0, 𝑇 ′ + 2𝜆2 𝑇 = 0
2
with solutions 𝑋 = 𝐴1 cos 𝜆𝑥 + 𝐵1 sin 𝜆𝑥 , 𝑇 = 𝑐1 𝑒 −2𝜆 𝑡 (𝑡 > 0 ∴ bounded)
2 2
∴ 𝑢(𝑥, 𝑡) = 𝑋𝑇 = 𝐶1 𝑒 −2𝜆 𝑡 (𝐴1 cos 𝜆𝑥 + 𝐵1 sin 𝜆𝑥) = 𝑒 −2𝜆 𝑡 (𝐴cos 𝜆𝑥 + 𝐵sin 𝜆𝑥)

2
Since 𝑢(0, 𝑡) = 0, 𝑒 −2𝜆 𝑡 (𝐴) = 0 ⇒ 𝐴=0
2
Then, 𝑢(𝑥, 𝑡) = 𝐵𝑒 −2𝜆 𝑡 sin 𝜆𝑥

2
Since 𝑢(3, 𝑡) = 0, 𝐵𝑒 −2𝜆 𝑡 sin 3𝜆 = 0
If 𝐵 = 0, the solution is identically zero, so we must have sin 3𝜆 = 0
or 3𝜆 = 𝑚𝜋 ⇒ 𝜆 = 𝑚𝜋/3, where 𝑚 = 0, ±1, ±2, ⋯ ⋯
2𝑚2 𝜋2 𝑡
∴ 𝑢(𝑥, 𝑡) = 𝐵𝑒 − 9 sin 𝑚𝜋𝑥/3

By the principle of superposition,

2𝑚12 𝜋2 𝑚1 𝜋𝑥 2𝑚2 𝜋 𝑚2 𝜋𝑥 2𝑚2 𝜋2 𝑚3 𝜋𝑥
− 9 𝑡 sin − 92 𝑡 − 93 𝑡
𝑢(𝑥, 𝑡) = 𝐵1 𝑒 + 𝐵2 𝑒 sin + 𝐵3 𝑒 sin
3 3 3
is also a solution.

By the last boundary condition,

𝑢(𝑥, 0) = 𝐵1 sin 𝑚1 𝜋𝑥/3 + 𝐵2 sin 𝑚2 𝜋𝑥/3 + 𝐵3 sin 𝑚3 𝜋𝑥/3
= 5sin4𝜋𝑥 − 3sin8𝜋𝑥 + 2sin10𝜋𝑥
⇒ 𝐵1 = 5, 𝑚1 = 12, 𝐵2 = −3, 𝑚2 = 24, 𝐵3 = 2, 𝑚3 = 30
2 2 2
∴ 𝑢(𝑥, 𝑡) = 5𝑒 −32𝜋 𝑡 sin4𝜋𝑥 − 3𝑒 −128𝜋 𝑡 sin8𝜋𝑥 + 2𝑒 −200𝜋 𝑡 sin10𝜋𝑥

This boundary value problem has the following interpretation as a heat flow problem.
A bar whose surface is insulated has a length of 3 units and a diffusivity of 2 units. If
its ends are kept at temperature zero units and its initial temperature 𝑢(𝑥, 0) =
5sin4𝜋𝑥 − 3sin8𝜋𝑥 + 2sin10𝜋𝑥, find the temperature at position 𝑥 at time 𝑡, i.e.,
find 𝑢(𝑥, 𝑡).
Partial Differential Equation + Fourier Series to Heat Conduction
Ex: Same example as the above. Find the temperature of the bar, if the initial
temperature 𝑢(𝑥, 0) is 25∘ C.

<Sol>: This problem is identical to the previous example, except that the initial
condition 𝑢(𝑥, 0) = 25. It is necessary to superimpose an infinite number of
solutions, i.e., we must replace the equation by
∞
2 𝜋2 𝑡/9 𝑚𝜋𝑥
𝑢(𝑥, 𝑡) = ∑ 𝐵𝑚 𝑒 −2𝑚 𝑠𝑖𝑛
3
𝑚=1

which for 𝑡 = 0 yields

∞
𝑚𝜋𝑥
25 = ∑ 𝐵𝑚 sin , 0<𝑥<3
3
𝑚=1

This amounts to expanding this equation in a Fourier sine series:

2 𝐿 𝑚𝜋𝑥 2 3 𝑚𝜋𝑥 50(1 − 𝑐𝑜𝑠 𝑚𝜋)
𝐵𝑚 = ∫ 𝑓(𝑥) 𝑠𝑖𝑛 𝑑𝑥 = ∫ 25 𝑠𝑖𝑛 𝑑𝑥 =
𝐿 0 𝐿 3 0 3 𝑚𝜋

The result can be written as:

∞
50(1 − 𝑐𝑜𝑠 𝑚𝜋) −2𝑚2𝜋2𝑡/9 𝑚𝜋𝑥 100 −2𝜋2𝑡 𝜋𝑥 1 −2𝜋2𝑡
𝑢(𝑥, 𝑡) = ∑ 𝑒 𝑠𝑖𝑛 = {𝑒 9 𝑠𝑖𝑛 + 𝑒 𝑠𝑖𝑛 𝜋 𝑥 + ⋯ ⋯ }
𝑚𝜋 3 𝜋 3 3
𝑚=1

This problem illustrates the importance of Fourier series in solving boundary value
problems.

