Scribd
Upload
2 views

bitstream_821478

The document contains exercises related to matrix theory, specifically focusing on properties of matrices, pseudo-inverses, and structured singular values. It includes tasks to verify mathematical properties, compute structured singular values for specific matrix forms, and analyze perturbations of matrices. The exercises emphasize the application of concepts such as the Moore-Penrose inverse, singular value decomposition, and the impact of perturbations on singular values.

Uploaded by

sharmin.105.kuet
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
2 views

bitstream_821478

The document contains exercises related to matrix theory, specifically focusing on properties of matrices, pseudo-inverses, and structured singular values. It includes tasks to verify mathematical properties, compute structured singular values for specific matrix forms, and analyze perturbations of matrices. The exercises emphasize the application of concepts such as the Moore-Penrose inverse, singular value decomposition, and the impact of perturbations on singular values.

Uploaded by

sharmin.105.kuet
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
You are on page 1/ 2

Exercises

Exercise 4.1 Verify that for any A, an m � n matrix, the following holds:
p1 kAk � kAk � pmkAk :
n 1 2 1

Exercise 4.5 Suppose the m � n matrix A is decomposed in the form


�� 0�
A�U 0 0 V0
where U and V are unitary matrices, and � is an invertible r � r matrix (| the SVD could be used to
produce such a decomposition). Then the \Moore-Penrose inverse", or pseudo-inverse of A, denoted
by A+ , can be de�ned as the n � m matrix
� �;1 0 �
A+ �V 0
0 0 U
(You can invoke it in Matlab with pinv(A).)
a) Show that A+A and AA+ are symmetric, and that AA+ A � A and A+AA+ � A+ . (These
four conditions actually constitute an alternative de�nition of the pseudo-inverse.)
b) Show that when A has full column rank then A+ � (A0 A);1A0 , and that when A has full
row rank then A+ � A0 (AA0 );1 .
c) Show that, of all x that minimize ky ; Axk2 (and there will be many, if A does not have full
column rank), the one with smallest length kxk2 is given by x^ � A+ y.

Exercise 4.7 Structured Singular Values


Given a complex square matrix A, de�ne the structured singular value function as follows.
�� (A) � min f� (�) 1j det(I ; �A) � 0g
�2� max
where � is some set of matrices.
a) If � � f�I : � 2 C g, show that ��(A) � �(A), where � is the spectral radius of A, de�ned
as: �(A) � maxi j�i j and the �i 's are the eigenvalues of A.
b) If � � f� 2 C n�n g, show that ��(A) � �max (A)
c) If � � fdiag(�1� � � � � �n ) j �i 2 C g, show that
�(A) � �� (A) � ��(D;1 AD) � �max (D;1 AD)
where
D 2 fdiag(d1 � � � � � dn ) j di � 0g
Exercise 4.8 Consider again the structured singular value function of a complex square matrix A
de�ned in the preceding problem. If A has more structure, it is sometimes possible to compute ��(A)
exactly. In this problem, assume A is a rank-one matrix, so that we can write A � uv0 where u� v are
complex vectors of dimension n. Compute �� (A) when

(a) � � diag(�1� : : : � �n )� �i 2 C .
(b) � � diag(�1� : : : � �n )� �i 2 R .
To simplify the computation, minimize the Frobenius norm of � in the de�ntion of ��(A).

Exercise 5.1 Suppose the complex m � n matrix A is perturbed to the matrix A + E .


(a) Show that
j �max (A + E ) ; �max (A) j � �max (E )
Also �nd an E that results in the inequality being achieved with equality.
(Hint: To show the inequality, write (A + E ) � A + E and A � (A + E ) ; E , take the 2-norm
on both sides of each equation, and use the triangle inequality.)
It turns out that the result in (a) actually applies to all the singular values of A and A + E , not
just the largest one. Part (b) below is one version of the result for the smallest singular value.
(b) Suppose A has less than full column rank, i.e. has rank� n, but A + E has full column rank.
Show (following a procedure similar to part (a) | but looking at min k(A + E )xk2 rather than
the norm of A + E , etc.) that
�min (A + E ) � �max (E )
Again �nd an E that results in the inequality being achieved with equality.
[The result in (b), and some extensions of it, give rise to the following sound (and widely used)
procedure for estimating the rank of some underlying matrix A, given only the matrix A + E
and knowledge of kE k2: Compute the SVD of A + E , then declare the \numerical rank" of A to
be the number of singular values of A + E that are larger than the threshold kE k2. The given
information is consistent with having an A of this rank.]
(c) Verify the above results using your own examples in MATLAB. You might also �nd it interesting
to verify numerically that for large m, n, the norm of the matrix E � s � randn(m� n) | which
is a matrix whose p are independent, zero-mean, Gaussian, with standard deviation s |
p entries
is close to s � ( m + n). So if A is perturbed by such a matrix, then a reasonable value to use
as a threshold when determining the numerical rank of A is this number.

You might also like