Student Solution Manual for

Introduction to Linear Algebra

Geza Schay and Dennis Wortman


1.1.1. P R = r p, P Q = q p, and QP = p q.
1 1
QC = 2 QP = 2 p 12 q, P C = 12 P Q = 12 q 12 p, and OC = 12 r = 12 p+ 12 q.
1.1.3. p + q = (2, 3, 1) + (1, 2, 2) = (3, 5, 1).

1.1.5. r = p + q, 2r = 2(p + q) = 2p + 2q.

2q 2r

r
q
p 2p

Figure 0.1.

3
1.1.7. M = i=1 mi = 2 + 3 + 5 = 10 and
3
1  1 3 6
r= mi ri = [2(2, 1, 4)+3(1, 5, 6)+5(2, 5, 4)] = ( , , 1).
M i=1 10 10 5

2
 1.1.9. The center R of the side BC has position vector r = 12 (b + c) and
the vector from R to A is a r. Thus the point P , 1/3 of the way from R to
A, has position vector
p = 12 (b + c)+ 13 (a r) = 12 b+ 12 c+ 13 a 13 12 (b + c) = 13 (a + b + c).

1.1.11. Opposite edges are those with no common vertex; for example, the
one from A to B, and the one from C to D. The midpoint of the former has
position vector p = 12 (a + b) and of the latter, q = 12 (c + d). The halfway
point M on the line joining those midpoints has the position vector

 
1 1 1 1 1
m = p+ (q p) = (p + q) = (a + b)+ (c + d)
2 2 2 2 2
1
= (a + b + c + d),
4

which is the position vector of the centroid.

3
 1.2.1. Let p = (5, 5) and q = (1, 7). Then
a. p + q = (5 + 1, 5 7) = (6, 2), p q = (5 1, 5 + 7) = (4, 12).

4
 b.


c. |p| =  52 + 52 = 50, |q| = 12 + (7)2 = 50,

|p + q| = 62 + (2)2 = 40, |p q| = 42 + 122 = 160.
d. |p + q|2 = 40, |p|2 +|q|2 = 50 + 50 = 100, and so |p + q|2 = |p|2 +|q|2 .

1.2.3. Let S be the midpoint of QP and T the midpoint of QR. Then
1 1 1 1
ST = QT QS = QR QP = (QR QP ) = P R.
2 2 2 2

5
 1.2.5.
pq 1 3 + (2) 5 + 4 2 1
cos = = = 0.0354.
|p||q| 21 38 798

Hence 1.535 radians 87.9

1.2.7.

The above diagram illustrates the regular octahedron inscribed in the unit
cube, joining the centers of the sides of the cube by line segments. Using the
notation from the diagram, we find four distinct types of angles formed by the
sides of the octahedron, represented, for example, by P AB, DAB, P AQ
and AP C. To determine those angles, first calculate the coordinates of the
relevant points:

1 1 1 1 1 1 1 1 1 1 1 1
P ( , , 1), A( , 0, ), B(1. , ), D(0, , ), Q( , , 0), C( , 1, ).
2 2 2 2 2 2 2 2 2 2 2 2
Hence
1 1 1 1
AP = (0, , ) and AB = ( , , 0),
2 2 2 2
6
and thus

AP AB 1/4 1
cos P AB = = = ; P AB = 60
|AP ||AB| 1/ 2 2

Also, AD = ( 12 , 12 , 0) and thus

AB AD
cos DAB = = 0; DAB = 90
|AB||AD|

Also AQ = (0, 12 , 12 ) and

AP AQ
cos P AQ = = 0; P AQ = 90
|AP ||AQ|

Lastly,
1 1 1 1
P A = (0, , ) and P C = (0, , ),
2 2 2 2
and so

PA PC
cos AP C = = 0; AP C = 90
|P A||P C|
Remark: we used vector methods in the above solution, but the angles could
also have been determined using simple geometric properties.
1.2.9. The parallel component is

pq 24 9 + 4 19
p1 = q = (12, 3, 4) = (12, 3, 4)
|q|2 122 + 32 + 42 169
and the perpendicular component is

19 110 564 93
p2 = p p1 = (2, 3, 1) (12, 3, 4) = ( , , ).
169 169 169 169
1.2.11. Call the end points of the given diameter A and B. Then, with
7
the notation of Figure 1.20,

AP BP = (p r) (p + r) = p p r r = |p|2 |r|2 = 0.

