1

3 Transform Techniques for solving PDEs

Let us apply the Separation of Variables technique naively to a problem

posed on an infinite domain, namely, the heat equation

𝑢𝑡 = 𝜅𝑢𝑥𝑥 , − ∞ < 𝑥 < ∞, 𝑡 > 0,

subject to the initial condition

𝑢(𝑥, 0) = 𝑓(𝑥 ) , − ∞ < 𝑥 < ∞.

Physically, the infinite length is a mathematical way of stating that the

rod is so long, the boundary effects are negligible, so no boundary
conditions except the boundedness of solution is considered.

Let 𝑢(𝑥, 𝑡) = 𝑋(𝑥)𝑇(𝑡) and substitute into the PDE to get 𝑋𝑇 ′ = 𝜅𝑋′′ 𝑇.

In turn, it becomes 𝑇′/𝜅𝑇 = 𝑋″/𝑋 = 𝛼 and yields two ODEs:

(1) 𝑋′′ − 𝛼𝑋 = 0 and (2) 𝑇′ − 𝛼𝜅𝑇 = 0.

The solution to (2)

𝑇(𝑡) = 𝐶 𝑒𝑥𝑝( 𝛼𝜅𝑡)

clearly indicates that 𝛼 = −𝜆2 ≤ 0 or that boundedness of solutions to

(1) as 𝑥 → ±∞ also restricts to 𝛼 = −𝜆2 ≤ 0 and thus

𝑋(𝑥) = 𝐴 𝑐𝑜𝑠( 𝜆𝑥) + 𝐵 𝑠𝑖𝑛( 𝜆𝑥).

So, any function of the form

 2

𝑒𝑥𝑝( − 𝜆2 𝜅𝑡)(𝐴 𝑐𝑜𝑠( 𝜆𝑥) + 𝐵 𝑠𝑖𝑛( 𝜆𝑥))

satisfy the PDE for arbitrary A, B and 𝜆. In fact, the eigenvalues 𝜆 form
a continous set rather than discrete.

Accordingly, the superposition is done by integration

∞
𝑢(𝑥, 𝑡) = ∫ 𝑒𝑥𝑝( − 𝜆2 𝜅𝑡)(𝐴(𝜆) 𝑐𝑜𝑠( 𝜆𝑥) + 𝐵(𝜆) 𝑠𝑖𝑛( 𝜆𝑥))𝑑𝜆
0

assuming the integral is suitably convergent.

Note that 𝜆-dependence of 𝐴 = 𝐴(𝜆) & 𝐵 = 𝐵(𝜆) are analogous to

indexing of 𝐴𝑘 & 𝐵𝑘 with the discrete set of eigenvalues 𝜆𝑘 .

In order to determine 𝐴(𝜆) & 𝐵(𝜆), need to satisfy the initial condition
∞
𝑢(𝑥, 0) = 𝑓(𝑥 ) = ∫ (𝐴(𝜆) 𝑐𝑜𝑠( 𝜆𝑥) + 𝐵(𝜆) 𝑠𝑖𝑛( 𝜆𝑥))𝑑𝜆.
0

This is called Fourier Integral representation of 𝑓(𝑥).

Thm.: If 𝑓(𝑥) is piecewise continuous on every finite interval and

absolutely integrable on −∞ < 𝑥 < ∞, then at every 𝑥 at which 𝑓(𝑥)
has right- and left-derivative,
∞
1 + −
(𝑓(𝑥 ) + 𝑓(𝑥 )) = ∫ (𝐴(𝜆) 𝑐𝑜𝑠( 𝜆𝑥) + 𝐵(𝜆) 𝑠𝑖𝑛( 𝜆𝑥))𝑑𝜆
2 0

when
 3

1 ∞
𝐴(𝜆) = ∫ 𝑓(𝑥) 𝑐𝑜𝑠( 𝜆𝑥)𝑑𝑥
𝜋 −∞

and
1 ∞
𝐵(𝜆) = ∫ 𝑓(𝑥) 𝑠𝑖𝑛( 𝜆𝑥)𝑑𝑥 .
𝜋 −∞

Def.: A function 𝑓(𝑥) is said to be absolutely integrable on the interval

−∞ < 𝑥 < ∞ if
∞
∫ |𝑓(𝑥)|𝑑𝑥 < ∞
−∞

i.e. it converges.