𝜕𝑢 𝜕2𝑢
Ex: Solve =2 , 𝑢(0, 𝑡) = 10, 𝑢(3, 𝑡) = 40, 𝑢(𝑥, 0) = 25, |𝑢(𝑥, 𝑡) | < 𝑀
𝜕𝑡 𝜕𝑥 2

<Sol>: This problem is identical to the previous example, except that the ends of the
bar arm are at temperatures 10°C and 40°C instead of 0°C. As far as the solution goes,
this makes quite a difference since we can no longer use separation of variables and
𝑚𝜋
conclude that 𝐴 = 0 and 𝜆 = as in the previous example. To solve this problem,
3

assume that 𝑢(𝑥, 𝑡) = 𝑣(𝑥, 𝑡) + 𝜓(𝑥), where 𝜓(𝑥) is to be suitably determined. In

terms of 𝑣(𝑥, 𝑡), the boundary value problem becomes:
 𝜕𝑣 𝜕2𝑣
= 2 2 + 2𝜓 ′′ (𝑥 ), 𝑣(0, 𝑡) + 𝜓(0) = 10, 𝑣(3, 𝑡) + 𝜓(3) = 40,
𝜕𝑡 𝜕𝑥
𝑣(𝑥, 0) + 𝜓(𝑥 ) = 25

This can be simplified by choosing 𝜓 ′′ (𝑥) = 0, 𝜓(0) = 10, 𝜓(3) = 40 from which
we find 𝜓(𝑥) = 10𝑥 + 10.

The resulting boundary value problem becomes:

𝜕𝑣 𝜕2𝑣
= 2 2, 𝑣(0, 𝑡) = 0, 𝑣(3, 𝑡) = 0,
𝜕𝑡 𝜕𝑥
𝑣(𝑥, 0) = 25 − (10𝑥 + 10) = 15 − 10𝑥
∞
2 𝜋2 𝑡/9 𝑚𝜋𝑥
∴ v(x, t) = ∑ 𝐵𝑚 𝑒 −2𝑚 𝑠𝑖𝑛
3
𝑚=1

which brings the boundary conditions back to zero and allows us to use separation of
variables again.

The last initial condition yields:

∞
𝑚𝜋𝑥
15 − 10x = ∑ 𝐵𝑚 𝑠𝑖𝑛
3
m=1
3
2 𝑚𝜋𝑥 30
from which 𝐵𝑚 = ∫ (15 − 10𝑥) 𝑠𝑖𝑛 𝑑𝑥 = (𝑐𝑜𝑠 𝑚𝜋 − 1)
3 0 3 𝑚𝜋

Since 𝑢(𝑥, 𝑡) = 𝑣(𝑥, 𝑡) + 𝜓(𝑥), we have

∞
30 2 2 𝑚𝜋𝑥
𝑢(𝑥, 𝑡) = 10𝑥 + 10 + ∑ (𝑐𝑜𝑠 𝑚𝜋 − 1)𝑒 −2𝑚 𝜋 𝑡/9 𝑠𝑖𝑛
𝑚𝜋 3
𝑚=1

When 𝑡 → ∞ , the summation term approaches zero. The term 10𝑥 + 10 is the
steady-state temperature, i.e. the temperature after a long time has elapsed. The
summation term represents the transient temperature, i.e., the temperature at the
beginning of time when the system is not yet stable.
Ex: A square plate with sides of unit length has its faces
insulated and its sides kept at 0°C. If the initial
temperature is specified, determine the subsequent
temperature at any point of the plate.

