Академический Документы
Профессиональный Документы
Культура Документы
5. Orthogonal matrices
orthogonal matrices
5-1
Orthonormal vectors
Example
0 1 1
1 1
0 , 1 , 1
1 2 0 2 0
A Rmn has orthonormal columns if its Gram matrix is the identity matrix:
T
T
A A = a1 a2 an a1 a2 an
aT1 a1 aT1 a2 T
a1 an
aT2 a1 aT2 a2 aT2 an
= .. .. ..
..
. . . .
aTn a1 aTn a2 aTn an
1 0 0
0 1 0
= .. .. . . ..
. . . .
0 0 1
(Ax)T (Ay) = xT AT Ay = xT y
preserves norms:
1/2
T
kAxk = (Ax) (Ax) = (xT x)1/2 = kxk
preserves angles:
T T
(Ax) (Ay) x y
(Ax, Ay) = arccos = arccos = (x, y)
kAxkkAyk kxkkyk
AT A = I
Ax = 0 = AT Ax = x = 0
orthogonal matrices
Orthogonal matrix
T
A A=I
= AAT = I
A is square
Aii = 1, Aij = 0 if j 6= i
permutation on {1, 2, 3, 4}
(1, 2, 3, 4) = (2, 4, 1, 3)
(1, 2, 3, 4) = (3, 1, 4, 2)
Rotation in a plane
Ax
cos sin
A= x
sin cos
x y = x (aT x)a
= (I aaT )x
H y
z = y + (y x)
= (I 2aaT )x
line through a and origin z = Ax
(see page 2-44)
Orthogonal matrices 5-10
Product of orthogonal matrices
if A1, . . . , Ak are orthogonal matrices and of equal size, then the product
A = A1 A2 Ak
is orthogonal:
AT A = (A1A2 Ak )T (A1A2 Ak )
= ATk AT2 AT1 A1A2 Ak
= I
Ax = b
solution is
x = A1b = AT b
orthogonal matrices
AT is a left inverse of A:
AT A = I
AAT 6= I
the span of a collection of vectors is the set of all their linear combinations:
Example
1 0 1 x 1 + x3
range( 1 1 2 ) = x1 + x2 + 2x3 | x1, x2, x3 R
0 1 1 x2 + x3
AAT b
n Ax = AAT b
range(A) = {Ax | x R }
z range(A), y range(A)
y = b AAT b
b
range(A) z = AAT b
b = Ax + y, AT y = 0 AT b = x, y = b AAT b
orthogonal matrices
A Cmn has orthonormal columns if its Gram matrix is the identity matrix:
H
H
A A = a1 a2 an a1 a2 an
aH H H
1 a1 a1 a2 a1 an
aH H
2 a1 a2 a2 aH
2 an
= .. .. ..
. . .
aH H
n a1 an a2 aH
n an
1 0 0
0 1 0
= .. .. . . ..
. . . .
0 0 1
Unitary matrix
Inverse
H
A A=I
= AAH = I
A is square
recall definition from page 3-37 (with = e 2j/n
and j = 1)
1 1 1 1
1 1 2 (n1)
2 4 2(n1)
W =
1
.. .. .. ..
. . . .
1 (n1) 2(n1) (n1)(n1)
the matrix (1/ n)W is unitary (proof on next page):
1 H 1
W W = WWH = I
n n
inverse of W is W 1 = (1/n)W H
inverse discrete Fourier transform of n-vector x is W 1x = (1/n)W H x
we show that W H W = nI
conjugate transpose of W is
1 1 1 1
1 2 n1
2 4 2(n1)
WH =
1
.. .. .. ..
. . . .
1 n1 2(n1) (n1)(n1)
n(ij) 1
H
(W W )ii = n, H
(W W )ij = = 0 if i 6= j
ij 1
(last step follows from n = 1)