A more compact form of the Fourier Integral representation can be

obtained by introducing the complex exponential:

𝑒 𝑖𝜆𝑥 = 𝑐𝑜𝑠( 𝜆𝑥) + 𝑖 𝑠𝑖𝑛( 𝜆𝑥)

where 𝑖 = √−1, into the integral representation as follows:

∞
1 + −
(𝑓(𝑥 ) + 𝑓(𝑥 )) = ∫ ( 𝐴(𝜆) 𝑐𝑜𝑠(
⏟ 𝜆𝑥) + 𝐵(𝜆) 𝑠𝑖𝑛(
⏟ 𝜆𝑥) )𝑑𝜆
2 0 1 1
(𝑒 𝑖𝜆𝑥 +𝑒 −𝑖𝜆𝑥 ) (𝑒 𝑖𝜆𝑥 −𝑒 −𝑖𝜆𝑥 )
2 2𝑖

and after some algebraic manipulations, it becomes

∞
1
(𝑓(𝑥 + ) + 𝑓(𝑥 − )) = ∫ 𝐶(𝜆)𝑒 𝑖𝜆𝑥 𝑑𝜆
2 −∞
 4

where
1
(𝐴(𝜆) − 𝑖𝐵(𝜆)) 𝜆>0
𝐶(𝜆) = { 2
1
(𝐴(−𝜆) + 𝑖𝐵(−𝜆)) 𝜆<0
2
that can be simplified by substituting for 𝐴(𝜆) & 𝐵(𝜆) from the
Theorem to get
1 ∞
𝐶 (𝜆) = ∫ 𝑓 (𝑥 )𝑒 −𝑖𝜆𝑥 𝑑𝑥 .
2𝜋 −∞

Def.: The Fourier Transform of a function 𝑓(𝑥) and its inverse are
defined by the pair
∞
ℑ[𝑓(𝑥)] ≡ 𝑓̂(𝜔) = ∫ 𝑓(𝑥)𝑒 −𝑖𝜔𝑥 𝑑𝑥
−∞

and
1 ∞
ℑ [𝑓̂(𝜔)] ≡ 𝑓(𝑥 ) =
−1
∫ 𝑓̂(𝜔)𝑒 𝑖𝜔𝑥 𝑑𝜔.
2𝜋 −∞

2
e.g. Find the Fourier Transform of the Gaussian 𝑓(𝑥 ) = 𝑒 −𝑘𝑥 , 𝑘 > 0:
∞ ∞
−𝑘𝑥 2 2
𝑓̂(𝜔) = ∫ 𝑒 𝑒 −𝑖𝜔𝑥 𝑑𝑥 = ∫ 𝑒 −𝑘𝑥 (𝑐𝑜𝑠( 𝜔𝑥 ) − 𝑖 𝑠𝑖𝑛( 𝜔𝑥))𝑑𝑥
−∞ −∞
∞
2
= ∫ 𝑒 −𝑘𝑥 𝑐𝑜𝑠( 𝜔𝑥)𝑑𝑥 .
−∞
 5

Note that
∞
𝑑 2
𝑓̂(𝜔) = − ∫ 𝑥𝑒 −𝑘𝑥 𝑠𝑖𝑛( 𝜔𝑥)𝑑𝑥
𝑑𝜔 −∞
∞ ∞
1 −𝑘𝑥 2 1 2
=[ 𝑒 𝑠𝑖𝑛( 𝜔𝑥)] − 𝜔 ∫ 𝑒 −𝑘𝑥 𝑐𝑜𝑠( 𝜔𝑥)𝑑𝑥
⏟2𝑘 −∞ 2𝑘 −∞
=0

i.e.
𝑑 1
𝑓̂(𝜔) = − 𝜔𝑓̂(𝜔)
𝑑𝜔 2𝑘
with
∞
−𝑘𝑥 2
𝜋 𝜋 −𝜔 2
𝑓̂(0) = ∫ 𝑒 𝑑𝑥 = √ ⇒ 𝑓̂(𝜔) = √ 𝑒 4𝑘 ,
−∞ 𝑘 𝑘

that is also Gaussian in 𝜔 (in Fourier space).

