2013 Assignments With Solutions

You might also like

Download as pdf or txt
Download as pdf or txt
You are on page 1of 43

MAT3701/201/1/2013

Tutorial Letter 201/1/2013

Linear Algebra
MAT3701

Semester 1

Department of Mathematical Sciences

Solutions to Assignment 1

BAR CODE

university
Learn without limits.
Open Rubric of south africa
ONLY FOR SEMESTER 1 STUDENTS
ASSIGNMENT 01
Based on Study Units 1 - 9
FIXED CLOSING DATE: 7 MARCH 2013
UNIQUE NUMBER: 362116
Please note that we will only mark a selection of the questions. It is therefore in your own interest to
do all the questions. The fact that a question is not marked does not mean that it is less important
than one that is marked. Worked solutions to all the questions will be sent to all students shortly
after the due date. For this assignment only questions 1, 3, 5 and 7 will be marked.

QUESTION 1

0 1 i 0
Let A D and B D ; and let
1 0 0 i
W1 D fX 2 M2 2 .C/ : AX D X Bg

and
W2 D fX 2 M2 2 .C/ : AX D i X g :

(a) Show that W1 is a subspace of M2 2 .C/ : (4)


(b) Given also that W2 is a subspace of M2 2 .C/ ; find a basis for W1 \ W2 : (7)
(c) Explain whether or not M2 2 .C/ D W1 W2 : (1)
[12]

SOLUTION

(a) ST1: W1 6D ; since e.g. 02 2 2 W1 :


ST2: Suppose X 1 2 W1 and X 2 2 W1 : Then

A .X 1 C X 2 / D AX 1 C AX 2
D X 1 B C X 2 B; since X 1 ; X 2 2 W1
D .X 1 C X 2 / B
Thus, X 1 C X 2 2 W1

ST3: Suppose z 2 C and X 2 W1 : Then

A .z X / D z .AX /
D z .X B/ ; since X 2 W1
D .z X / B
Thus, z X 2 W1

Since all three conditions are satisfied, W1 is a subspace of M2 2 .C/ :

2
MAT3701/201

(b) Let X D [X 1 X 2 ] ; where X 1 and X 2 denote the first and second columns of X; respectively.
Then
X 2 W1 \ W2 , AX D X B and AX D i X
i 0
, A [X 1 X 2 ] D [X 1 X 2 ] and A [X 1 X 2 ] D i [X 1 X 2 ]
0 i
, [AX 1 AX 2 ] D [ i X 1 i X 2 ] and [AX 1 AX 2 ] D [i X 1 i X 2 ]
, AX 1 D i X 1 D i X 1 and AX 2 D i X 2
, X 1 D 0 and AX 2 D i X 2 ...(i)
x1
Let X 2 D ; then
x2
0 1 x1 x1
AX 2 D i X 2 , Di
1 0 x2 x2
, x2 D i x1 and x1 D i x2
, x1 D i x2

Thus,
i x2 i
X2 D D x2 ; x2 2 C
x2 1
and therefore
0 i
X D [X 1 X 2 ] D x2 ; x2 2 C
0 1
0 i
so that is a basis for W1 \ W2 .
0 1
(c) No, since W1 \ W2 6D f0g :

QUESTION 2

Suppose U1 ; U2 and U are subspaces of a vector space V over a field F such that V D U1 U2 and
U1 U: Prove that U D U1 .U \ U2 / :

SOLUTION

It suffices to show that U D U1 C .U \ U2 / and U1 \ .U \ U2 / D f0g :


Let u 2 U: Since V D U1 U2 ; it follows that u D u 1 C u 2 ; where u 1 2 U1 and u 2 2 U2 : Since
U1 U; we have that u u 1 2 U; and therefore u 2 D u u 1 2 U \ U2 .
It follows that U U1 C .U \ U2 / : But it also follows from U1 U and .U \ U2 / U that U1 C
.U \ U2 / U; and therefore U D U1 C .U \ U2 / :
Finally,
U1 \ .U \ U2 / D U1 \ .U2 \ U /
D .U1 \ U2 / \ U
D f0g \ U; since U1 \ U2 D f0g
D f0g ;
and we conclude that U D U1 .U \ U2 / :

3
QUESTION 3

Classify each of the following statements as true or false. If true, construct a T with the given
properties. If false, explain why such a T does not exist.

(a) There exists a linear transformation T : C 3 ! C 2 such that rank .T / D nullity .T / : (4)

(b) There exists a linear transformation T : C 3 ! C 2 such that T is onto and N .T / D spanf.1; 1; 1/g :
(4)

(c) There exists a linear transformation T : C 3 ! C 2 such that N .T / D f.0; 0; 0/g : (4)
[12]

SOLUTION

(a) False. Since rank .T / C nullity.T / D 3; it is impossible for rank.T / D nullity.T / :

(b) True. Define T : C 3 ! C 2 by T .x; y; z/ D .x y; y z/ :


Then

.x; y; z/ 2 N .T / , T .x; y; z/ D 0
, x y D 0 and y z D 0
, .x; y; z/ D .y; y; y/ D y .1; 1; 1/ ; y 2 C

Thus, N .T / Dspanf.1; 1; 1/g :


Since nullity.T / D 1; it follows that rank.T / D 3–nullity.T / D 2; and since T maps to C 2 ;
which is also two–dimensional, T must be onto.

(c) False. Since T maps to C 2 it follows that rank.T / 2; and since also rank.T / Cnullity.T / D 3;
we have that nullity.T / 1.

QUESTION 4

Let T : V ! V be a linear operator and suppose 2 F is an eigenvalue of T: Prove that

(a) C 1 is an eigenvalue of T C I V I

(b) 2 is an eigenvalue of T 2 I

(c) if T is nonsingular, then 6D 0I

(d) if T is nonsingular, then 1 is an eigenvalue of T 1:

4
MAT3701/201

SOLUTION

Since 2 F is an eigenvalue of T; there exists a vector v 6D 0 in V such that T .v/ D v:


(a) .T C I V / .v/ D T .v/ C I V .v/
D vCv
D . C 1/ v;
Hence C 1 is an eigenvalue of T C I V since v 6D 0:
(b) T 2 .v/ D T .T .v//
D T . v/
D T .v/
D 2 v;
so that 2 is an eigenvalue of T 2 :
(c) If T is nonsingular, then T .v/ 6D 0 since v 6D 0: But v D T .v/ 6D 0; so that 6D 0.
(d) Applying T 1 to both sides of T .v/ D v yields
1 1 1
vDT .T .v// D T . v/ D T .v/ ;

so that T 1 .v/ D 1v since 6D 0 from (c). Thus, 1 is an eigenvalue of T 1 since v 6D 0:

QUESTION 5
1
Let T : M2 2 .C/ ! M2 2 .C/ be the linear operator defined by T .A/ D 2 A C At :
(a) Show that T is a projection. (4)
(b) Find a basis for the space onto which T projects. (4)
(c) Find a basis for the space along which T projects. (4)
(d) Find a basis for M2 2 .C/ such that [T ] is diagonal. Write down [T ] : (2)
[14]

SOLUTION

(a) T 2 .A/ D T 12 A C At
D 12 T A C At
D 12 hT .A/ C T At i
t
D 21 12 A C At C 12 At C At
D 21 12 A C At C 1
2 At C A
D 12 A C At
D T .A/
Therefore, T is a projection