The last equality holds because p and r are both radius vectors, and so their
lengths are equal.
1.2.13.
a. Note first that

|p + q|2 = (p + q) (p + q) = p p+2p q + q q p p+2|p q|+q q.

In turn, by Cauchys inequality,

p p + 2|p q| + q q p p + 2|p||q| + q q.

Since p p = |p|2 and q q = |q|2 , we have

p p + 2|p||q| + q q = |p|2 + 2|p||q| + |q|2 = (|p| + |q|)2 .

Finally, this chain of steps implies that |p + q|2 (|p| + |q|)2 , from which
the Triangle Inequality follows by taking square roots of both sides.
b. In Cauchys inequality, equality occurs when q = 0 or p = q, for any
, but the first inequality above becomes an equality for p = q only if
0. Thus, the chain of steps above remains true with equality replacing
the inequalities if and only if the vectors p and q are parallel and point in the
same direction.
1.2.15.
a.

p = p1 +p2 = |p| cos i+|p| sin j = (|p| cos , |p| sin ) = |p|(cos , sin ).

b.
p |p|(cos , sin )
up = = = (cos , sin ).
|p| |p|

8

1.2.17. If p = (3, 4, 12), then |p| = 9 + 16 + 144 = 13 and therefore
pi p1 3 pj p2 4
cos 1 = = = , cos 2 = = = ,
|p| |p| 13 |p| |p| 13
and
pk p3 12
cos 3 = = = .
|p| |p| 13
1.2.19. We have to verify that the first three parts of Theorem 1.2.2
remain valid for the product defined by p q = 2p1 q1 + p2 q2 :
1. p q =2p1 q1 + p2 q2 = 2q1 p1 + q2 p2 = q p,
2. p (q + r) =2p1 (q1 + r1 ) + p2 (q2 + r2 ) = 2p1 q1 + 2p1 r1 + p2 q2 + p2 r2
=2p1 q1 + p2 q2 + 2p1 r1 + p2 r2 = p q + p r,
3. c(p q) = c(2p1 q1 + p2 q2 ) = 2(cp1 )q1 + (cp2 )q2 = (cp) q and
c(p q) = c(2p1 q1 + p2 q2 ) = 2p1 (cq1 ) + p2 (cq2 ) = p (cq).

If we stretch the first component of every vector by a factor of 2, then
this product applied to the original vectors gives the lengths of and the angles
between the altered vectors the same way as the standard dot product would
when applied to the altered vectors.
1.3.1. The vector parametric form of the equation of the line is p = p0 +
tv = (1, 2, 4) + t(2, 3, 5) = (1 + 2t, 2 + 3t, 4 5t), and thus the scalar
parametric form is x = 1+2t, y = 2+3t, z = 45t, and the nonparametric
form is x1
2
= y+2
3
= z4
5
.

1.3.3. A direction vector of the line through P0 and P1 is v = P0 P 1 =
9
(5 7, 6 (2), 3 5) = (2, 8, 8). Hence the vector parametric form
of the equation of the line is p = p0 + tv = (7, 2, 5) + t(2, 8, 8) =
(72t, 2+8t, 58t), and thus the scalar parametric form is x = 72t, y =
2 + 8t, z = 5 8t, and the nonparametric form is x7 2
= y+2
8
= z5
8
.
1.3.5. A direction vector of the line through the given points P0 and P1