In the presence of (assumed) even or odd symmetry, we get Fourier-

Cosine and -Sine Transforms:

Def.: The Fourier Cosine Transform pair of a function 𝑓(𝑥) on

0 < 𝑥 < ∞ are

∞
ℑ𝐶 [𝑓(𝑥 )] ≡ 𝑓̂𝐶 (𝜔) = ∫ 𝑓(𝑥) 𝑐𝑜𝑠( 𝜔𝑥)𝑑𝑥 ,
0

2 ∞
ℑ−1 ̂
𝐶 [𝑓𝐶 (𝜔 )] ≡ 𝑓(𝑥 ) = ∫ 𝑓̂𝐶 (𝜔) 𝑐𝑜𝑠( 𝜔𝑥)𝑑𝜔.
𝜋 0
 6

Def.: The Fourier Sine Transform pair of a function 𝑓(𝑥) on 0 < 𝑥 < ∞
are
∞
ℑ𝑆 [𝑓(𝑥 )] ≡ 𝑓̂𝑆 (𝜔) = ∫ 𝑓(𝑥) 𝑠𝑖𝑛( 𝜔𝑥)𝑑𝑥 ,
0
∞
2
ℑ𝑆−1 [𝑓̂𝑆 (𝜔)] ≡ 𝑓(𝑥 ) = ∫ 𝑓̂𝑆 (𝜔) 𝑠𝑖𝑛( 𝜔𝑥)𝑑𝜔.
𝜋 0

Note that for 𝑓𝑒 (𝑥 ) = 𝑓(|𝑥 |) even extension of 𝑓(𝑥) to −∞ < 𝑥 < ∞,

∞ ∞ ∞
−𝑖𝜔𝑥 −𝑖𝜔𝑥
ℑ[𝑓𝑒 (𝑥 )] = ∫ 𝑓𝑒 (𝑥 )𝑒 𝑑𝑥 = ∫ 𝑓(𝑥)𝑒 𝑑𝑥 + ∫ 𝑓(𝑥)𝑒 𝑖𝜔𝑥 𝑑𝑥
−∞ 0 0
∞
= 2 ∫ 𝑓(𝑥) 𝑐𝑜𝑠( 𝜔𝑥)𝑑𝑥 = 2ℑ𝐶 [𝑓(𝑥 )].
0

Similarly, for 𝑓𝑜 (𝑥 ) = sign(𝑥)𝑓(|𝑥 |) odd extension of 𝑓(𝑥) to −∞ <

𝑥 < ∞,

ℑ[𝑓𝑜 (𝑥 )] = −2𝑖ℑ𝑆 [𝑓(𝑥 )].

e.g. Find the Fourier-Sine and Cosine Transform of 𝑓(𝑥 ) = 𝑒 −𝑘𝑥 ,

for 𝑘 > 0:
∞ ∞
𝑓̂𝐶 (𝜔) = ∫ 𝑒 −𝑘𝑥
𝑐𝑜𝑠( 𝜔𝑥)𝑑𝑥 = 𝑅𝑒 {∫ 𝑒 −𝑘𝑥 𝑒 −𝑖𝜔𝑥 𝑑𝑥 }
0 0
∞
1 ( ) 𝑘
= 𝑅𝑒 {− 𝑒 − 𝑘+𝑖𝜔 𝑥 } = 2 2
.
𝑘 + 𝑖𝜔 0 𝑘 + 𝜔

Similarly,
 7

∞
1 𝜔
̂
𝑓𝑆 (𝜔) = ∫ 𝑒 −𝑘𝑥
𝑠𝑖𝑛( 𝜔𝑥)𝑑𝑥 = − 𝐼𝑚 { }= 2 .
𝑘 + 𝑖𝜔 𝑘 + 𝜔 2
0

Note the symmetry in Fourier-Sine and -Cosine transforms that the

above results may be used in reverse by replacing 𝜔 ↔ 𝑥 and
introducing the scaling factor 2/𝜋, i.e.
𝑘 𝜋 −𝑘𝜔 𝑥 𝜋 −𝑘𝜔
ℑ𝐶 [ 2 ]= 𝑒 and ℑ𝑆 [ 2 ]= 𝑒 .
𝑘 + 𝑥2 2 𝑘 + 𝑥2 2