5
(b) The space onto which T projects is equal to R .T / :
If we denote the standard basis for M2 2 .C/ by

1 0 0 1 0 0 0 0
D E 11 D ; E 12 D ; E 21 D ; E 22 D
0 0 0 0 1 0 0 1

then

R.T / D span fT .E 11 / ; T .E 12 / ; T .E 21 / ; T .E 22 /g
1 0 1 0 1 1 0 1 0 0
D span ; ; ;
0 0 2 1 0 2 1 0 0 1
1 0 0 1 0 0
D span ; ;
0 0 1 0 0 1

so that a basis for the space onto which T projects is given by

1 0 0 1 0 0
1 D ; ; :
0 0 1 0 0 1

(c) The space along which T projects is equal to N .T / : Now,

A 2 N .T / , T .A/ D 0
1
, A C At D 0
2
, At D A

a b
Let A D ; then
c d

At D A
a c a b
, D
b d c d
, a D d D 0 and b D c
0 1
, ADb ; b2C
1 0

0 1
Thus, N .T / Dspan ; and it follows that a basis for the space along which T
1 0
projects is given by
0 1
2 D
1 0

(d) Since T is a projection,

R .T / D E 1 .T / ; N .T / D E 0 .T / and M2 2 .C/ D R .T / N .T / ;

6
MAT3701/201

hence
1 0 0 1 0 0 0 1
D 1 [ 2 D ; ; ;
0 0 1 0 0 1 1 0

is a basis for M2 2 .C/ such that [T ] is diagonal, and


2 3
1 0 0 0
6 0 1 0 0 7
[T ] D 64 0 0 1
7
0 5
0 0 0 0

since the first three matrices in are eigenvectors of T corresponding to D 1 and the last
matrix in is an eigenvector of T corresponding to D 0:

QUESTION 6

Let T : M2 2 .C/ ! M2 2 .C/ be the linear operator defined by

a b c d d
T D :
c d aCb b

(a) Show that T satisfies the test for diagonalizability.

(b) Find a basis for M2 2 .C/ consisting of eigenvectors of T , and write down [T ] :

(c) Show that T satisfies the equation T 2 D I .

SOLUTION

(a) We determine the matrix of T with respect to the basis

1 0 0 1 0 0 0 0
D E 11 D I E 12 D I E 21 D I E 22 D
0 0 0 0 1 0 0 1

of M2 2 .C/ :

0 0
T .E 11 / D D 0 E 11 C 0 E 12 C 1 E 21 C 0 E 22
1 0
0 0
T .E 12 / D D 0 E 11 C 0 E 12 C 1 E 21 C 1 E 22
1 1
1 0
T .E 21 / D D 1 E 11 C 0 E 12 C 0 E 21 C 0 E 22
0 0
1 1
T .E 22 / D D 1 E 11 C 1 E 12 C 0 E 21 C 0 E 22
0 0

7
Thus it follows that 2 3
0 0 1 1
6 0 0 0 1 7
[T ] D 6
4 1
7:
1 0 0 5
0 1 0 0

The characteristic polynomial of T is therefore given by

P . / D det I4 [T ]
0 1 1
0 0 1
D
1 1 0
0 1 0
1 1 0 1
D 1 0 1 1 (expansion by second row)
0 0 0 1 0
1 1
D 2
1 (expanding both by the 3rd row)
1 1
2 1
D 1
1
2
2
D 1
D . 1/2 . C 1/2

which clearly splits. The eigenvalues of T are D 1 and D 1; both of multiplicity two.

2 3
1 0 1 1
6 0 1 0 1 7
E 1 : I4 [T ] D 6
4
7
1 1 1 0 5
0 1 0 1
2 3
1 0 1 1
6 0 1 0 1 7
!6
4 0
7
1 0 1 5 R3 ! R3 C R1
0 1 0 1
2 3
1 0 1 1
6 0 1 0 1 7 . . . (ii)
!6
4 0 0
7
0 0 5 R3 ! R3 C R2
0 0 0 0 R4 ! R4 C R2

It follows that
dim .E 1 / D 2 D multiplicity of D 1:

8
MAT3701/201

2 3
1 0 1 1
6 0 1 0 1 7
E : I4 [T ] D 6
4
7
1
1 1 1 0 5
0 1 0 1
2 3
1 0 1 1
6 0 1 0 1 7
!6
4
7
0 1 0 1 5 R3 ! R3 R1
0 1 0 1
2 3
1 0 1 1
6 0 1 0 1 7
!6
4
7
0 0 0 0 5 R3 ! R3 R2 . . . (iii)
0 0 0 0 R4 ! R4 R2
Thus
dim .E 1/ D 2 D multiplicity of D 1:

Since in both cases the dimension of the eigenspace equals the multiplicity of its associated
eigenvalue, it follows that T is diagonalisable.
(b) We use (ii) to obtain a basis for E 1 [T ] : Let x3 D and x4 D ; then x2 D and x1 D :
Thus
.x1 ; x2 ; x3 ; x4 / D f. ; ; ; /: ; 2 Cg
D f .1; 0; 1; 0/ C . 1; 1; 0; 1/ : ; 2 Cg
Dspanf.1; 0; 1; 0/ ; . 1; 1; 0; 1/g

So that f.1; 0; 1; 0/ ; . 1; 1; 0; 1/g is a basis for E 1 [T ] :


Since
1 0
E 11 C E 21 D
1 0
and
1 1
E 11 C E 12 C E 22 D
0 1
it follows that
1 0 1 1
1 D ;
1 0 0 1
is a basis for E 1 .T / :
We use (iii) to obtain a basis for E 1 [T ] : Let x3 D and x4 D ; then x2 D and
x1 D C : Thus

.x1 ; x2 ; x3 ; x4 / D f. C ; ; ; / : ; 2 Cg
D f . 1; 0; 1; 0/ C .1; 1; 0; 1/ : ; 2 Cg
Dspanf. 1; 0; 1; 0/ ; .1; 1; 0; 1/g

9
so that f. 1; 0; 1; 0/ ; .1; 1; 0; 1/g is a basis for E 1 [T ] :
Since
1 0
E 11 C E 21 D
1 0

and
1 1
E 11 E 12 C E 22 D
0 1

it follows that
1 0 1 1
2 D ;
1 0 0 1

is a basis for E 1 .T / :
Therefore
1 0 1 1 1 0 1 1
D 1 [ 2 D ; ; ;
1 0 0 1 1 0 0 1

is a basis for M2 2 .C/ consisting of eigenvectors of T; and


2 3
1 0 0 0
6 0 1 0 0 7
[T ] D 64 0 0
7:
1 0 5
0 0 0 1

(c)
a b c d d
T2 DT
c d aCb b
aCb b b
D
c d Cd d
a b
D
c d

so that T 2 D I M2 2 .C/ :

QUESTION 7

Let T : M2 2 .C/ ! M2 2 .C/ be the linear operator defined by T .X / D AX , where


0 1
AD ; and let W be the cyclic subspace generated by I2 :
1 0

(a) Find the T –cyclic basis for W: (4)

(b) Find the characteristic polynomial of TW : (4)

10
MAT3701/201

(c) Explain whether or not


1 1 1 2
U D span ;
1 1 2 1
is T –invariant. (4)
[12]

SOLUTION

(a)
T .I2 / D AI2
0 1
D
1 0
0 1
T 2 .I2 / DT
1 0
1 0
D
0 1
D I2
Note that fI2 ; T .I2 /g is linearly independent since I2 is not a multiple of T .I2 / but that
I2 ; T .I2 / ; T 2 .I2 / is linearly dependent since T 2 .I2 / C I2 D 0: : : :(iv)
Therefore fI2 ; T .I2 /g is the cyclic basis for W:
(b) It follows from (iv) that
p .t/ D . 1/2 t 2 C 1 D t 2 C 1

is the characteristic polynomial of TW :


1 1 1 2
(c) Let B D and C D and note that U D spanfB; Cg :
1 1 2 1
Now,
0 1 1 1
T .B/ D
1 0 1 1
1 1
D
1 1
D 3B C 2C (show)
2U
and
0 1 1 2
T .C/ D
1 0 2 1
2 1
D
1 2
D 5B C 3C (show)
2U
Therefore U is T -invariant.