The equation for the temperature 𝑢(𝑥, 𝑦, 𝑡) at any point (𝑥, 𝑦) at time t is
𝜕𝑢 𝜕2𝑢 𝜕2𝑢
= 𝐾( 2 + 2 )
𝜕𝑡 𝜕𝑥 𝜕𝑦
B.C.: 𝑢(0, 𝑦, 𝑡) = 𝑢(1, 𝑦, 𝑡) = 𝑢(𝑥, 0, 𝑡) = 𝑢(𝑥, 1, 𝑡) = 0, 𝑢(𝑥, 𝑦, 0) = 𝑓(𝑥, 𝑦)
|𝑢(𝑥, 𝑦, 𝑡)| < 𝑀, where 0 < 𝑥 < 1, 0 < 𝑦 < 1, 𝑡 > 0

<Sol>: To solve the boundary value problem, let 𝑢 = 𝑋𝑌𝑇 , where 𝑋, 𝑌, 𝑇 are
functions of 𝑥, 𝑦, 𝑡 respectively.
⇒ 𝑋𝑌𝑇′ = 𝐾(𝑋′′𝑌𝑇 + 𝑋𝑌′′𝑇)
Dividing by 𝐾𝑋𝑌𝑇 yields:
𝑇′ 𝑋′′ 𝑌′′
= +
𝐾𝑇 𝑋 𝑌

Since the left side is a function of t alone, while the right side is a function of x and y,
we see that each side must be a constant, say −𝜆2 (which is needed for boundedness).
𝑇 ′ + 𝐾𝜆2 𝑇 = 0
𝑋′′ 𝑌′′
+ = −𝜆2
𝑋 𝑌

Since the left side depends only on x while the right side depends only on y, each side
must be a constant, say −𝜇2 . Thus,
𝑋 ′′ 𝑌 ′′
=− − 𝜆2 = −𝜇2
𝑋 𝑌
⇒ 𝑋 ′′ + 𝜇2 𝑋 = 0, 𝑌 ′′ + (𝜆2 − 𝜇2 )𝑌 = 0

Solution to these equations are given by:

2𝑡
𝑋 = 𝑎1 𝑐𝑜𝑠 𝜇𝑥 + 𝑏1 𝑠𝑖𝑛 𝜇𝑥 , 𝑌 = 𝑎2 𝑐𝑜𝑠 √𝜆2 − 𝜇2 𝑦 + 𝑏2 𝑠𝑖𝑛 √𝜆2 − 𝜇2 𝑦, 𝑇 = 𝑎3 𝑒 −𝑘𝜆
2
∴ 𝑢(𝑥, 𝑦, 𝑡) = (𝑎1 𝑐𝑜𝑠 𝜇𝑥 + 𝑏1 𝑠𝑖𝑛 𝜇𝑥 ) (𝑎2 𝑐𝑜𝑠 √𝜆2 − 𝜇2 𝑦 + 𝑏2 𝑠𝑖𝑛 √𝜆2 − 𝜇2 𝑦) (𝑎3 𝑒−𝑘𝜆 𝑡 )

B. C. : 𝑢(0, 𝑦, 𝑡) = 0 ⇒ 𝑎1 = 0
𝑢(𝑥, 0, 𝑡) = 0 ⇒ 𝑎2 = 0
 2
⇒ 𝑢(𝑥, 𝑦, 𝑡) = 𝐵𝑒 −𝑘𝜆 𝑡 𝑠𝑖𝑛 𝜇𝑥 𝑠𝑖𝑛 √𝜆2 − 𝜇2 𝑦, where 𝐵 = 𝑏1 𝑏2 𝑎3

B. C. : 𝑢(1, 𝑦, 𝑡) = 0 ⇒ 𝑠𝑖𝑛𝜇 = 0 ⇒ 𝜇 = 𝑚𝜋, 𝑚 = 1, 2, 3, ⋯ ⋯

𝑢(𝑥, 1, 𝑡) = 0 ⇒ 𝑠𝑖𝑛 √𝜆2 − 𝜇2 = 0 ⇒ √𝜆2 − 𝜇2 = 𝑛𝜋, 𝑛 = 1, 2, 3, ⋯ ⋯

2 )𝜋2 𝑡
⇒ 𝑢(𝑥, 𝑦, 𝑡) = 𝐵𝑒 −𝑘(𝑚+𝑛 𝑠𝑖𝑛 𝑚𝜋𝑥 𝑠𝑖𝑛 𝑛𝜋𝑦

By the superposition theorem, we can arrive at the possible solution:

∞ ∞
2 )𝜋2 𝑡
𝑢(𝑥, 𝑦, 𝑡) = ∑ ∑ 𝐵𝑚𝑛 𝑒 −𝑘(𝑚+𝑛 𝑠𝑖𝑛 𝑚𝜋𝑥 𝑠𝑖𝑛 𝑛𝜋𝑦
𝑚=1 𝑛=1

∞ ∞

I. C. : 𝑢(𝑥, 𝑦, 0) = 𝑓(𝑥, 𝑦) ⇒ 𝑓(𝑥, 𝑦) = ∑ ∑ 𝐵𝑚𝑛 𝑠𝑖𝑛 𝑚𝜋𝑥 𝑠𝑖𝑛 𝑛𝜋𝑦

𝑚=1 𝑛=1
1 1
where 𝐵𝑚𝑛 = 4 ∫ ∫ 𝑓(𝑥, 𝑦) 𝑠𝑖𝑛 𝑚𝜋𝑥 𝑠𝑖𝑛 𝑛𝜋𝑦 𝑑𝑥𝑑𝑦
0 0

Thus, the formal solution to this problem is

∞ ∞
2 )𝜋2 𝑡
𝑢(𝑥, 𝑦, 𝑡) = ∑ ∑ 𝐵𝑚𝑛 𝑒 −𝑘(𝑚+𝑛 𝑠𝑖𝑛 𝑚𝜋𝑥 𝑠𝑖𝑛 𝑛𝜋𝑦
𝑚=1 𝑛=1
1 1
where 𝐵𝑚𝑛 = 4 ∫ ∫ 𝑓(𝑥, 𝑦) 𝑠𝑖𝑛 𝑚𝜋𝑥 𝑠𝑖𝑛 𝑛𝜋𝑦 𝑑𝑥𝑑𝑦
0 0

Applications to Vibrating Strings and Membranes

Ex: A string of length L is stretched between
points (0,0) and (𝐿, 0) on the x-axis. At time
𝑡 = 0, it has a shape given by 𝑓(𝑥), 0 < 𝑥 < 𝐿,
and it is released from rest. Find the displacement of the string at any later time.

𝜕2𝑦 2
𝜕2𝑦
=𝑎 , 0 < 𝑥 < 𝐿, 𝑡>0
𝜕𝑡 2 𝜕𝑥 2
where 𝑦(𝑥, 𝑡) = displacement from x-axis at time t
B. C. : 𝑦(0, 𝑡) = 𝑦(𝐿, 𝑡) = 0, 𝑡>0
I. C. : 𝑦(𝑥, 0) = 𝑓(𝑥), 0<𝑥<𝐿
𝑦𝑡 (𝑥, 0) = 0 (The initial velocity of the string is 0)
<Sol>: Let 𝑢 = 𝑋𝑇, then 𝑋𝑇 ′′ = 𝑎2 𝑋 ′′ 𝑇 ⇒ 𝑇 ′′ /𝑎2 𝑇 = 𝑋 ′′ /𝑋
2
Calling the separation constant －𝜆 , we have

𝑇 ′′ + 𝜆2 𝑎2 𝑇 = 0, 𝑥 ′′ + 𝜆2 𝑥 = 0
⇒ 𝑇 = 𝐴1 𝑠𝑖𝑛 𝜆𝑎𝑡 + 𝐵1 𝑐𝑜𝑠 𝜆𝑎𝑡 , 𝑋 = 𝐴2 𝑠𝑖𝑛 𝜆𝑥 + 𝐵2 𝑐𝑜𝑠 𝜆𝑥
∴ 𝑦(𝑥, 𝑡) = 𝑋𝑇 = (𝐴2 𝑠𝑖𝑛 𝜆𝑥 + 𝐵2 𝑐𝑜𝑠 𝜆𝑥)(𝐴1 𝑠𝑖𝑛 𝜆𝑎𝑡 + 𝐵1 𝑐𝑜𝑠 𝜆𝑎𝑡)

B. C. : 𝑦(0, 𝑡) = 0 ⇒ 𝐵2 = 0
⇒ 𝑦(𝑥, 𝑡) = 𝐴2 𝑠𝑖𝑛 𝜆𝑥 (𝐴1 𝑠𝑖𝑛 𝜆𝑎𝑡 + 𝐵1 𝑐𝑜𝑠 𝜆𝑎𝑡) = 𝑠𝑖𝑛 𝜆𝑥 (𝐴 𝑠𝑖𝑛 𝜆𝑎𝑡 + 𝐵 𝑐𝑜𝑠 𝜆𝑎𝑡)
𝑚𝜋
𝑦(𝐿, 𝑡) = 0 ⇒ 𝑠𝑖𝑛 𝜆𝐿 = 0 ⇒ 𝜆=
𝐿

Since 𝑦𝑡 (𝑥, 𝑡) = 𝑠𝑖𝑛 𝜆𝑥 (𝐴𝜆𝑎 𝑠𝑖𝑛 𝜆𝑎𝑡 − 𝐵𝜆𝑎 𝑐𝑜𝑠 𝜆𝑎𝑡)