Some Properties

(1) ℑ [𝑓′(𝑥 )] = 𝑖𝜔 ℑ [𝑓(𝑥 )],

ℑ [𝑓′′(𝑥 )] = −𝜔2 ℑ [𝑓(𝑥)],

ℑ𝐶 [𝑓′(𝑥 )] = 𝜔 ℑ𝑆 [𝑓(𝑥 )] − 𝑓 (0+ ),

ℑ𝐶 [𝑓′′(𝑥 )] = −𝜔2 ℑ𝐶 [𝑓(𝑥 )] − 𝑓′(0+ ),

ℑ𝑆 [𝑓′(𝑥 )] = −𝜔ℑ𝐶 [𝑓(𝑥 )],

ℑ𝑆 [𝑓′′(𝑥 )] = −𝜔2 ℑ𝑆 [𝑓 (𝑥 )] + 𝜔𝑓(0+ ).

(2) Convolution:
∞
ℑ−1 [𝑓̂(𝜔)𝑔̂(𝜔)] = ∫−∞ 𝑓(𝜏)𝑔(𝑥 − 𝜏)𝑑𝜏.

In the case, 𝑓 and 𝑔 are even-extended functions, then

 8

1 1 ∞
𝐻𝑒 (𝑥 ) = [𝐻(𝑥 ) + 𝐻 (−𝑥 )] = ∫ 𝑓(𝜏)[𝑔(𝑥 − 𝜏) + 𝑔(𝑥 + 𝜏)] 𝑑𝜏
2 2 0

is an even-extended function where

∞
𝐻(𝑥 ) ≡ ∫ 𝑓(𝜏)𝑔(𝑥 − 𝜏)𝑑𝜏.
0

It follows using ℑ[𝐻𝑒 (𝑥 )] = 2ℑ𝐶 [𝐻(𝑥 )] that

1 ∞
ℑ−1 ̂
𝐶 [𝑓𝐶 (𝜔)𝑔
̂𝐶 (𝜔)] = ∫ 𝑓(𝜏)[𝑔(𝑥 − 𝜏) + 𝑔(𝑥 + 𝜏)]𝑑𝜏.
2 0

By similar arguments, it can be shown that

1 ∞
ℑ𝑆−1 [𝑓̂𝑆 (𝜔)𝑔̂𝐶 (𝜔)] = ∫ 𝑓(𝜏)[𝑔(𝑥 − 𝜏) − 𝑔(𝑥 + 𝜏)]𝑑𝜏,
2 0

where 𝑓 and 𝑔 are odd- and even-extended, respectively.

e.g. Consider, again, the heat equation

𝑢𝑡 = 𝜅𝑢𝑥𝑥 , − ∞ < 𝑥 < ∞, 𝑡>0

subject to the initial condition:

𝑢(𝑥, 0) = 𝑓(𝑥 ) , − ∞ < 𝑥 < ∞.

Taking the Fourier transform of the equation, we get

2𝑡 2
𝑢̂𝑡 = −𝜅𝜔2 𝑢̂ ⇒ 𝑢̂(𝜔, 𝑡) = 𝐶 (𝜔)𝑒 −𝜅𝜔 ⇒ 𝑢̂(𝜔, 𝑡) = 𝑓̂(𝜔)𝑒 −𝜅𝜔 𝑡 .

where 𝑢̂(𝜔, 𝑡) = ℑ[𝑢(𝑥, 𝑡)] and 𝑢̂(𝜔, 0) = 𝑓̂(𝜔).

Now, using the convolution property and

 9

1 𝑥2
−𝜅𝜔2 𝑡 −
ℑ−1 [𝑒 ]= 𝑒 4𝜅𝑡
2√𝜅𝜋𝑡
we get
1 ∞ 2
𝑢(𝑥, 𝑡) = ℑ −1 [
𝑢̂(𝜔, 𝑡)] = ∫ 𝑓̂(𝜔)𝑒 −𝜅𝜔 𝑡 𝑒 −𝑖𝜔𝑥 𝑑𝑥
2𝜋 −∞
∞ (𝑥−𝜏)2
1 −
= ∫ 𝑓 (𝜏)𝑒 4𝜅𝑡 𝑑𝜏.
2√𝜅𝜋𝑡 −∞

Case 1 𝑓(𝑥) = 𝑈0 :
∞ (𝑥−𝜏)2 ∞
1 − 1 2
𝑢(𝑥, 𝑡) = 𝑈0 ∫ 𝑒 4𝜅𝑡 𝑑𝜏 = 𝑈0 ∫ 𝑒 −𝑠 𝑑𝑠 = 𝑈0 .
2√𝜅𝜋𝑡 −∞ √𝜋 −∞

Note that even though 𝑓(𝑥) = 𝑈0 is not absolutely integrable, i.e. it

does not have Fourier tranform, the convolution procedure gives the
correct answer.