11
QUESTION 8

Let 2 3
1
0 2 0
6 7
6 7
AD6
6
1
2
1
2 1 7
7:
4 5
1
2 0 0
(a) Show that A is a regular transition matrix.
(b) Find lim Am :
m!1

(c) Describe the Gerschgorin discs in which the eigenvalues of A lie.

SOLUTION

(a) A is a transition matrix since all its entries are nonnegative and all three its column sums are
equal to 1: It is also regular since all the entries of
2 32 1 1 1 3 2 3 1 1 3
0 21 0 4 4 2 8 4 4
A3 D A A2 D 4 21 12 1 5 4 34 12 12 5 D 4 21 58 12 5
1
2 0 0 0 14 0 1
8
1
8
1
4

are positve.
(b) Apply Theorem 5.20 (f).
First calculate E 1 .A/
2 3 2 3 2 1 3 2 3 32
x1 x1 1 2 x1 0 0
A 4 x2 5 D 4 x2 5 , 4 1
2
1
2
5 4 x2 5 D 4 0 5
1
x3 x3 1 x3 0
2 0 1
2 3 2 3
x1 2
, 4 x2 5 D t 4 4 5 ; t 2 R (show)
x3 1
2 3
2
14 5
The fixed probability vector of A is given by P D 4 ; and it follows that
7 1
2 3
2 2 2
1
lim Am D 4 4 4 4 5 :
m!1 7 1 1 1

(c) Let Ci denote the Gerschgorin disk associated with the i–th row of A: Then
1
C1 D C3 D z 2 C : jzj
2
1 3
C2 D z2C : z
2 2

12
MAT3701/202/1/2013

Tutorial Letter 202/1/2013

Linear Algebra
MAT3701

Semester 1

Department of Mathematical Sciences

Solutions to Assignment 2

BAR CODE

university
Learn without limits.
Open Rubric of south africa
ONLY FOR SEMESTER 1 STUDENTS
ASSIGNMENT 02
Based on Study Units 10 - 17
FIXED CLOSING DATE: 2 APRIL 2013
UNIQUE NUMBER: 396673
Please note that we will only mark a selection of the questions. It is therefore in your own interest to
do all the questions. The fact that a question is not marked does not mean that it is less important
than one that is marked. Worked solutions to all the questions will be sent to all students shortly
after the due date. For this assignment only questions 2, 4, 6 and 8 will be marked.

QUESTION 1

1 i
Define B : C 2 C 2 ! C by B .x; y/ D x Ay, where A D and x and y are column vectors
i 1
in C 2 : Explain which of the inner product axioms are satisfied by B; and which are not.

SOLUTION:
x1 y1 z1
Let x D ; yD and z D be column vectors in C 2 and let c 2 C.
x2 y2 z2
IP1

B .x C z; y/ D .x C z/ Ay
D x1 C z 1 ; x2 C z 2 Ay
D .x1 C z 1 ; x2 C z 2 / Ay
D ..x1 ; x2 / C .z 1 ; z 2 // Ay
D .x1 ; x2 / Ay C .z 1 ; z 2 / Ay
D x Ay C z Ay
D B .x; y/ C B .z; y/

Thus IP1 is satisfied.


IP2

B .cx; y/ D .cx/ Ay
D .cx1 ; cx2 / Ay
D c .x 1 ; x 2 / Ay
D cx Ay
D cB .x; y/ :

1
Thus IP2 is not satisfied in general. Take, e.g., x D y D and c D i; then B .cx; y/ D i and
0
cB .x; y/ D i; i.e., B .cx; y/ 6D cB .x; y/ :

2
MAT3701/202

IP3
1 i y1
B .x; y/ D .x1 ; x2 /
i 1 y2
D x y1 C i x1 y2 i x2 y1 C x 2 y2
D x 1 y1 C i x1 y2 i x 2 y1 C x2 y2
D x1 y1 i x1 y2 C i x2 y1 C x2 y 2
and
1 i x1
B .y; x/ D y1; y2
i 1 x2
D y 1 x1 C i y 1 x2 i y 2 x1 C y 2 x2
Thus B .x; y/ D B .y; x/ ; so IP3 is satisfied.
IP4
1 i x1
B .x; x/ D .x 1 ; x 2 /
i 1 x2
D x 1 x1 C i x 1 x2 i x 2 x1 C x 2 x2
1
Thus IP4 is not satisfied in general. Take, e.g. x D ; then B .x; x/ D 0.
i

QUESTION 2

Let f a ; f b ; f c be the Lagrange polynomials associated with the distinct real numbers a; b; c
respectively.
(a) Show that
hg; hi D g .a/ h .a/ C g.b/h .b/ C g .c/ h .c/
is an inner product on P2 .R/ over R: (9)
(b) Show that f f a ; f b ; f c g is an orthonormal basis for P2 .R/ with respect to h; i : (4)
(c) Show that, with respect to h; i ; the orthogonal projection P : P2 .R/ ! P2 .R/ on W D
spanf f a ; f b g is given by P .g/ D g .a/ f a C g .b/ f b : (4)
[17]

SOLUTION:

(a) Let f; g and h denote polynomials in P2 .R/ and let r 2 R.


IP1
h f C g; hi D . f C g/ .a/ h .a/ C . f C g/ .b/ h .b/ C . f C g/ .c/ h .c/
D . f .a/ C g .a// h .a/ C . f .b/ C g .b// h .b/ C . f .c/ C g .c// h .c/
D . f .a/ h .a/ C f .b/ h .b/ C f .c/ h .c// C .g .a/ h .a/ C g .b/ h .b/ C g .c/ h .c//
D h f; hi C hg; hi

Thus IP1 is satisfied.

3
IP2

hrg; hi D .rg/ .a/ h .a/ C .rg/ .b/ h .b/ C .rg/ .c/ h .c/
D .rg .a// h .a/ C .rg .b// h .b/ C .rg .c// h .c/
D r .g .a/ h .a/ C g .b/ h .b/ C g .c/ h .c//
D r hg; hi

Thus IP2 is satisfied.


IP3
Over R this axiom becomes hg; hi D hh; gi since g; h D hg; hi :
Now

hg; hi D g .a/ h .a/ C g .b/ h .b/ C g .c/ h .c/


D h .a/ g .a/ C h .b/ g .b/ C h .c/ g .c/
D hh; gi

Thus IP3 is satisfied.


IP4
hg; gi D g .a/2 C g .b/2 C g .c/2 0

and

hg; gi D 0 , g .a/ D g .b/ D g .c/ D 0


, gD0

since a nonzero polynomial of degree at most 2 can have at most 2 zeros.