I. C. : 𝑦𝑡 (𝑥, 0) = 0 ⇒ 𝑠𝑖𝑛 𝜆 𝑥(𝐴𝜆𝑎) = 0 ⇒ 𝐴=0
𝑚𝜋𝑥 𝑚𝜋𝑎𝑡
∴ 𝑦(𝑥, 𝑡) = 𝐵 𝑠𝑖𝑛 𝑐𝑜𝑠
𝐿 𝐿

By the superposition theorem,

∞
𝑚𝜋𝑥 𝑚𝜋𝑎𝑡
𝑦(𝑥, 𝑡) = ∑ 𝐵𝑚 𝑠𝑖𝑛 𝑐𝑜𝑠
𝐿 𝐿
𝑚=1

∞
𝑚𝜋𝑥
I. C. : 𝑦(𝑥, 0) = 𝑓(𝑥) = ∑ 𝐵𝑚 𝑠𝑖𝑛
𝐿
𝑚=1

2 𝐿 𝑚𝜋𝑥
where 𝐵𝑚 = ∫ 𝑓(𝑥) 𝑠𝑖𝑛 𝑑𝑥
𝐿 0 𝐿

∞
2 𝐿 𝑚𝜋𝑥 𝑚𝜋𝑥 𝑚𝜋𝑎𝑡
∴ 𝑦(𝑥, 𝑡) = ∑ ( ∫ 𝑓(𝑥) 𝑠𝑖𝑛 𝑑𝑥) 𝑠𝑖𝑛 𝑐𝑜𝑠
𝐿 0 𝐿 𝐿 𝐿
𝑚=1

The terms in this series represent the natural or normal modes of vibration. The
𝑚𝜋𝑎𝑡
frequency of the mth normal mode fm is obtained from the term involving 𝑐𝑜𝑠
𝐿

𝑚𝜋𝑎
and is given by 2𝜋𝑓𝑚 = .
𝐿

𝑚𝑎 𝑚 𝜏
⇒ 𝑓𝑚 = = √
2𝐿 2𝐿 𝜇
Since all the frequencies are integer multiples the lowest frequency f1, the vibrations of
the string will yield a musical tone, as in the case of a violin or piano string. The first
three normal modes are illustrated as follows:

Fundamental modes

first harmonic second harmonic third harmonic

The frequencies and wavelengths of the music pitch A (La):

A3 (La): 220 Hz 157 cm
A4 (La): 440 Hz 78 cm
A5 (La): 880 Hz 39 cm
A6 (La): 1760 Hz 20 cm

Solutions Using Fourier Integrals

Ex: A semi-infinite thin bar (𝑥 ≥ 0) whose surface is insulated has an initial temperature
equal to 𝑓(𝑥) . A temperature of zero is suddenly applied to the end 𝑥 = 0 and
maintained. (a) Set up the boundary value problem for the temperature 𝑢(𝑥, 𝑡) at any
point 𝑥 at time 𝑡, (b) Show that
2 ∞ ∞ 2
𝑢(𝑥, 𝑦) = ∫ ∫ 𝑓(𝑣)𝑒 −𝑘𝜆 𝑡 𝑠𝑖𝑛 𝜆𝑣 𝑠𝑖𝑛 𝜆𝑥 𝑑𝜆 𝑑𝑣
𝜋 0 0

<Sol>: (a) The boundary value problem is

𝜕𝑢 𝜕2𝑢
=𝑘 2 , 𝑥 >0, 𝑡>0
𝜕𝑡 𝜕𝑥
𝑢(𝑥, 0) = 𝑓(𝑥) , 𝑢(0, 𝑡) = 0 , |𝑢(𝑥, 𝑡)| < 𝑀
where the last condition is used since the temperature must be bounded for physical
reasons.

(b) A solution obtained by separation of variables is:

2
𝑢(𝑥, 𝑡) = 𝑒 −𝑘𝜆 𝑡 (𝐴 𝑐𝑜𝑠 𝜆𝑥 + 𝐵 𝑠𝑖𝑛 𝜆𝑥)
2
B. C. : 𝑢(0, 𝑡) = 0 ⟹ 𝐴𝑒 −𝑘𝜆 𝑡 = 0 ⟹ 𝐴=0
−𝑘𝜆2 𝑡
⟹ 𝑢(𝑥, 𝑡) = 𝐵𝑒 𝑠𝑖𝑛 𝜆𝑥
We can integrate over 𝜆 from 0 to ∞ and still has a solution. This is the analog of
the superposition theorem for discrete values of 𝜆 used in connection with Fourier
series.
∞
2
𝑢(𝑥, 𝑡) = ∫ 𝐵(𝜆)𝑒 −𝑘𝜆 𝑡 𝑠𝑖𝑛 𝜆𝑥 𝑑𝜆
0
∞
I. C. : 𝑢(𝑥, 0) = 𝑓(𝑥) = ∫ 𝐵(𝜆) 𝑠𝑖𝑛 𝜆𝑥 𝑑𝜆
0

Since 𝑓(𝑥) must be an odd function, we have

2 ∞ 2 ∞
𝐵(𝜆) = ∫ 𝑓(𝑥) 𝑠𝑖𝑛 𝜆𝑥 𝑑𝑥 = ∫ 𝑓(𝑣) 𝑠𝑖𝑛 𝜆𝑣 𝑑𝑣
𝜋 0 𝜋 0
2 ∞ ∞ 2
∴ 𝑢(𝑥, 𝑡) = ∫ ∫ 𝑓(𝑣)𝑒 −𝑘𝜆 𝑡 𝑠𝑖𝑛 𝜆𝑣 𝑠𝑖𝑛 𝜆𝑥 𝑑𝜆 𝑑𝑣
𝜋 0 0

Ex: Find a bounded solution to Laplace’s

equation 𝛻 2 𝑣 = 0 for the half plane 𝑦 > 0 if
𝑣 takes on the value 𝑓(𝑥) on the 𝑥-axis.

