Scribd
Upload
84 views1 page

Generalized Inverse in Stat 471

This document summarizes key concepts from a lecture on generalized inverses: 1. The generalized inverse (pseudoinverse) of a matrix X is (X'X)-1X' and can be used to find the least squares solution to an overdetermined system. 2. The Moore-Penrose generalized inverse X- satisfies certain matrix equations and is a unique inverse. 3. The Moore-Penrose generalized inverse can be computed using the singular value decomposition (SVD) of X. The inverse involves taking the reciprocal of non-zero singular values.

Uploaded by

crazynup
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
84 views1 page

Generalized Inverse in Stat 471

This document summarizes key concepts from a lecture on generalized inverses: 1. The generalized inverse (pseudoinverse) of a matrix X is (X'X)-1X' and can be used to find the least squares solution to an overdetermined system. 2. The Moore-Penrose generalized inverse X- satisfies certain matrix equations and is a unique inverse. 3. The Moore-Penrose generalized inverse can be computed using the singular value decomposition (SVD) of X. The inverse involves taking the reciprocal of non-zero singular values.

Uploaded by

crazynup
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd

Stat 471: Lecture 7

Generalized Inverse.
Moo K. Chung
mchung@[Link]
September 22, 2003
1.

2 1
4 2

y
1
y
2

1
1

.
There is no exact solution but it is possible to
get a solution in the least-squares sense. Note
that
2y
1
+ y
2
1 =
1
, 4y
1
+ y
2
1 =
2
.
Minimize the sum of squared errors, min(X
c)

(X c), where c = (1, 1)

, = (y
1
, y
2
)

and
X =

2 1
4 2

A = A,

A = 2A.
Then it can be shown that

= (X

X)
1
X

c.
The generalized inverse (psudo inverse) of X is
X

= (X

X)
1
X

.
2. Moore-Penrose generalized inverse X

is a
unique matrix that satises the following ma-
trix equations
XX

X = X, X

XX

= X

(XX

= XX

, (X

X)

= X

X.
(see C.R. Rao and S.K. Mitra, Generalized in-
verse of matrices and its applications, Wiley,
New Work, 1971; Penrose, R., A Generalized
Inverse for Matrices. Proc. Cambridge Phil.
Soc. 51, 406-413, 1955.)
>>X=[2 1
4 2]
>>Y=pinv(X)
Y =
0.0800 0.1600
0.0400 0.0800
>> X*Y*X
ans =
2.0000 1.0000
4.0000 2.0000
3. Computing the generalized inverse is based on
the singular value decomposition (SVD). Let X
be n p matrix with n p. Then SVD of X is
X = UDV

,
where U
np
has orthonormal columns, D
pp
=
Diag(d
1
, , d
p
) is diagonal with non-negative
elements and V
pp
is orthogonal. For symmetric
positive denite matrix X
pp
, we have diagonal-
ization X = QDQ

(lecture 07). For nonsingu-


lar matrix X
nn
, X = UDV

and U is orthog-
onal. Let D

= Diag(d

1
, , d

p
), d

i
= 1/d
i
if d
i
= 0 and d

i
= 0 if d
i
= 0. Then it can
be shown that the Moore-Penrose generalized
inverse is given by
X

= V D

.
>> [U,D,V]=svd(X)
D =
5.0000 0
0 0.0000
>> U*D*V
ans =
2.0000 1.0000
4.0000 2.0000
>> V*[1/5 0
0 0]*U
ans =
0.0800 0.1600
0.0400 0.0800

You might also like