Thus IP4 is satisfied and it follows that h ; i is an inner product on P2 .R/.

(b)

h f a ; f a i D f a .a/ f a .a/ C f a .b/ f a .b/ C f a .c/ f a .c/


D 1:1 C 0:0 C 0:0
D 1

Similarly h f b ; f b i D h f c ; f c i D 1
Further,

h f a ; f b i D f a .a/ f b .a/ C f a .b/ f b .b/ C f a .c/ f b .c/


D 1:0 C 0:1 C 0:0
D 0

Similarly h f a ; f c i D h f b ; f c i D 0
Thus, it follows that f f a ; f b ; f c g is an orthonormal basis for P2 .R/.

4
MAT3701/202

(c) By definition,
P .g/ D hg; f a i f a C hg; f b i f b
D .g .a/ f a .a/ C g .b/ f a .b/ C g .c/ f a .c// f a
C .g .a/ f b .a/ C g .b/ f b .b/ C g .c/ f b .c// f b
D .g .a/ :1 C g .b/ :0 C g .c/ :0/ f a C .g .a/ :0 C g .b/ :1 C g .c/ :0/ f b
D g .a/ f a C g .b/ f b

QUESTION 3

Let 9
x1 x2 D3 =
x1 x3 D 0 ... (i)
;
2x1 x2 x3 D 3
(a) Express the general solution to (i) in the form
x D v C wt;

where t is a parameter and v and w are fixed vectors in R 3 :


(b) Use your expression in (a) to express kxk2 as a quadratic polynomial in t; and find the value
of t which minimises kxk2 : Substitute this value for t in (a) to obtain the minimal solution of
(i).
(c) Check your answer to (b) by finding the minimal solution according to the method described
in Friedberg: Section 6.3.

SOLUTION:
We follow the solution in SG: example 2, p. 123, which is similar to this question.
(a) Use Gaussian elimination to solve (i). In matrix form:
2 3 2 3 2 3
1 1 0 : 3 1 1 0 : 3 1 1 0 : 3
4 1 0 5
1 : 0 ! 1 4 0 1 : 0 5 4
! 0 1 1 : 3 5 R2 R1
2 1 1 : 3 1 1 0 : 3 R3 R2 0 0 0 : 0 R3 R1

The general solution is therefore given by


x3 D t;
x2 D t 3;
x1 D t

that is
x D .x1 ; x2 ; x3 / D .0; 3; 0/ C .1; 1; 1/ t ... (ii)
so that
v D .0; 3; 0/ and w D .1; 1; 1/

5
(b) From SG, top of p. 124,

kxk2 D kvk2 C 2hv; wit C kwk2 t 2


D k.0; 3; 0/k2 C 2h.0; 3; 0/ ; .1; 1; 1/it C k.1; 1; 1/k2 t 2
D 9 6t C 3t 2

Thus
d kxk2
D 6t 6D0,t D1
dt
From (ii), the minimal solution is therefore

x D .0; 3; 0/ C .1; 1; 1/ D .1; 2; 1/

(c) Following the method in Friedberg: section 6.3, example 3,


2 32 3
1 1 0 1 1 2
AA D 4 1 0 1 5 4 1 0 1 5
2 1 1 0 1 1
2 3
2 1 3
D 4 1 2 3 5I
3 3 6

so we consider the system

2x C y C 3z D 3
x C 2y C 3z D 0
3x C 3y C 6z D 3

A solution to this is, for example, 2 3


4
uD4 1 5
2

Hence, the minimal solution is given by


2 32 3 2 3
1 1 2 4 1
sDA uD 4 1 0 1 54 1 5D4 2 5;
0 1 1 2 1

which agrees with the solution in (b).

6
MAT3701/202

QUESTION 4 (Friedberg: Section 6.3, Exercise 9)

Prove that if V D W W ? and T is the projection on W along W ? ; then T D T : Hint: Recall that
N .T / D W ? : [8]

SOLUTION:

Let v1 D w1 Cw1? and v2 D w2 Cw2? be arbitrary vectors in V , where w1 ; w2 2 W and w1? ; w2? 2 W ? :
Then

hT .v1 / ; v2 i D w1 ; w2 C w2?
D hw1 ; w2 i since w1 ; w2? D 0
D hv1 ; w2 i since w1? ; w2 D 0
D hv1 ; T .v2 /i since T is the projection on W along W ?

Thus T D T :

QUESTION 5 (Friedberg: Section 6.3, Exercise 14)

Let V be an inner product space, and let y; z 2 V: Define T : V ! V by T .x/ D hx; yi z for all
x 2 V . First prove that T is linear. Then show that T exists, and find an explicit expression for it.

SOLUTION:

Let v1 and v2 denote vectors in V and let a 2 F:


Then

T .v1 C v2 / D hv1 C v2 ; yi z
D ..v1 ; y/ C hv2 ; yi/ z
D hv1 ; yi z C hv2 ; yi z
D T .v1 / C T .v2 /

and

T .av1 / D hav1 ; yi z
D a hv1 ; yi z
D aT .v1 / :

Thus T is a linear operator on V .


For all v1 ; v2 2 V;

hT .v1 / ; v2 i D hhv1 ; yi z; v2 i
D hv1 ; yi hz; v2 i since hv1 ; yi 2 F
D hv1 ; hz; v2 i yi since hz; v2 i 2 F
D hv1 ; hv2 ; zi yi
D v1 ; T .v2 / ;

7
hence it follows that T exists and

T .v2 / D hv2 ; zi y for all v2 2 V;

equivalently,
T .x/ D hx; zi y for all x 2 V:

QUESTION 6

It is given that A 2 M3 3 .C/ is a normal matrix with eigenvalues 1 and i and corresponding
eigenspaces
1
E 1 D span p . 1; 1; 0/
2
and
1 p 1 p
E i D span 1; 1; 2 ; 1; 1; 2 :
2 2

(a) Find the spectral decomposition of A: (9)

(b) Find A. (4)


[13]

SOLUTION:

(a) Since the bases of both eigenspaces are orthonormal, the matrix
2 1 1 1 3
p
6 2 2 2 7
6 1 1 1 7
6 7
PD6 p 7
6 2 2 2 7
4 1 1 5
0 p p
2 2
2 3
1 0 0
with these basis vectors as columns will be unitary, and A D P D P with D D 4 0 i 0 5 :
0 0 i
Hence
2 3 2 3
1 0 0 0 0 0
4
A D 1 P 0 0 0 5P Ci P4 0 1 0 5P
0 0 0 0 0 1
D 1 P1 C i P2 :::(i)

8
MAT3701/202

is the spectral decomposition of A, where


2 32 1 1 3
2 3
1 p p 0 1 1
2 3 p 0 0 76 7 0 7
6 2 2
1 0 0 6 2 76 1 1 7
1 7 6 2 6 2
D P4 0 0 0 5P D6 1 766 p 7D6 1 1 7
0 7
P1 6
4 p 0 0 7
5 6 2 2 2 7 4 2 5
0 0 0 2 4 1 1 1 5 2
0 0 0 p 0 0 0
2 2 2
2 32 1 1 3
1 1 2 3
p p 0 1 1
2 3 6 0 2 2 76 2 2 7 0 7
0 0 0 6 1 1 76 1 1 7
1 7 6 2 6 2
4 5 6 76 7
D P 0 1 0 P D6 0 76 p 7D6 1 1
0 7
P2
6 2 2 76 2 2 2 7 4 2 5
0 0 1 4 1 1 54 1 1 1 5 2
0 p p p 0 0 1
2 2 2 2 2

are the associated orthogonal projections.