<Sol>: The boundary value problem is given by

𝜕2𝑣 𝜕2𝑣
+ =0, 𝑣(𝑥, 0) = 𝑓(𝑥) , |𝑣(𝑥, 𝑦)| < 𝑀
𝜕𝑥 2 𝜕𝑦 2
𝑋 ′′ 𝑌 ′′
Let 𝑣 = 𝑋𝑌 , =− = −𝜆2
𝑋 𝑌
⟹ 𝑋 ′′ + 𝜆2 𝑋 = 0 , 𝑌 ′′ − 𝜆2 𝑌 = 0
⟹ 𝑋 = 𝑎1 𝑐𝑜𝑠 𝜆𝑥 + 𝑏1 𝑠𝑖𝑛 𝜆𝑥 , 𝑌 = 𝑎2 𝑒 𝜆𝑦 + 𝑏2 𝑒 −𝜆𝑦

Then the solution is 𝑣(𝑥, 𝑦) = (𝑎1 𝑐𝑜𝑠 𝜆𝑥 + 𝑏1 𝑠𝑖𝑛 𝜆𝑥)(𝑎2 𝑒 𝜆𝑦 + 𝑏2 𝑒 −𝜆𝑦 ).

If 𝜆 > 0, the term in 𝑒 𝜆𝑦 is unbounded as y → ∞; so that to keep 𝑣(𝑥, 𝑦) bounded

we must have 𝑎2 = 0.
∴ 𝑣(𝑥, 𝑦) = 𝑒 −𝜆𝑦 (𝐴 𝑐𝑜𝑠 𝜆𝑥 + 𝐵 𝑠𝑖𝑛 𝜆𝑥)

Integrating over 𝜆:
∞
𝑣(𝑥, 𝑦) = ∫ 𝑒 −𝜆𝑦 [𝐴(𝜆) 𝑐𝑜𝑠 𝜆𝑥 + 𝐵(𝜆) 𝑠𝑖𝑛 𝜆𝑥]𝑑𝜆
0
∞
B. C. : 𝑣(𝑥, 0) = 𝑓(𝑥) = ∫ [𝐴(𝜆) 𝑐𝑜𝑠 𝜆𝑥 + 𝐵(𝜆) 𝑠𝑖𝑛 𝜆𝑥]𝑑𝜆
0
From the Fourier’s integral theorem, we find:
1 ∞ 1 ∞
𝐴(𝜆) = ∫ 𝑓(𝑢) 𝑐𝑜𝑠 𝜆𝑢 𝑑𝑢 , 𝐵(𝜆) = ∫ 𝑓(𝑢) 𝑠𝑖𝑛 𝜆𝑢 𝑑𝑢
𝜋 −∞ 𝜋 −∞
1 ∞ ∞
∴ 𝑣(𝑥, 𝑦) = ∫ ∫ 𝑒 −𝜆𝑦 𝑓(𝑢) 𝑐𝑜𝑠 𝜆(𝑢 − 𝑥) 𝑑𝑢 𝑑𝜆
𝜋 𝜆=0 𝑢=−∞

PDE + Sturm-Liouville System

(a) Show that separation of variables in the boundary value problem
𝜕𝑢 𝜕 𝜕𝑢
𝑔(𝑥) = [𝑘(𝑥) ] + ℎ(𝑥)𝑢 , 0<𝑥<𝐿, 𝑡>0
𝜕𝑡 𝜕𝑥 𝜕𝑥
𝑢(0, 𝑡) = 0 , 𝑢(𝐿, 𝑡) = 0 , 𝑢(𝑥, 0) = 𝑓(𝑥) , |𝑢(𝑥, 𝑡)| < 𝑀
leads to Sturm-Liouville system, (b) Give a physical interpretation of the equation, (c)
How would you proceed to solve the boundary value problem?

