Applications and Review of Fourier Transform/Series: Solving Linear Partial Differential Equations (PDE's)
Applications and Review of Fourier Transform/Series: Solving Linear Partial Differential Equations (PDE's)
Applications and Review of Fourier Transform/Series: Solving Linear Partial Differential Equations (PDE's)
(Reference – The Fourier Transform and its Application, second edition, R.N. Bracewell,
McGraw-Hill Book Co., New York, 1978.)
Fourier analysis is a fundamental tool used in all areas of science and engineering. The fast
fourier transform (FFT) algorithm is remarkably efficient for solving large problems. Nearly
every computing platform has a library of highly-optimized FFT routines. In the field of Earth
science, fourier analysis is used in the following areas:
In this remote sensing course we will use fourier analysis to understand and evaluate apertures
(antennas and telescopes) as well as to filter images.
Fourier analysis deals with complex numbers so perhaps it is time to dust off your book on
advanced calculus. Here is a very brief review of the things you’ll need. A complex number
z = x + iy is composed of real and imaginary numbers. Remember i = −1 . Functions can have
real and imaginary components as well. For example a general, complex-valued function of a
real variable is f (x) = u(x) + iw(x) . The most important complex-valued function for this
discussion is the complex exponential function e iθ = cos θ €+ isin θ . After a little algebra it is easy
€ iθ −iθ iθ −iθ
e +e e −e
to show that cosθ = and sin θ = . This is the level of math you’ll need to
€ 2 2i
understand fourier analysis. €
€ €
2
∞
−i2 πkx
F(k) = ∫ f(x)e dx F(k) = ℑ[ f(x)] - forward transform
−∞
∞
i2 πkx
f(x) = ∫ F(k)e dk f(x) = ℑ−1[ F(k)] - inverse transform
−∞
where x is distance and k is wavenumber where k = 1/λ and λ is wavelength. These equations
are more commonly written in terms of time t and frequency ν where ν = 1/T and T is the period.
∞ ∞
−i 2 π (k ⋅x )
F(k) = ∫ ∫ f (x)e d 2x F(k) = ℑ2 [ f (x)]
−∞−∞
∞ ∞
i2π (k⋅x )
f (x) = ∫ ∫ F(k)e d 2k f (x) = ℑ2−1 [ F(k)]
−∞−∞
where x = (x, y) is the position vector, k = (kx, ky) is the wavenumber vector, and
(k . x) = kx x + ky y.
The next two page show some examples of fourier transform pairs. These figures were taken
from Bracewell [1978].
3
4
5
Any function f(x) can be decomposed into odd O(x) and even E(x) components.
1 1
E(x) = [ f(x) + f(− x)] O(x) = [ f(x) − f(− x)]
2 2
∞
−i2 πkx
F(k) = ∫ f(x)e dx
−∞
e − iθ = cosθ − isin θ
∞ ∞
∞ ∞
You have probably seen fourier cosine and sine transforms, but it is better to use the complex
exponential form.
6
7
The following are some important properties of fourier transforms that you should derive for
yourself at least once. You’ll find derivations in Bracewell. Once you have derived and
understand these properties, you can treat them as tools. Very complicated problems can be
simplified using these tools. For example, when solving a linear partial differential equation, one
uses the derivative property to reduce the differential equation to an algebraic equation.
1 k
similarity property ℑ[ f(ax)] = F
| a | a
df
differentiation property ℑ = i2πk F(k)
dx
∞
convolution property ℑ ∫ f (u)g( x -u )du = F(k)G( k)
-∞
∞ 2 ∞ 2
Note: These properties are equally valid in 2-dimensions or even n-dimensions. The properties
also apply to discrete data. See Chapter 18 in Bracewell.
8
Fourier series
The coefficients of the 2-dimensional Fourier series are computed by the following integration.
L W
1 m n
Fnm = ∫ ∫ f (x, y)exp −i2π L x + W y dydx
LW o o
The function is reconstructed by the following summations over the fourier coefficients.
∞ ∞
m
m n
f (x, y) = ∑ ∑F
n= −∞ m= −∞
n exp i2π x + y
L W
The finite size of the area leads to a discrete set of wavenumbers kx = m/L, ky = n/W and a
discrete set of fourier coefficients Fnm. In addition to the finite size of the area, geophysical data
commonly have a characteristic sampling interval Δx and Δy.
The Nyquist wavenumbers is kx = 1/(2 Δx) and ky = 1/(2 Δy) so there is a finite set of fourier
coefficients -I/2 < m < I/2 and -J/2 < n < J/2. Recall the trapesoidal rule of integration.
L I−1
L
L I −1
I∑
∫ f (x)dx ≅ f (xi )
o i =0
9
m 1 I −1 J −1
j
m n
F =
n
IJ
∑∑ f i exp −i2π i + j
I J
i =o j = o
I/2 J/2 i j
fi j = ∑ ∑F n
m
exp i2π m + n
n= − I / 2 m= − J / 2 I J
The forward and inverse discrete fourier transforms are almost identical sums so one can use the
same computer code for both operations.