(b) Substituting the values for P1 and P2 in (i) yields


2 3
1 1 1 1
6 2 C 2i C i 0 7
2 2
6 7
AD6 1 1 1 1
4 C i C i 0 75
2 2 2 2
0 0 i

QUESTION 7

Eliminate the x y–term (cross term) in

2x 2 C 2x y C 2y 2 D 1 ... (i)

by a rotation of the axes, that is, find a rotation matrix P 2 M2 2 .R/ such that (i), expressed in
terms of x 0 ; y 0 defined by
x x0
DP ;
y y0
contains no cross term.
Express x; y in terms of x 0 ; y 0 , and state the (counterclockwise) angle of rotation.

SOLUTION:

We follow the solution to the example in SG, p. 146. The quadratic form in (i) can be expressed as

X t AX D 1

where
2 1 x
AD and XD
1 2 y

9
The characteristic polynomial of A is given by

2 t 1
det .A t I2 / D
1 2 t
D .2 t/2 1
D .t 1/ .t 3/

resulting in the eigenvalues t D 1 and t D 3:


To find E 1 ; the eigenspace of A associated with t D 1; we solve the homogeneous system with
coefficient matrix
1 1 1 1
!
1 1 0 0 R2 R1
Thus
E 1 D span f.1; 1/g
with orthonormal basis
1
p .1; 1/
2
Similarly
E 3 D span f.1; 1/g
with orthonormal basis
1
p .1; 1/
2
Thus
Pt A P D D
where
1 1 1
PDp
2 1 1
(note that det .P/ D 1 to ensure the coordinate transformation is a rotation), and

1 0
DD
0 3

We therefore let
x x0 1 1 1 x0
DP Dp ;
y y0 2 1 1 y0
that is
1
x D p x 0 C y0
2
1
y D p x 0 C y0 ;
2
so that (i) becomes
2 2
x0 C 3 y0 D1

10
MAT3701/202

Since " #
p1 p1 cos sin
PD 2 2 D 4 4
p1 p1 sin cos
2 2 4 4

the rotation of the x y–axes is through 45o :

QUESTION 8

Let 2 3
1 1 1
AD4 1 0 1 5:
1 1 1

(a) Find a nonsingular matrix Q such that Q t AQ D D; where D is diagonal. (9)

(b) Write down the rank, index and signature of A. (3)


[12]

SOLUTION:

(a) Use the algorithm after Example 6 in Friedberg (see also SG: Section 16.3).
2 3
:
1 1 1 :: 1 0 0
6 :: 7
6 1 0 1 : 0 1 0 7
4 5
::
1 1 1 : 0 0 1
2 3
:
1 0 1 :: 1 0 0
6 :: 7 [R2 ! R2 R1 ]
! 6 4 0 1 0 : 1 1 0 75 and
: [C2 ! C2 C1 ]
1 0 1 :: 0 0 1
2 3
::
1 0 0 : 1 0 0
6 : 7 [R3 ! R3 R1 ]
6
! 4 0 1 0 :: 1 1 0 7 and
5
:: [C3 ! C3 C1 ]
0 0 0 : 1 0 1

Therefore
2 3t 2 3 2 3
1 0 0 1 1 1 1 0 0
QD 4 1 1 0 5 4
D 0 1 0 5 and D D 4 0 1 0 5
1 0 1 0 0 1 0 0 0

(b) rank.A/ D 2
index.A/ D 1
signature.A/ D 0

11
MAT3701/201/2/2013

Tutorial Letter 201/2/2013

LINEAR ALGEBRA
MAT3701

Semester 2

Department of Mathematical Sciences

This tutorial letter contains solutions


for assignment 01.

BAR CODE

university
Learn without limits.
Open Rubric of south africa
ASSIGNMENT 01
Solution
UNIQUE ASSIGNMENT NUMBER: 256311

Please note that we will only mark a selection of the questions. It is therefore in your own interest to
do all the questions. The fact that a question is not marked does not mean that it is less important
than one that is marked. Worked solutions to all the questions will be sent to all students shortly
after the due date. For this assignment, questions 1, 3, 5, and 7 will be marked.

Question 1

Let V, W and U denote the subspaces of M2×2 (C) defined by


  
a b
V = : a+b+c+d=0 ;
c d
  
a b
W = : a+b+c=0 ;
c 0

and  
−1 0
U = span .
0 1

(a) Explain whether or not V = W ⊕ U . (10)

(b) Is dim (V + W ) = dim (V + U )? Explain your answer. (2)


[12]

Solution

(a) It follows that W ⊆ V by letting d = 0 in the definition of V , and since


        
−1 0 −1 0 −z 0
U = span = z :z∈C = :z∈C ,
0 1 0 1 0 z

it also follows that U ⊆ V by letting a = −z, b = c = 0 and d = z in the definition of V .


Hence,
W + U ⊆ V. (i)
From the definition of V we have d = −a − b − c, so the general form of a matrix in V is
 
a b
v= with a, b, c ∈ C.
c −a − b − c

For V = W ⊕ U to be true it suffices to show that V = W + U and W ∩ U = {0}.

2
MAT3701/201

Now,      
a b z1 z2 −z4 0
v= = +
c −a − b − c z3 0 0 z4
with z1 + z2 + z3 = 0 and z1 , z2 , z3 , z4 ∈ C implies that

z1 = −z2 − z3
z1 − z4 = a
z2 = b
z3 = c
z4 = −a − b − c

hence

z1 = −b − c
z2 = b
z3 = c
z4 = −a − b − c

so that      
a b −b − c b a+b+c 0
v= = + ,
c −a − b − c c 0 0 −a − b − c
and it follows that V ⊆ W + U . This, together with (i), shows that V = W + U .
Finally, since    
z1 z2 −z4 0
=
z3 0 0 z4
with z1 + z2 + z3 = 0 and z1 , z2 , z3 , z4 ∈ C implies that z1 = z2 = z3 = z4 = 0, it follows that
W ∩ U = {0} and therefore V = W ⊕ U .

(b) We showed in (a) that W ⊆ V and U ⊆ V , hence it follows that V = V + W and V = V + U ,


respectively, and therefore dim (V + W ) = dim (V + U ).

Question 2

This question was not marked.


It is still important to work through the solutions and compare them with your own attempts.

Let U and W1 be subspaces of a finite–dimensional vector space V over F such that U ∩ W1 = {0} and
U ⊕ W1 6= V.

(a) Let W2 = W1 + span{v1 } , where v1 ∈


/ U ⊕ W1 . Prove that U ∩ W2 = {0} .

(b) Use (a) to prove the existence of a subspace W of V such that V = U ⊕ W and W1 ⊆ W.

3
Solution

(a) Suppose u ∈ U ∩ W2 . Then u = w1 + av1 , where w1 ∈ W1 and a ∈ F . Hence av1 = u − w1 ∈


span{v1 } ∩ U ⊕ W1 = {0}, since v1 ∈
/ U ⊕ W1 . Thus a = 0, and therefore it follows from u = w1 + av1
that also u = w1 = 0 since U ∩ W1 = {0}. Thus, U ∩ W2 = {0} .

