5.

8 Discrete Fourier Transform

357

Furthermore, the fact that k 1 + k + 2k + + (n2)k + (n1)k = k + 2k + + (n1)k + 1 implies 1 + k + 2k + + (n1)k 1 k = 0 and, consequently, whenever k = 1. (5.8.2)

1 + k + 2k + + (n1)k = 0

Fourier Matrix
The n n matrix whose (j, k)-entry is jk = jk for 0 j, k n1 is called the Fourier matrix of order n, and it has the form 1 1 Fn = 1 . . . 1 2 . . . 1 2 4 . . . n2 .. . 1 n1 n2 . . . . nn

1 n1

Note. Throughout this section entries are indexed from 0 to n 1. For example, the upper left-hand entry of Fn is considered to be in the (0, 0) position (rather than the (1, 1) position), and the lower righthand entry is in the (n 1, n 1) position. When the context makes it clear, the subscript n on Fn is omitted. The Fourier matrix is a special case of the Vandermonde matrix introduced in Example 4.3.4. Using (5.8.1) and (5.8.2), we see that the inner product of any two columns in Fn , say, the rth and sth , is
n1 n1 n1
50

F Fs = r
j=0

jr js =
j=0

jr js =
j=0

j(sr) = 0.

In other words, the columns in Fn are mutually orthogonal. Furthermore, each column in Fn has norm n because
n1 n1

Fk
50

2 2

=
j=0

| jk |2 =
j=0

1 = n,

Some authors dene the Fourier matrix using powers of rather than powers of , and some include a scalar multiple 1/n or 1/ n. These dierences are supercial, and they do not aect the basic properties. Our denition is the discrete counterpart of the integral operator x(t)ei2f t dt that is usually taken as the denition of the continuous Fourier F (f ) =

transform.