You are on page 1of 5

The Little Fourier Transform

Kyle Macfarlan

Introduction

This is transcribed from my class notes on March 28, 2013. The lecturer was Dr. Michael Harold, although I am fairly sure he is using lecutre notes from Dr. Neal Amundson. I am not good at taking notes, much less at keeping them for future reference, but the material covered today was unique and elegant and I wanted to be sure I kept a record of it.

Background

The Little Fourier Transform (LFT) is something of a misnomer, as it was not developed by Fourier himself, but was rather given the name by Dr. Amundson because of its similarity to the more well known Fourier Transforms. The LFT capitalizes on the biorthogonality property of eigenvectors of an n x n matrix A and its transpose AT . Ax = x A y = y xT i xj and xT i yj = 0 for j n 1 (4) In other words, an eigenvector xi is orthogonal to all other eigenvectors of A and all but one eigenvector of AT . This allows for a simple solution to linear equationsalgebraic and dierential. The notation used for the LFT and inverse LFT operators is
T z. f (z) := (yi , z) = yi 1 f (f (z)) := z. T

(1) (2) i = j (3)

=0

(5) (6)

Derivation of the Inverse LFT

Any operator must have an inverse, if it is to be useful in solving problems. Because any arbitrary vector z and be expressed as a linear combination of orthogonal vectors, z can be written as
n

z=
i=1

i xi .

(7)

T Then apply the LFT to this equation (i.e. pre-multiply byyj ) to obtain the following n equations: n T z= yj i=1 T xi i yj

for j = 1, 2, . . . , n

(8)

Because of the biorthogonality property of y and x,


T yj xi =

0 if i = j, = 0 if i = j.

(9)

And thus we obtain


T T yj z = j yj xj

for j = 1, 2, . . . , n

(10) (11) (12)

j = and z =

T yj z T yj xj n T yj z xi . T yj xj

i=1

Therefore, we can dene the Inverse LFT as


n

(z) =
i=1

f (z) xi . T yj xj

(13)

Application of the LFT to Linear Algebraic Equations

The solution of a linear system of equations Az = b is a common, and wellprescribed procedure. Typical solution methods are Cramers rule or GaussJordan Elimination. However, the LFT can be used to solve such systems. 2

Although this method is more tedious and computationally intensive than Gauss-Jordan Elimination, it is a good introduction to solving other linear systems. Az = b f (Az = b)
T T yj Az = yj b

(14) (15) j = 1, 2, . . . , n (16)

Recall from the eigenvalue equation AT yj = j yj . This can be rewritten T T . Substitute this into A = j yj by taking the transpose of both sides as yj the previous result.
T T z = yj b j yj f (z) T f (z) = yj z= T yj b j

j = 1, 2, . . . , n

(17)

j = 1, 2, . . . , n

(18)

Then apply the Inverse LFT to obtain


n

z=
i=1

T b yj xi T j yj xi

(19)

Application of LFT to Linear, First Order ODEs

The LFT can also be used to solve a system of coupled, linear, rst order homogeneous ordinary dierential equations. It can also be used to solve non-homogeneous problems, but that scenario is not considered here. Consider the system of dierential equations and initial conditions dz = Az dt z(t = 0) = z0 . (20) (21)

Note that except in special cases, this system cannot be solved by successive substitution; the equations are coupled. The beauty of the LFT is that it decouples the equations and allows them to be solved directly. 3

dz = Az) dt T dz T yj = yj Az dt d T T yj z = j yj z dt f(

(22) j = 1, 2, . . . , n j = 1, 2, . . . , n (23) (24) (25)

And now it is evident that we have n uncoupled equations. We can now solve them each directly ( dy = cy y = kect ). dt
T yj z = Kj ej t f (z ) T Kj = yj z0

j = 1, 2, . . . , n j = 1, 2, . . . , n

(26) (27)

Here we determined the integration constants Kj from the LFT of the initial conditions.
T T f (z) = yj z = yj z0 ej t) n

j = 1, 2, . . . , n

(28) (29)

z(t) =
i=1

T yi z 0 i t e xi T yi xi

Extension to Complex Domain

Of course, not every problem is in the real domain. Oscillatory behavior is represented by exponentials of complex numbers. It is therefore important to extend this analysis to the complex domain. Thankfully, this is a relatively straightforward analysis. I will skip the derivation, but the crux of the matter is that while we were able to be a little lazy with the concept of the transpose before, we must now use the rigorous adjoint matrix operator A . In the real domain, this is equivalent to the transpose, but in the complex domain it is equivalent to the complex conjugate transpose. A = AT 4 (30)

We can then dene the LFT and inverse LFT as


f (z) = yj z n

j = 1, . . . , n

(31) (32)

z(t) =

yi z 0 i t e xi . yi xi i=1

So the only dierence is that instead of the transpose, we have the complex conjugate transpose of y

Conclusion

The main results of this paper are Equations 29 and 32. This provides a straightforward, if slightly tedious, method for solving a system of coupled ODEs. Although calculating all of the inner products may sound tedious, it is more straightforward than calculating exponential matrices. Equation 29 can also be easily implemented in programs like MATLAB, allowing for even quicker solution time. The Little Fourier Transform is a powerful technique for solving systems of linear equations. While it may be inecient for solving algebraic systems, it provides a relatively simple formula for the solution of systems of linear ODEs. But perhaps its most important application is in the classroom. Students often have trouble conceptually understanding orthogonal vectors and functions and how to apply them in transforms. The LFT is a perfect introduction to this concept by linking it with problems students have already seen.

You might also like