Case 2 𝑓(𝑥) = 𝑈0 𝐻(𝑥) where 𝐻(𝑥) is the unit-step function.

𝑥
∞ (𝑥−𝜏)2
1 1 2√𝜅𝑡 2
𝑢(𝑥, 𝑡) = 𝑈0 ∫ 𝑒 − 4𝜅𝑡 𝑑𝜏 = 𝑈0 ∫ 𝑒 −𝑠 𝑑𝑠
2√𝜅𝜋𝑡 0 √𝜋 −∞
𝑥
0
1 2 2√𝜅𝑡 2
= 𝑈0 {∫ 𝑒 −𝑠 𝑑𝑠 + ∫ 𝑒 −𝑠 𝑑𝑠}
√𝜋 −∞ 0

1 𝑥
= 𝑈0 [1 + erf ( )].
2 2√𝜅𝑡
where
 10

𝑥
2 2
erf(𝑥 ) = ∫ 𝑒 −𝑠 𝑑𝑠 ⟹ −1 < erf(𝑥 ) < 1.
√𝜋 0

Note that heat initially concentrated in one half (𝑥 > 0) diffuses

immediately (𝑡 > 0) into the other half (𝑥 < 0) since 𝑢(𝑥, 𝑡) is positive
everywhere for every 𝑡 > 0. This is the infinite speed of propagation
characteristic of Parabolic PDEs.

Case 3 𝑓(𝑥) = 𝑥(𝐿 − 𝑥) for 0 < 𝑥 < 𝐿 and vanishes everwhere else.
𝐿 (𝑥−𝜏)2
1 −
𝑢(𝑥, 𝑡) = ∫ 𝜏(𝐿 − 𝜏)𝑒 4𝜅𝑡 𝑑𝜏 .
2√𝜅𝜋𝑡 0

Check the Integral Tables for an explicit form.

In these examples, the advantage of the transform technique is apparent

when the data (external conditions) of the problem (PDE) is piecewise-
defined.

e.g. Consider the vibration problem:

𝑢𝑡𝑡 = 𝑐 2 𝑢𝑥𝑥 , − ∞ < 𝑥 < ∞, 𝑡 > 0,

subject to the initial conditions:

𝑢(𝑥, 0) = 𝑓(𝑥 ) , − ∞ < 𝑥 < ∞,

𝑢𝑡 (𝑥, 0) = 𝑔(𝑥 ) , − ∞ < 𝑥 < ∞.

 11

Taking the Fourier transform of the equation, we get

𝑢̂𝑡𝑡 = −𝑐 2 𝜔2 𝑢̂ ⇒ 𝑢̂(𝜔, 𝑡) = 𝐴(𝜔) 𝑐𝑜𝑠( 𝑐𝜔𝑡) + 𝐵(𝜔) 𝑠𝑖𝑛( 𝑐𝜔𝑡).

Since

𝑢̂(𝜔, 0) = 𝑓̂(𝜔) = 𝐴(𝜔), and 𝑢̂𝑡 (𝜔, 0) = 𝑔̂(𝜔) = 𝑐𝜔𝐵(𝜔),

we have
1
𝑢̂(𝜔, 𝑡) = 𝑓̂(𝜔) ⏟
𝑐𝑜𝑠( 𝑐𝜔𝑡) + 𝑔̂(𝜔) ⏟𝑠𝑖𝑛( 𝑐𝜔𝑡) .
1 𝑖𝑐𝜔𝑡 −𝑖𝑐𝜔𝑡
𝑐𝜔 1
(𝑒 +𝑒 ) (𝑒 𝑖𝑐𝜔𝑡 −𝑒 −𝑖𝑐𝜔𝑡 )
2 2𝑖

