Documenti di Didattica
Documenti di Professioni
Documenti di Cultura
(A)
(A)
(3) (10 pts) Let A, B Rnn be such that B is nonsingular and they satisfy the inequality
kAk kB 1 k 0.5. Show that B A is nonsingular.
Solution: Observe that B A = B(I B 1 A). Then B A is nonsinular if and only if
I B 1 A is nonsingular. Let X := B 1 A. We use that kXk = kB 1 Ak kB 1 kkAk < 1
and by the basic lemma we proved in class we conclude that I X has an inverse.
(4) (10 pts) Let (A) be the spectral radius of A. Is there a matrix such that (A) < kAk for
every subordinate matrix norm?
Solution: Yes, there is. Here is an example of such a matrix:
0 1
A=
.
0 0
We have that (A) = 0 and it is less than any subordinate matrix norm of A (any norm
of A is larger that zero). In fact, all nonzero upper or lower triangular matrices with zero
diagonal elements have (A) = 0, while any subordinate norm will be greater than 0.
1
(5) (10 pts) Assume that A and B are such that for some subordinate matrix norm k k the
following holds true: kI BAk < 1. Show that both A and B are nonsingular.
Solution: Let X = BA. Then X has an inverse because kI Xk < 1. Therefore, X is
nonsingular, i.e., det(X) 6= 0. Because det(X) = det(A) det(B) 6= 0 we conclude that both
A and B are nonsingular.
1. Some Not So Simple Problems
(6) (20 pts) Consider the system Ax = b where A Rnn is such that ajj = 1 >
1 j n. Prove that the Jacobi iteration method converges.
i6=j
|aij |, for
Solution: We proved in class tha Jacobi converges for any Strictly Diagonally Dominant
matrix. It is also in the book.
(7) (20 pts) Show that the matrix 1-norm subordinate to the vector 1-norm is given by:
n
n
X
X
kAk1 = max
|aij |, where kxk1 =
|xi |.
1jn
i=1
i=1
3
1
0 ... 0
0
1
3
1
.
.
.
0
0
0
1
3 ... 0
0
C=
... ... ... ... ... ...
0
0
0 ... 3
1
0
0
0 ... 1
3
Solution: Look at the general notes posted on my page for the spectrum of such matrices.
(9) (20 pts) Prove that the Gauss-Seidel iterative method for the system Cx = b converges.
Here C is defined in Problem 8.
Solution: Let C = DLU . The we have that G = (DL)1 U , and x(k+1) = Gx(k) +(D
L)1 b. Take an eigenvector x for an eigenvalue . Then Gx = x implies U x = (D L)x
which is equivalent to (L + U )x = Dx. Take xk such that |xk | = max1in |xi |. Then,
using equation k in the systems (L + U )x = Dx, we have
3|| = |||ckk | ||
k1
X
|ckj | +
j=1
n
X
|ckj |.
j=k+1
Therefore, we obtain
3
k1
X
j=1
n
1
1 X
|ckj | = 1 +
|ckj | +
||
||
j=k+1
and this implies that || 0.5 < 1. Then, the spectral radius of G is less than one, hence
the CG method converges.