is v = P0 P 1 = (1 1, 2 (2), 3 4) = (0, 0, 7). Hence the vector
parametric form of the equation of the line is p = p0 + tv = (1, 2, 4) +
t(0, 0, 7) = (1, 2, 4 7t), and thus the scalar parametric form is x = 1 ,
y = 2, z = 4 7t, and the nonparametric form is x = 1, y = 2.
1.3.7. As noted in the given hint, the direction vector w of the line is
orthogonal to the normal vectors to the two planes, u = (3, 4, 3) and v =
(0, 0, 1). Proceeding as in Example 1.3.7 in the text, the equation of the plane
through the origin determined by the vectors u and v can be expressed as
p = su + tv, where s and t are parameters, or, in scalar form, as x = 3s, y =
4s, z = 3s+t. Eliminating the parameters results in the equation 4x+3y =
0 (the variable z may take any value since the parameter t is free), and thus
w = (4, 3, 0). Therefore, the vector parametric form of the equation of the
line is p = p0 + tw = (5, 4, 8) + t(4, 3, 0), and thus the scalar parametric
form is x = 5 + 4t, y = 4 + 3t, z = 8, and the nonparametric form is
x5
4
= y43
, z = 8.

10
 1.3.9. a.

b. The t-scales on the two lines have their zeros at different points. Or,
alternatively, if t denotes time, then the two equations describe two moving
points that are at the intersection at different times.
c. p = (2, 6) + s(2, 3) and p = (2, 6) + s(1, 4). (Alternatively, any
nonzero scalar multiples of the direction vectors (2, 3) and (1, 4) will do
in their places.)
1.3.11. The vector p = ra + sb + tc represents the point A when r = 1
(and s = t = 0), the point B when s = 1, and the point C when t = 1.
When r = 0, p represents a point on the line segment joining B and C (by
the results of Exercise 1.3.10), and similarly for the cases s = 0 and t = 0, p
represents a point on the other two sides of the triangle formed by A, B, and
C.
When none of the variables r, s, or t is 0 or 1, then p represents a point
in the interior of that triangle. This can be seen by first noting that since
r = 1 s t, the equation p = ra + sb + tc can be expressed in the form
p = (1 s t)a + sb + tc = a + s(b a) + t(c a) Now, let D be the
point on side AB represented by d = a+s(b a), and let E be the point on
side BC represented by e = a+s(b a) + (1 s)(c a) = sb + (1 s)c.
11
 t t
Then p = d + 1s (1 s)(c a) = d + 1s (e d), and thus P is on the
t
line segment joining D and E, 1s of the way from D towards E (note that
t
1s
< 1 since s + t < 1).
1.3.13. The vector p0 = (3, 2, 1) must lie in the required plane, and so
we may choose u = p0 . For v we may choose the same v that we have for
the given line: v = (2, 1, 3), and for the fixed point of the plane we may
choose O. Thus the parametric equation can be written as p = s(3, 2, 1) +
t(2, 1, 3).
The nonparametric equations are obtained by eliminating s and t from x =
3s+2t, y = 2s+t, z = s3t. This elimination results in 5x+11y+7z = 0.
1.3.15. In this case it is easier to start with the nonparametric form of the
equation of the plane. Since the required plane is to be orthogonal to the line
given by p = (3, 2, 1) + t(2, 1, 3), we may take its normal vector to be the
direction vector of the line, that is, take n = (2, 1, 3). Next, we may take

p0 = OP 0 = (5, 4, 8), and therefore the equation n p = n p0 becomes
(2, 1, 3) (x, y, z) = (2, 1, 3) (5, 4, 8) = 38, that is, 2x + y 3z = 38.
To write parametric equations, we need to determine two nonparallel vec-
tors that lie in the plane. To do so, note that the points (19, 0, 0) and (0, 38, 0)
lie in the plane and thus u = (19, 0, 0) (5, 4, 8) = (14, 4, 8) and
v = (0, 38, 0) (5, 4, 8) = (5, 34, 8) are nonparallel vectors in the plane.
Hence p = (5, 4, 8) + s(14, 4, 8) + t(5, 34, 8) is a parametric vector
equation of the plane. (Note that this answer is not unique: there are infinitely
many other correct answers depending on the choices for p0 , u and v.)
1.3.17. In this case it is again easier to start with the nonparametric form
of the equation of the plane. Since normal vectors to parallel planes must be
parallel, we may use as the normal vector n to the required plane that of the
given plane, namely n = (7, 1, 2). Also, because P0 (5, 4, 8) is a point in the

required plane, we may take p0 = OP 0 = (5, 4, 8), and hence the equation
n p = n p0 becomes (7, 1, 2) (x, y, z) = (7, 1, 2) (5, 4, 8) = 23, that is,
7x + y + 2z = 23.
To write parametric equations, choose any two distinct vectors a and b,
other than p0 , whose components satisfy the above equation, and let u = a p0
and v = b p0 . Since such vectors a and b represent points of the plane, u
and v are nonzero vectors lying in the plane. We need to be careful to choose
the vectors a and b so that u and v are not parallel. For instance, if we