(b) If U ⊕ W2 6= V, choose v2 ∈
/ U ⊕ W2 and let W3 = W2 + span{v2 }. Then U ∩ W3 = {0} according
to (a), and W1 ⊂ W2 ⊂ W3 . If V = U ⊕ W3 , then we are done. If not, repeat the process until it
eventually terminates since V is finite dimensional.

Question 3

Let A, B ∈ Mn×n (C) .

(a) Prove that if B is invertible, then there exists a scalar c ∈ C such that A + cB is not invertible.
Hint: Examine det (A + cB) . (5)

(b) Find nonzero 2 × 2 matrices A and B such that both A and A + cB are invertible for all
c ∈ C. (5)
[10]

Solution

(a) A + cB = (AB −1 + cIn )B, so that det (A + cB) = det(AB −1 + cIn ) det(B). Choose −c equal to
an eigenvalue of AB −1 , then det (A + cB) = 0 and therefore A + cB is not invertible.

(b) Choose A = I2 and B = [ 00 10 ], for example.

Question 4

This question was not marked.


It is still important to work through the solutions and compare them with your own attempts.

Let T : C 3 → C 3 be the linear operator defined by

T (z1 , z2 , z3 ) = (−z1 − iz3 , 0, −2iz1 + 2z3 ) .

(a) Show that T is a projection.

(b) Find a basis for the space onto which T projects.

(c) Find a basis for the space along which T projects.

4
MAT3701/201

Solution

(a)

T 2 (z1 , z2 , z3 ) = T (−z1 − iz3 , 0, −2iz1 + 2z3 )


= (z1 + iz3 − 2z1 − 2iz3 , 0, 2iz1 − 2z3 − 4iz1 + 4z3 )
= (−z1 − iz3 , 0, −2iz1 + 2z3 )
= T (z1 , z2 , z3 )

Thus T 2 = T, which means that T is a projection.

(b) The space onto which T projects is

R(T ) = {T (z1 , z2 , z3 ) : (z1 , z2 , z3 ) ∈ C 3 }


= {(−z1 − iz3 , 0, −2iz1 + 2z3 ) : (z1 , z2 , z3 ) ∈ C 3 }
= {−z1 (1, 0, 2i) − z3 (i, 0, −2) : (z1 , z2 , z3 ) ∈ C 3 }
= span{(1, 0, 2i)}, since the second vector is i times the first.

A basis is therefore {(1, 0, 2i)}.

(c) The space along which T projects is N (T ). Now,

T (z1 , z2 , z3 ) = 0 ⇔ (−z1 − iz3 , 0, −2iz1 + 2z3 ) = 0



−z1 − iz3 = 0

−2iz1 + 2z3 = 0

⇔ z1 + iz3 = 0

since the second equation is 2i times the first.


Thus,

N (T ) = {z1 (1, 0, i) + z2 (0, 1, 0) : z1 , z2 ∈ C}


= span{(1, 0, i), (0, 1, 0)}

A basis is therefore {(1, 0, i), (0, 1, 0)}.

Question 5

Let P : V → V be a projection on the vector space V over F. Prove that

(a) I − P is a projection; (2)


(b) N (I − P ) = R (P ) ; (4)
(c) R (I − P ) = N (P ) ; (4)
(d) V = R (P ) ⊕ R (I − P ) . (4)
[14]

5
Solution

(a)
(I − P )2 = I − 2P + P 2 = I − P,
hence I − P is a projection.

(b)
v ∈ N (I − P ) ⇒ (I − P )v = 0 ⇒ v = P v ⇒ v ∈ R(P ),
hence
N (I − P ) ⊆ R (P ) . (ii)
Conversely,

v ∈ R(P ) ⇒ v = P w for some w ∈ V ⇒ P v = P 2 w = P w = v ⇒ (I − P )v = 0 ⇒ v ∈ N (I − P ),

hence R (P ) ⊆ N (I − P ). This, together with (ii), implies that N (I − P ) = R (P ).

(c) Since I − P is also a projection by (a), the result follows by replacing P with I − P in (b).

(d)
I = P + (I − P ) ⇒ v = P v + (I − P )v for all v ∈ V ⇒ V ⊆ R (P ) + R (I − P ) .
Since R (P ) ⊆ V and R (I − P ) ⊆ V , we also have R (P ) + R (I − P ) ⊆ V , and therefore

V = R (P ) + R (I − P ) . (iii)

Further,

v ∈ R (P ) ∩ R (I − P ) ⇒ P v = 0 from (c) since v ∈ R (I − P ) and P v = v from (b) since v ∈ R (P )


⇒ v = 0.

Hence,
R (P ) ∩ R (I − P ) = {0}, (iv)
and it follows from (iii) and (iv) that V = R (P ) ⊕ R (I − P ).

Question 6

This question was not marked.


It is still important to work through the solutions and compare them with your own attempts.

Let T : P2 (C) → P2 (C) be the linear operator defined by

T a + bx + cx2 = 2a + (a + b) x + (a + c) x2 .


(a) Show that T satisfies the test for diagonalizability.

(b) Find a basis τ for P2 (C) consisting of eigenvectors of T, and write down [T ]τ .

6
MAT3701/201

(c) Show that 2I − T is a projection.

Solution

(a) Let β = {1, x, x2 } (the standard basis for P2 (C)).


Then,

T (1) = 2 + x + x2
T (x) = x
T x2 = x2


Thus,  
2 0 0
[T ]β =  1 1 0 
1 0 1
It follows that the characteristic polynomial of T is

p (x) = (x − 2) (x − 1)2 ,

which splits. Hence, its eigenvalues are 2, 1 and 1.


Thus, T has a repeated eigenvalue λ = 1, and
 
1 0 0
[T ]β − I =  1 0 0  (v)
1 0 0
which is of rank one. Since  
3 − rank [T ]β − I = 2
is equal to the multiplicity of λ = 1, it follows that T is diagonalisable.

(b) Considering (v) as the coefficient matrix of a system of homogeneous linear equations, it follows
that
x = 0 and y and z are arbitrary.
Hence,

E1 = {(0, y, z) : y, z ∈ C}
= span {(0, 1, 0) , (0, 0, 1)}

and therefore the eigenspace of T corresponding to λ = 1 is

E1 (T ) = span x, x2 .


Similarly,
E2 (T ) = span 1 + x + x2 .


Thus,  
1 0 0
[T ]τ =  0 1 0 
0 0 2

7
where
τ = x, x2 , 1 + x + x2 .


(c)

(2IP2 (C) − T )2 = 2IP2 (C) − T (vi)


⇔ (2IP2 (C) − T )2 τ = 2IP2 (C) − T τ
   
 2  
⇔ 2IP2 (C) − T τ = 2IP2 (C) − T τ (vii)

Since
   
2IP2 (C) − T τ = 2IP2 (C) τ − [T ]τ
= 2I3 − [T ]τ
 
1 0 0
=  0 1 0 ,
0 0 0

it follows by direct multiplication that (vii) holds and therefore also (vi).

Question 7

Explain in each of the following cases whether or not W is T –invariant under the given linear operator
T.
   
1 1 1 2
(a) W = span , and T : M2×2 (C) → M2×2 (C) defined by T (X) = AX, where
 −1
 1 −2 1
1 1
A= ; (5)
−1 1

(b) W = {a + bx + cx2 : a + b + c = 0} and T : P2 (C) → P2 (C) defined by

T a + bx + cx2 = (a + b) + (b + c) x + (c + a) x2 ;


(5)

(c) W = N (T ) + R (T ) and T : C 3 → C 3 defined by

T (a, b, c) = (a + b, b + c, a − c) .