<Sol>: Let 𝑢 = 𝑋𝑇
𝑑 𝑑𝑋
𝑔(𝑥)𝑋𝑇 ′ = 𝑇 [𝑘(𝑥) ] + ℎ(𝑥)𝑋𝑇
𝑑𝑥 𝑑𝑥

Dividing by 𝑔(𝑥)𝑋𝑇
𝑇′ 1 𝑑 𝑑𝑋 ℎ(𝑥)
= [𝑘(𝑥) ] +
𝑇 𝑔(𝑥)𝑋 𝑑𝑥 𝑑𝑥 𝑔(𝑥)

Setting each side equal to – 𝜆:

T ′ + 𝜆𝑇 = 0
𝑑 𝑑𝑋
[𝑘(𝑥) ] + [ℎ(𝑥) + 𝜆𝑔(𝑥)]𝑋 = 0
𝑑𝑥 𝑑𝑥
B. C. : 𝑢(0, 𝑡) = 0 ⟹ 𝑋(0) = 0 , 𝑢(𝐿, 𝑡) = 0 ⟹ 𝑋(𝐿) = 0

(b) 𝑢(𝑥): temperature at any point 𝑥 at time 𝑡

𝑘(𝑥): thermal conductivity (nonconstant)
𝑔(𝑥): specific heat multiplied by the density
ℎ(𝑥)𝑢: Newton’s law of cooling type radiation into a medium of temperature zero
is taking place at the surface of the bar, with a proportionality factor that
depends on position.

(c) From the Sturm-Liouville system, we can find eigenvalue 𝜆𝑛 and normalized
eigenfunctions 𝑋𝑛 (𝑥), where 𝑛 = 1, 2, 3, ⋯
𝑇 ′ + 𝜆𝑇 = 0 ⟹ 𝑇 = 𝐶𝑒 −𝜆𝑡
Thus, a solution obtained by superposition is
∞

𝑢(𝑥, 𝑡) = ∑ 𝐶𝑛 𝑒 −𝜆𝑛 𝑡 𝑋𝑛 (𝑥)

𝑛=1

I. C. ∶ 𝑢(𝑥, 0) = 𝑓(𝑥) = ∑ 𝐶𝑛 𝑋𝑛 (𝑥)

𝑛=1

which leads to
𝐿
𝐶𝑛 = ∫ 𝑓(𝑥)𝑋𝑛 (𝑥)𝑑𝑥
0
∞ 𝐿
∴ 𝑢(𝑥, 𝑡) = ∑ {∫ 𝑓(𝑥)𝑋𝑛 (𝑥)𝑑𝑥 } 𝑒 −𝜆𝑛𝑡 𝑋𝑛 (𝑥)
𝑛=1 0

Solutions Using Laplace Transform

Ex: Solve by Laplace transform the boundary-value problem
𝜕𝑢 𝜕2𝑢
=4 2
𝜕𝑡 𝜕𝑥
𝑢(0, 𝑡) = 0 , 𝑢(3, 𝑡) = 0 , 𝑢(𝑥, 0) = 10 𝑠𝑖𝑛 2𝜋𝑥 − 6 𝑠𝑖𝑛 4𝜋𝑥

<Sol>: Taking the Laplace transform of the given differential equation with respect to
𝑡, we have
∞ ∞
−𝑠𝑡
𝜕𝑢 −𝑠𝑡
𝜕2𝑢
∫ 𝑒 ( ) 𝑑𝑡 = ∫ 𝑒 (4 2 ) 𝑑𝑡
0 𝜕𝑡 0 𝜕𝑥
∞
−𝑠𝑡
𝑑 2 ∞ −𝑠𝑡
⟹ 𝑠∫ 𝑒 𝑢𝑑𝑡 − 𝑢(𝑥, 0) = 4 2 ∫ 𝑒 𝑢𝑑𝑡
0 𝑑𝑥 0
𝑑2𝑈
⟹ 𝑠𝑈 − 𝑢(𝑥, 0) = 4
𝑑𝑥 2

Using the given condition 𝑢(𝑥, 0) = 10 𝑠𝑖𝑛 2𝜋𝑥 − 6 𝑠𝑖𝑛 4𝜋𝑥, this becomes
𝑑2𝑈
4 − 𝑠𝑈 = 6 𝑠𝑖𝑛 4𝜋𝑥 − 10 𝑠𝑖𝑛 2𝜋𝑥
𝑑𝑥 2
B. C. : 𝑢(0, 𝑡) = 0 ⟹ 𝑈(0, 𝑠) = 0 , 𝑢(3, 𝑡) = 0 ⟹ 𝑈(3, 𝑠) = 0

Solve:
𝑑2𝑈
4 2 − 𝑠𝑈 = 6 𝑠𝑖𝑛 4𝜋𝑥 − 10 𝑠𝑖𝑛 2𝜋𝑥
𝑑𝑥
𝑈(0, 𝑠) = 0 , 𝑈(3, 𝑠) = 0
 10 𝑠𝑖𝑛 2𝜋𝑥 6 𝑠𝑖𝑛 4𝜋𝑥
⟹ 𝑈(𝑥, 𝑠) = −
𝑠 + 16𝜋 2 𝑠 + 64𝜋 2