Now,

−1 [
1 ∞
𝑢(𝑥, 𝑡) = ℑ 𝑢̂(𝜔, 𝑡)] = ∫ 𝑢̂(𝜔, 𝑡) 𝑒 𝑖𝜔𝑥 𝑑𝜔
2𝜋 −∞
1 ∞
= ∫ 𝑓̂(𝜔)(𝑒 𝑖𝜔(𝑥+𝑐𝑡) + 𝑒 𝑖𝜔(𝑥−𝑐𝑡) ) 𝑑𝜔
4𝜋 −∞
1 ∞ 1
+ ∫ 𝑔̂(𝜔) (𝑒 𝑖𝜔(𝑥+𝑐𝑡) − 𝑒 𝑖𝜔(𝑥−𝑐𝑡) ) 𝑑𝜔
4𝜋 −∞ 𝑖𝑐𝜔
1 1 𝑥+𝑐𝑡
= (𝑓(𝑥 + 𝑐𝑡) + 𝑓(𝑥 − 𝑐𝑡)) + ∫ 𝑔(𝜏) 𝑑𝜏,
2 2𝑐 𝑥−𝑐𝑡

where we used
𝑥
1
ℑ [∫ 𝑔(𝜏)𝑑𝜏] = 𝑔̂(𝜔).
−∞ 𝑖𝜔

This is called D’Alembert solution.

 12

e.g. Consider the Laplace equation:

𝑢𝑥𝑥 + 𝑢𝑦𝑦 = 0, 𝑥 > 0, 𝑦 > 0,

subject to the boundary conditions:

𝑢(0, 𝑦) = 𝑔(𝑦), 𝑦 > 0,

𝑢𝑦 (𝑥, 0) = 𝑓 (𝑥 ), 𝑥 > 0.

First, decompose 𝑢 = 𝑣 + 𝑤 such that

𝑣𝑥𝑥 + 𝑣𝑦𝑦 = 0, 𝑥 > 0, 𝑦 > 0,

𝑣 (0, 𝑦) = 𝑔(𝑦), 𝑦 > 0,

𝑣𝑦 (𝑥, 0) = 0, 𝑥 > 0,

and

𝑤𝑥𝑥 + 𝑤𝑦𝑦 = 0, 𝑥 > 0, 𝑦 > 0,

𝑤(0, 𝑦) = 0, 𝑦 > 0,

𝑤𝑦 (𝑥, 0) = 𝑓 (𝑥 ), 𝑥 > 0.
 13

In half-space problems, it is important to decide on the use of the type

of transform, Fourier-Cosine or Fourier-Sine.

For this purpose, consider the following demonstration in this respect:

(1) 𝑋 ″ + 𝜔2 𝑋 = 0, 𝑥 > 0, 𝑋(0) = 0 ⇒ 𝑋(𝑥) = 𝑠𝑖𝑛( 𝜔𝑥),

(2) 𝑋 ″ + 𝜔2 𝑋 = 0, 𝑥 > 0, 𝑋 ′ (0) = 0 ⇒ 𝑋(𝑥) = 𝑐𝑜𝑠( 𝜔𝑥).

Thus, problem type (1) is suitable for Fourier-Sine transform, while

problem type (2) for Fourier-Cosine transform.

It follows from the respective homogeneous boundary data, namely

𝑋(0) = 0 and 𝑋 ′ (0) = 0, and the Transform Properties (1) above that:
=0
⏞(0+ ) = −𝜔2 ℑ [𝑋],
ℑ𝑆 [𝑋 ″ ] = −𝜔2 ℑ𝑆 [𝑋] + 𝜔 𝑋 𝑆

ℑ𝐶 [𝑋 ″ ] = −𝜔2 ℑ𝐶 [𝑋] − 𝑋
⏟′ (0+ ) = −𝜔2 ℑ𝐶 [𝑋].
=0

Thus, above, Fourier-Cosine transform in 𝑦 is suitable to solve for

𝑣 = 𝑣(𝑥, 𝑦), while Fourier-Sine transform in 𝑥 is suitable to solve for
𝑤 = 𝑤(𝑥, 𝑦).

Note: Application of Fourier Cosine transform for a function 𝑓(𝑥)

requires 𝑓 ′ (0) is to be known, while application of Fourier Sine
transform requires 𝑓 (0) is to be known.
 14

Applying Fourier-Cosine transform in 𝑦 to solve for 𝑣 = 𝑣(𝑥, 𝑦), and

recalling that

ℑ𝐶 [𝑣𝑦𝑦 ] = −𝜔2 ℑ𝐶 [𝑣] − 𝑣𝑦 (𝑥, 0+ ),

we get

𝑣̂𝑥𝑥 − 𝜔2 𝑣̂ = 0 ⇒ 𝑣̂(𝑥, 𝜔) = 𝐴(𝜔)𝑒 𝜔𝑥 + 𝐵(𝜔)𝑒 −𝜔𝑥 .