(4)
[14]

Solution

(a) Since
         
1 1 1 1 0 2 1 1 1 2
T =A = = −2 +2 ∈W
−1 1 −1 1 −2 0 −1 1 −2 1

8
MAT3701/201

and
       
1 2 −1 3 1 1 1 2
T = = −5 +4 ∈W
−2 1 −3 −1 −1 1 −2 1
it follows that W is a T-invariant subspace.

(b) Suppose a + bx + cx2 ∈ W, then a + b + c = 0. Hence,

T a + bx + cx2 = (a + b) + (b + c) x + (c + a) x2 lies in W,


since
(a + b) + (b + c) + (c + a) = 2 (a + b + c) = 0.
Thus, W is a T-invariant subspace.

(c) W is a T-invariant subspace, since N (T ) and R (T ) are invariant subspaces for any linear trans-
formation T, and hence also
W = N (T ) + R (T ) .

Question 8

This question was not marked.


It is still important to work through the solutions and compare them with your own attempts.

Let T : V → V be a linear operator on a finite–dimensional vector space V over F, and suppose


v ∈ V is such that
v, T (v) , T 2 (v)
is linearly independent and
T 3 (v) = v + T (v) + T 2 (v) . (v)
Let
W = span v, T (v) , T 2 (v) .


(a) Show that W is T –invariant.

(b) Find the characteristic polynomial of TW .

(c) Explain whether TW : W → W is one–to–one.

(d) Explain whether TW : W → W is onto.

Solution

(a) It suffices to show that the image under T of each generator of W is again in W . This is clearly
the case for the first two generators of W . For the third generator we have

T (T 2 (v)) = T 3 (v) = v + T (v) + T 2 (v) , from (v)

9
and since the right-hand side lies again in W , it follows that W is T –invariant.

(b) It follows from (v) that the characteristic polynomial of TW is

cTW (t) = (−1)3 (t3 − t2 − t − 1).

(c) TW is one-to-one, since it follows from (b) that 0 is not an eigenvalue of TW .

(d) Since TW is one-to-one according to (c) and since W is finite dimensional, it follows that TW is
also onto.

10
MAT3701/202/2/2013

Tutorial Letter 202/2/2013

LINEAR ALGEBRA
MAT3701

Semester 2

Department of Mathematical Sciences

This tutorial letter contains solutions


for assignment 02.

BAR CODE

university
Learn without limits.
Open Rubric of south africa
ASSIGNMENT 02
Solution
UNIQUE ASSIGNMENT NUMBER: 392986

Please note that we will only mark a selection of the questions. It is therefore in your own interest to
do all the questions. The fact that a question is not marked does not mean that it is less important
than one that is marked. Worked solutions to all the questions will be sent to all students shortly
after the due date. For this assignment, questions 3, 5, 6, and 8 will be marked.

Question 1

This question was not marked.


It is still important to work through the solutions and compare them with your own attempts.

Let A ∈ Mn×m (C) such that rank(A) = m, and let P = A (A∗ A)−1 A∗ . Prove that

(a) A∗ A is invertible;
(b) Lp is an orthogonal projection;
(c) Lp projects onto the column space of A.

Solution

(a) Note that A∗ A is an m × m matrix. To show that A∗ A is invertible, it suffices to show that
N (A∗ A) = {0}.
Now,

A∗ Ax = 0 ⇒ x∗ A∗ Ax = 0
⇒ (Ax)∗ Ax = 0
⇒ Ax = 0
⇒ x = 0, since rank (A) = m.

Thus, N (A∗ A) = {0}, and therefore A∗ A is invertible.

(b) Since

P 2 = [A (A∗ A)−1 A∗ ][A (A∗ A)−1 A∗ ]


= A[(A∗ A)−1 A∗ A] (A∗ A)−1 A∗
= A (A∗ A)−1 A∗
= P,

2
MAT3701/202

it follows that
L2P = LP 2 = LP ,
thus LP is a projection. But P is also self-adjoint, since
P ∗ =[A (A∗ A)−1 A∗ ]∗
=[A∗ ]∗ [(A∗ A)−1 ]∗ A∗
=A[(A∗ A)∗ ]−1 A∗
=A (A∗ A)−1 A∗
=P,
and therefore
(LP )∗ = LP ∗ = LP .
Thus, LP is also self-adjoint, and therefore an orthogonal projection.

(c) LP projects onto R(LP ) = R(P ). Note that


R(P ) ⊆ R(A), (i)
since
P v = A (A∗ A)−1 A∗ v = A((A∗ A)−1 A∗ v) ∈ R(A)
for all v ∈ C m . Since
P A = A (A∗ A)−1 A∗ A = A,
it also follows that
R(A) ⊆ R(P ). (ii)
Thus, it follows from (i) and (ii) that R(LP ) = R(A), and therefore Lp projects onto the column space
of A.

Question 2

This question was not marked.


It is still important to work through the solutions and compare them with your own attempts.

Let T : C 3 → C 3 denote the orthogonal projection on


n √   √ o
W = span 1, 2, i , 0, 2, 0 .

Find T (z1 , z2 , z3 ) .

Solution

We may solve the problem by first using Gram-Schmidt to find an orthonormal basis for W , and then
proceed from there to find the formula for T . Alternatively, it follows from Question 1 that T = LP ,
where  
√1 √0
P = A (A∗ A)−1 A∗ with A =  2 2
i 0

3
(the columns of A consist of the generators of W ). Since
 ∗  −1
1
√ √ 0 1
√ √0
(A∗ A)−1 =  2 2  2 2
i 0 i 0
 −1
4 2
=
2 2
 
1 2 −2
= ,
4 −2 4
we have
 √
 
√1 √0 1 2 −2 1
 
√ 2 −i
P =  2 2
4 −2 4 0 2 0
i 0
 
1 0 −i
1
= 0 2 0 ,
2
i 0 1

so that       
z1 1 0 −i z1 z1 − iz3
1 1
T (z1 , z2 , z3 ) = LP (z1 , z2 , z3 ) = P z2  = 0 2 0  z2  =  2z2  ,
2 2
z3 i 0 1 z3 iz1 + z3
that is,
1
T (z1 , z2 , z3 ) = (z1 − iz3 , 2z2 , iz1 + z3 ).
2

Question 3

Consider the following system of linear equations over R

x 1 + x2 + x3 = 1
x1 − x2 = 1
2x2 + x3 = 1

(a) Show that the system has no solution. (2)

(b) Find the least squares approximate solution to the system. (10)

[12]

Solution (a) Adding the last two equations yields

x1 + x2 + x3 = 2,

which shows that the system has no solutions when compared with the first equation.

4
MAT3701/202

(b) Use the method described in Section 12.6 of the Study Guide. Denote the coefficient matrix and
constant column of the system by
   
1 1 1 1
A = 1 −1 0 and y = 1 ,
  
0 2 1 1

respectively. Then    
2 0 1 2
∗ ∗
A A = 0 6 3 and A y = 2 ,
  
1 3 2 2
and it follows that the least squares approximate solution x0 is obtained from A∗ Ax0 = A∗ y, which
yields  
1
x0 =  31  (show).
0

Question 4

This question was not marked.