10 6
∴ u(x, t) = 𝐿−1 {𝑈(𝑥, 𝑠)} = 𝐿−1 { 2
} 𝑠𝑖𝑛 2𝜋𝑥 − 𝐿−1 { } 𝑠𝑖𝑛 4𝜋𝑥
𝑠 + 16𝜋 𝑠 + 64𝜋 2
2 2
= 10𝑒 −16𝜋 𝑡 𝑠𝑖𝑛 2𝜋𝑥 − 6𝑒 −64𝜋 𝑡 𝑠𝑖𝑛 4𝜋𝑥

Theoretically we could have taken the Laplace transform of the given differential
equation with respect to 𝑥 rather than 𝑡 . However, this would lead to various
difficulties as is evident upon carrying out the procedures. In practice, we take
Laplace transform with respect to the various independent variables and then choose
that variable which leads to the greatest simplification.

Solutions Using Bessel Functions of the First Kind

A circular plate of unit radius has its plane faces insulated. If
the initial temperature is 𝐹(𝜌) and if the rim is kept at
temperature zero, find the temperature of the plate at any time.

<Sol>: Since the temperature is initial independent of ∅, the boundary value problem
for determining 𝑢(𝜌, 𝑡) is
𝜕𝑢 𝜕 2 𝑢 1 𝜕𝑢
= 𝑘( 2 + )
𝜕𝑡 𝜕𝜌 𝜌 𝜕𝜌
𝑢(1, 𝑡) = 0 , 𝑢(𝜌, 0) = 𝐹(𝜌) , |𝑢(𝜌, 𝑡)| < 𝑀

Let 𝑢 = 𝑃(𝜌)𝑇(𝑡) = 𝑃𝑇, then

1
𝑃𝑇 ′ = 𝑘 (𝑃′′ 𝑇 + 𝑃′ 𝑇)
𝜌

Dividing by 𝑘PT,
𝑇 ′ 𝑃′′ 1 𝑃′
= + = −𝜆2
𝑘𝑇 𝑃 𝜌𝑃
from which 𝑇 ′ + 𝑘𝜆2 𝑇 = 0
1
𝑃′′ + 𝑃′ + 𝜆2 𝑃 = 0 ⟹ 𝜌2 𝑃′′ + 𝜌𝑃′ + 𝜆2 𝜌2 𝑃 = 0
𝜌
2
⟹ 𝑇 = 𝐶1 𝑒 −𝑘𝜆 𝑡 , 𝑃 = 𝐴1 𝐽0 (𝜆𝜌) + 𝐵1 𝑌0 (𝜆𝜌)
Since 𝑢 = 𝑃𝑇 is bounded at 𝜌 = 0, ⟹ 𝐵1 = 0
−𝑘𝜆2 𝑡
∴ 𝑢(𝜌, 𝑡) = 𝐴𝑒 𝐽0 (𝜆𝜌) , where 𝐴 = 𝐴1 𝐶1

2
B. C. ∶ 𝑢(1, 𝑡) = 0 = 𝐴𝑒 −𝑘𝜆 𝑡 𝐽0 (𝜆) ⟹ 𝐽0 (𝜆) = 0 , 𝜆 = 𝜆1 , 𝜆2 are positive roots.

Thus, a solution is
2
𝑢(𝜌, 𝑡) = 𝐴𝑒 −𝑘𝜆𝑚 𝑡 𝐽0 (𝜆𝑚 𝜌) , 𝑚 = 1, 2, 3, ⋯

By superposition, a solution is
∞
2
𝑢(𝜌, 𝑡) = ∑ 𝐴𝑚 𝑒 −𝑘𝜆𝑚 𝑡 𝐽0 (𝜆𝑚 𝜌)
𝑚=1

I. C. ∶ 𝑢(𝜌, 0) = 𝐹(𝜌) = ∑ 𝐴𝑚 𝐽0 (𝜆𝑚 𝜌)

𝑚=1

Knowing that
1 1
2 2
𝐴𝑚 = ∫ 𝑥𝐽𝑛 (𝜆𝑚 𝑥)𝑓(𝑥)𝑑𝑥 = 2 ∫ 𝜌𝐹(𝜌)𝐽0 (𝜆𝑚 𝜌)𝑑𝜌
𝐽𝑛+1 2 (𝜆𝑚 ) 0 𝐽1 (𝜆𝑚 ) 0

∞ 1
2 2
∴ 𝑢(𝜌, 𝑡) = ∑ {[ 2 ∫ 𝜌𝐹(𝜌)𝐽0 (𝜆𝑚 𝜌)𝑑𝜌] 𝑒 −𝑘𝜆𝑚 𝑡 𝐽0 (𝜆𝑚 𝜌)}
𝐽1 (𝜆𝑚 ) 0
𝑚=1