12
choose a = (0, 23, 0) and b = (1, 16, 0), then u = (0, 23, 0) (5, 4, 8) =
(5, 19, 8) and v = (1, 16, 0) (5, 4, 8) = (4, 12, 8). Hence we conclude
that p = (5, 4, 8) + s(5, 19, 8) + t(4, 12, 8) is a parametric vector equa-
tion of the plane. (As a check, notice that s = 1, t = 0 give p = a, and
s = 0, t = 1 give p = b.)
1.3.19. We may choose u = p1 p0 = (1, 6, 3)(5, 4, 8) = (4, 2, 5)
and v = p2 p0 = (7, 2, 5) (5, 4, 8) = (2, 6, 13). Thus we obtain
p = (5, 4, 8) + s(4, 2, 5) + t(2, 6, 13) as a parametric vector equation of
the plane. The corresponding scalar equations are

x = 5 4s + 2t, y = 4 + 2s 6t, z = 8 + 5s + 13t.

To obtain a nonparametric equation, eliminate s and t from the last three

equations as in Example 1.3.5. Thus

28(x 5) + 31(y 4) + 10(z + 8) = 0.

or, equivalently,
28x + 31y + 10z = 184.

1.3.21. We can proceed as in Example 1.3.3 in the text to decompose the

vector equation (5, 4, 1) + s(2, 1, 7) = (9, 9, 2) + t(2, 4, 5) into
three scalar equations:
5 + 2s = 9 + 2t, 4 + s = 9 4t, 1 7s = 2 + 5t.
Solving this system yields s = 3 and t = 4. Hence the lines intersect at the
point (1, 7, 22).
1.3.23. In terms of scalar components, the equation of the given line is
x = 3 3s, y = 2 + 5s, z = 6 + 7s.
Substituting these equations into that of the given plane results in 3(3 3s) +
2(2 + 5s) 2(6 + 7s) = 3, or, simplifying, 13s = 10; the solution is
s = 10
13
. Hence the point of intersection is (69/13, 76/13, 8/13).
1.3.25. Rewriting the equation of the given line in components and re-
placing the parameter s by r yield
13
 x = 3 + 7r, y = 2 5r, z = 4 + 4r
and rewriting the equation of the given plane in components gives
x = 3s + 2t, y = 2 3t, z = 1 + 3s + 4t.
Combining those sets of equations and simplifying result in the system
7r + 3s 2t = 3

5r 3t = 4

4r 3s 4t = 5.

The solution of this system is r = 6 (and s = 49/9 and t = 34/3), so

the point of intersection is (39, 32, 28).

1.3.27. Pick a point P on the plane, say, P (0, 5, 0). Then P P 0 =
(3, 1, 0). From the given equation, n = (0, 1, 2). Thus, as in Example
1.3.6,

|P P 0 n| |3 0 + (1) 1 + 0 (2)| 1
D= = = .
|n| 0+1+4 5
1.3.29. First find the plane through O that is parallel to the direction
vectors of the lines, as in Example 1.3.7, p = s(4, 0, 3) + t(5, 0, 2), or, in
scalar parametric form, x = 4s + 5t, y = 0, z = 3s 2t. The equation
of the plane in scalar form is y = 0, from which we obtain a normal vector,
n = (0, 1, 0), to the plane, and hence also to the two given lines.
The point P (2, 1, 5) lies on the first line and Q(0, 2, 3) lies on the second

one. Thus P Q = (2, 3, 2), and as in Example 1.3.6,

|P Q n| |(2) 0 + (3) 1 + (2) 0|
D= = = 3.
|n| 0+1+0
1.3.31. Following the hint given in Exercise 1.3.30, let Q be the point

(3, 2, 4) on the line L, obtained by setting s = 0 in the equation, so QP 0 =
14
(1, 2, 4) (3, 2, 4) = (2, 4, 8). The direction of the line L is u =

(7, 5, 4), and thus the component of QP 0 parallel to L is given by.

 
u u 38 19
QP 0 = (7, 5, 4) = (7, 5, 4).
|u| |u| 90 45

Then the component of QP 0 orthogonal to L is
19 1
(2, 4, 8) (7, 5, 4) = (223, 85, 284),
45 45
1 
and therefore the distance from P0 to P is , 45 (223, 85, 284) 8.24.