It is still important to work through the solutions and compare them with your own attempts.

For a linear operator T on an inner product space V, prove that T ∗ T = T0 implies T = T0 . Is the
same result true if we assume that T T ∗ = T0 ?

Solution

For all v ∈ V ,
hT ∗ T v, vi = h0, vi = 0 ⇒ hT v, T vi = 0 ⇒ T v = 0,
so that T = T0 .
If T T ∗ = T0 , then it follows from above that T ∗ = T0 , and therefore

T = (T ∗ )∗ = T0∗ = T0 .

Question 5

Let V be a complex inner product space, and let T be a linear operator on V. Define
1 1
T1 = (T + T ∗ ) and T2 = (T − T ∗ ) .
2 2i
(a) Prove that T1 and T2 are self–adjoint and that T = T1 + iT2 . (5)

5
(b) Suppose also that T = U1 + iU2 , where U1 and U2 are self–adjoint. Prove that U1 = T1 and
U2 = T2 . (5)

(c) Prove that T is normal if and only if T1 T2 = T2 T1 . (5)

[15]

Solution

(a)
1 1 1
T1∗ = (T + T ∗ )∗ = (T ∗ + (T ∗ )∗ ) = (T ∗ + T ) = T1 ,
2 2 2
and
1 −1 ∗ −1 ∗
T2∗ = (T − T ∗ )∗ = (T − (T ∗ )∗ ) = (T − T ) = T2 ,
2i 2i 2i
thus T1 and T2 are self-adjoint.
Further,
1 1
T1 + iT2 = (T + T ∗ ) + i (T − T ∗ ) = T,
2 2i
as required.

(b) (
T + T ∗ = 2U1 ⇒ T1 = U1
T = U1 + iU2 ⇒ T ∗ = U1∗ − iU2∗ = U1 − iU2 ⇒
T − T ∗ = 2iU2 ⇒ T2 = U2
as required.

(c)
T T ∗ = (T1 + iT2 )(T1 − iT2 ) = T12 − iT1 T2 + iT2 T1 + T22
and
T ∗ T = (T1 − iT2 )(T1 + iT2 ) = T12 + iT1 T2 − iT2 T1 + T22 .
Therefore,
T T ∗ = T ∗ T ⇔ −T1 T2 + T2 T1 = T1 T2 − T2 T1 ⇔ T1 T2 = T2 T1 ,
and the result follows.

Question 6

Express the rigid motion f : R2 → R2 defined by

f (a, b) = (1 − a, 2 + b)

in the form f = g ◦ T , where g is a translation and T an orthogonal operator on R2 . Find the vector
v0 by which g translates. If T is a rotation, find the angle θ through which it rotates; and if T is a
reflection about a line L through the origin, find the equation of L and the angle α it makes with the
positive x–axis. [10]

Solution

6
MAT3701/202

f (a, b) = (1, 2) + (−a, b)


= (g ◦ T ) (a, b) ,

where
g (a, b) = (1, 2) + (a, b) ,
hence the translation is by (1, 2) = f (0) , and

T (a, b) = (−a, b)

Let β be the standard ordered basis for R2 , then


 
−1 0
[T ]β = ;
0 1

hence T is orthogonal, since


[T ]∗β [T ]β = [T ]β [T ]∗β = I2
Also,  
det [T ]β = −1,
so T is a reflection about a line L through the origin.
Since    
−1 0 cos π sin π
[T ]β = = ,
0 1 sin π − cos π
π
L makes an angle α = 2
with the positive x–axis. Hence, the equation of L is

x=0

Alternatively, L can be determined as follows. Note that L is given by the eigenspace E1 (T ) of T


corresponding to 1. To compute E1 (T ), the corresponding homogeneous system to solve is, in matrix
form,  
−2 0
[T ]β − I = ;
0 0
hence,
E1 [T ] = span {(0, 1)}
Thus, the equation of L is x = 0, and the angle it makes with the positive x–axis is α = π2 .

Question 7

This question was not marked.


It is still important to work through the solutions and compare them with your own attempts.

Let Hu , u a unit vector, be a Householder operator on a finite–dimensional inner product space


V, i.e., Hu (x) = x − 2 hx, ui u for all x ∈ V. Prove the following results.

7
(a) Hu is linear.

(b) Hu (x) = x if and only if x is orthogonal to u.

(c) Hu (u) = −u.

(d) Hu∗ = Hu and Hu2 = I, and hence Hu is a unitary [orthogonal] operator on V .

(Note: If V is a real inner product space, then in the language of Section 6.11, Hu is a reflection.)

Solution

(a) Hu is a linear operator, since for x, y ∈ V and a ∈ F ,

Hu (x + y) = x + y − 2hx + y, uiu
= x + y − 2(hx, ui + hy, ui)u, property of the inner product
= (x − 2hx, uiu) + (y − 2hy, uiu)
= Hu (x) + Hu (y),

and
Hu (ax) = ax − 2hax, uiu = ax − 2ahx, uiu = a(x − 2hx, uiu) = aHu (x).

(b)

Hu (x) = x
⇔ x − 2 hx, ui u = x
⇔ hx, ui u = 0
⇔ hx, ui = 0, since u 6= 0
⇔ x is orthogonal to u

(c)
Hu (u) = u − 2 hu, ui u = −u, since hu, ui = 1.

(d) For all x, y ∈ V ,

hHu∗ (x), yi = hx, Hu (y)i


= hx, y − 2 hy, ui ui
= hx, yi + hx, −2 hy, ui ui
= hx, yi − 2hy, uihx, ui
= hx, yi − 2hu, yihx, ui
= hx, yi + h−2hx, uiu, yi
= hx − 2 hx, ui u, yi,

thus
Hu∗ (x) = x − 2 hx, ui u = Hu (x), i.e., Hu∗ = Hu .

8
MAT3701/202

Finally,

Hu2 (x) = Hu (x − 2 hx, ui u)


= x − 2 hx, ui u − 2hx − 2 hx, ui u, uiu
= x − 2 hx, ui u − 2(hx, ui − 2 hx, ui)u, since hu, ui = 1
=x
= I(x),

that is,
Hu2 = I.

Question 8

This question was not marked.


It is still important to work through the solutions and compare them with your own attempts.
 
2 −2
Let A = and suppose k·k denotes the Euclidean norm.
1 1

(a) Find kAk , kA−1 k and cond (A) . (9)


e such that Ax = b, kbk = 1, and kb − Ae
(b) Suppose that we have vectors x and x xk ≤ 0.001. Use (a)
−1
to determine upper bounds for ke x − A bk (the absolute error) and ke x − A−1 bk / kA−1 bk (the
relative error). (4)

[13]

Solution

(a)     
∗ 2 1 2 −2 5 −3
A A= = ,
−2 1 1 1 −3 5
thus
λ−5 3
det(λI − A∗ A) =
3 λ−5
= (λ − 5)2 − 32
= (λ − 8)(λ − 2),

so that
√ √ 1
kAk = 8 = 2 2, A−1 = √ , and cond (A) = kAk · A−1 = 2.
2
(b)
1
e − A−1 b = A−1 (Ae
x x − b) ≤ A−1 kAe
x − bk ≤ √ (0.001),
2

9
and
x − A−1 bk
ke ke
x − xk kAe
x − bk
−1
= ≤ cond (A) ≤ 0.002.
kA bk kxk kbk

10

You might also like