For boundedness as 𝑥 → ∞ ⇒ 𝐴(𝜔) = 0,

𝑣̂ (0, 𝜔) = 𝑔̂(𝜔) ⇒ 𝐵(𝜔) = 𝑔̂(𝜔) ⇒ 𝑣̂(𝑥, 𝜔) = 𝑔̂(𝜔)𝑒 −𝜔𝑥 .

2 ∞
⇒ 𝑣(𝑥, 𝑦) = ℑ−1
𝐶 [𝑣
̂(𝑥, 𝜔)] = ∫ 𝑔̂(𝜔)𝑒 −𝜔𝑥 𝑐𝑜𝑠( 𝜔𝑥)𝑑𝜔
𝜋 0
𝑥 ∞ 1 1
= ∫ 𝑔(𝜏) ( 2 + ) 𝑑𝜏
𝜋 0 𝑥 + (𝑦 − 𝜏)2 𝑥 2 + (𝑦 + 𝜏)2

where we have used the Convolution Property (2) above and

2 𝑥
ℑ−1
𝐶 [𝑒
−𝜔𝑥 ]
= .
𝜋 𝑥2 + 𝑦2

Applying Fourier-Sine transform in 𝑥 to solve for 𝑤 = 𝑤(𝑥, 𝑦), and

recalling that

ℑ𝑆 [𝑤𝑥𝑥 ] = −𝜔2 ℑ𝑆 [𝑤] + 𝜔 𝑤 (0+ , 𝑦),

we get
 15

̂𝑦𝑦 − 𝜔2 𝑤
𝑤 ̂ (𝜔, 𝑦) = 𝐴(𝜔)𝑒 𝜔𝑦 + 𝐵(𝜔)𝑒 −𝜔𝑦 .
̂ =0 ⇒ 𝑤

For boundedness as 𝑦 → ∞ ⇒ 𝐴(𝜔) = 0,

𝑓̂(𝜔) 1
̂𝑦 (𝜔, 0) = 𝑓̂(𝜔) ⇒ 𝐵 (𝜔) = −
𝑤 ̂ (𝜔, 𝑦) = − 𝑓̂(𝜔)𝑒 −𝜔𝑦 .
⇒ 𝑤
𝜔 𝜔

∞
2 1
( ) −1 [ (
⇒ 𝑤 𝑥, 𝑦 = ℑ𝑆 𝑤 )]
̂ 𝜔, 𝑦 = − ∫ 𝑓̂(𝜔)𝑒 −𝜔𝑦 𝑠𝑖𝑛( 𝜔𝑥)𝑑𝜔
𝜋 0 𝜔
1 ∞ 𝑥−𝜏 𝑥+𝜏
= ∫ 𝑓(𝜏) (atan ( ) − atan ( )) 𝑑𝜏
𝜋 0 𝑦 𝑦

where we have used the Convolution Property (2) above and

1 −𝜔𝑦 2 𝑥
ℑ−1
𝑠 [ 𝑒 ] = atan ( ).
𝜔 𝜋 𝑦

Thus, 𝑢(𝑥, 𝑦) = 𝑣(𝑥, 𝑦) + 𝑤(𝑥, 𝑦).

Note that the Robin type boundary condition involves

(3) 𝑋 ″ + 𝜔2 𝑋 = 0, 𝑥 > 0, 𝑋 ′ (0) + 𝛾𝑋(0) = 0 ⇒

⇒ 𝑋(𝑥) = 𝜔 𝑐𝑜𝑠( 𝜔𝑥) + 𝛾 𝑠𝑖𝑛( 𝜔𝑥).

Thus, solving PDEs in half-space (𝑥 > 0) with Robin-type boundary

condition requires a new transform-pair to be defined as follows:
 16

∞
𝜔 𝑐𝑜𝑠( 𝜔𝑥) + 𝛾 𝑠𝑖𝑛( 𝜔𝑥)
ℑ𝑅 [𝑓(𝑥 )] ≡ 𝑓̂𝑅 (𝜔) = ∫ 𝑓(𝑥) 𝑑𝑥 ,
0 √𝜔 2 + 𝛾2

2 ∞ 𝜔 𝑐𝑜𝑠( 𝜔𝑥) + 𝛾 𝑠𝑖𝑛( 𝜔𝑥)

ℑ−1 ̂
𝑅 [𝑓𝑅 (𝜔)]
̂
≡ 𝑓 (𝑥 ) = ∫ 𝑓𝑅 (𝜔) 𝑑𝜔.
𝜋 0 2
√𝜔 + 𝛾 2

Laplace Transform is another alternative for problems on unbounded

domains. Let us demonstrate on some sample problems:

e.g. Consider the heat equation

𝑢𝑡 = 𝜅𝑢𝑥𝑥 , 𝑥 > 0, 𝑡 > 0,

subject to the boundary condition:

𝑢(0, 𝑡) = 𝑈0 , 𝑡 > 0,

the initial condition:

𝑢(𝑥, 0) = 0, 𝑥 > 0.

Taking the Laplace transform of the equation in t, we get

𝑠 𝑠
√ 𝑥 −√ 𝑥
𝑠𝑢̑ = 𝜅𝑢̑ 𝑥𝑥 ⇒ 𝑢̑ (𝑥, 𝑠) = 𝐴(𝑠)𝑒 𝜅 + 𝐵 (𝑠)𝑒 𝜅 .

For boundedness as 𝑥 → ∞ ⇒ 𝐴(𝑠) = 0,

1 1 1 𝑠
−√ 𝑥
( )
𝑢̑ 0, 𝑠 = 𝑈0 ⇒ ( )
𝐵 𝜔 = 𝑈0 ⇒ 𝑢̑ (𝑥, 𝑠) = 𝑈0 𝑒 𝜅 ,
𝑠 𝑠 𝑠
 17

1 𝑠 𝑥
−√ 𝑥
⇒ 𝑢(𝑥, 𝑡) = ℒ −1 [𝑢̑ (𝑥, 𝑠)] = 𝑈0 ℒ −1 [ 𝑒 𝜅 ] = 𝑈0 erfc ( ),
𝑠 2√𝜅𝑡

where
∞
2 2
erfc(𝑡) = 1 − erf(𝑡) = ∫ 𝑒 −𝜏 𝑑𝜏
√𝜋 𝑡

and
𝑎 1
ℒ [erfc ( )] = 𝑒 −𝑎√𝑠
2√𝑡 𝑠

e.g. Consider the semi-infinite string falling under gravity while one end
being fixed:

𝑢𝑡𝑡 = 𝑐 2 𝑢𝑥𝑥 − 𝑔, 𝑥 > 0, 𝑡 > 0,

subject to the boundary condition

𝑢(0, 𝑡) = 0, 𝑡 > 0,

the initial conditions

𝑢(𝑥, 0) = 𝑢𝑡 (𝑥, 0) = 0, 𝑥 > 0.

Taking the Laplace transform of the equation in t, we get

2 2
1 𝑠𝑥
−
𝑠𝑥 1
𝑠 𝑢̑ = 𝑐 𝑢̑ 𝑥𝑥 − 𝑔 ⇒ 𝑢̑ (𝑥, 𝑠) = 𝐴(𝑠) 𝑒𝑐 + 𝐵(𝑠) 𝑒 𝑐 − 𝑔
𝑠 𝑠3

For boundedness as 𝑥 → ∞ ⇒ 𝐴(𝑠) = 0,

 18

1 𝑔 −
𝑠𝑥
𝑢̑ (0, 𝑠) = 0 ⇒ 𝐵(𝜔) = 3 𝑔 ⇒ 𝑢̑ (𝑥, 𝑠) = − 3 (1 − 𝑒 𝑐 ).
𝑠 𝑠

𝑔 2 𝑔 𝑥 2 𝑥
⇒ 𝑢(𝑥, 𝑡) = − 𝑡 + (𝑡 − ) 𝐻 (𝑡 − ).
2 2 𝑐 𝑐

The shape of the string at 𝑡 = 𝑡0 .

Note that this problem can not be solved by Fourier-Sine transform

because a constant function 𝑔 is not absolutely integrable on 𝑥 > 0.