1.3.33. Let P0 denote a point on the plane S, and let p0 = OP 0 denote
the radius vector of P0 . Then d = n p0 and, since |n| = 1, we note that
f (q) = n q d = n q n p0 = n (q p0 ) = |q p0 | cos ,
where denotes the angle between the vectors n and q p0 . Thus f(q) =
|q p0 || cos | is the length of the component of q p0 perpendicular to
the plane S; that is, it is the distance between S and Q. We also note that
cos 0 if 0 2 , when n points from S towards Q, and cos 0 if

2
, when n points from Q towards S.

1.3.35. Applying the result of Exercise 1.3.33, we have q = OP 0 =
(3, 4, 0), n = (0,1,2)
0+1+4
= (0,1,2)

5
, and hence d = 55 , determined by dividing

both sides of the equation of the given plane by 5. Therefore, f (q) =
n q d = 45 55 = 15 and thus D = 15 .
1.3.37. In the solution to Exercise 1.3.29, it was determined that v =
(0, 1, 0) is orthogonal to the given lines. Thus an equation of the plane con-
taining v and the line p = (2, 1, 5) + s(4, 0, 3) is given by p = (2, 1, 5) +
s(4, 0, 3) + t(0, 1, 0). In terms of scalar components: x = 2 4s, y =
1+t, z = 5+3s. The scalar form of the equation p = (0, 2, 6)+r(5, 0, 2)
of the second line, is then x = 5r, y = 2, z = 6 2r. At the point
of intersection of the plane S and that second line, we have 2 4s = 5r,
1 + t = 2, 5 + 3s = 6 2r, for which the solution is s = 1/7, t = 3,
and r = 2/7. Thus (10/7, 2, 38/7) is the point of intersection of S and
the second line. Therefore the equation of the normal transverse L is p =
(10/7, 2, 38/7) + t(0, 1, 0). It is a line that intersects both the given lines
15
and is orthogonal to both of them.

2.1.1. First row-reduce the corresponding augmented matrix to echelon

form:

2 2 3  0
1 5 2  1
4 0 6  2
 
2 2 3  0 2 2 3  0
0 4 7/2  1 0 4 7/2  1 .
0 4 0  2 0 0 7/2  2

Next apply back substitution to solve the corresponding system. The last
row corresponds to the equation 72 x3 = 1, so x3 = 27 . The second row
corresponds to the equation 4x2 + 72 x3 = 1 and thus x2 = 78 x3 + 14 = 12 .
Finally, the first row corresponds to 2x1 + 2x2 3x3 = 0, from which we
obtain x1 = x2 + 32 x3 = 1314
. Hence, in vector form the solution is

13/14
x = 1/2 .
2/7

2.1.3. First row-reduce the corresponding augmented matrix to echelon

form:
     
2 2 3  0 2 2 3  0
.
1 5 2  1 0 4 7/2  1

Next apply back substitution to solve the corresponding system. The sec-
ond row corresponds to the equation 4x2 + 72 x3 = 1. Here x3 is free; set
x3 = t. Then we obtain 4x2 + 72 t = 1; hence x2 = 14 78 t. Finally, the
first row corresponds to 2x1 + 2( 14 78 t) 3t = 0, from which we obtain
x1 = 14 + 19
8
t. Hence, in vector form the solution is

19/8 1/4
x = t 7/8 + 1/4 .
1 0

16
 2.1.5. First row-reduce to echelon form:


1 0 1  0

2 3 1  0

6 6 0  0
 
1 0 1  0 1 0 1  0

0 3 
3  0 0 3 3  0 .

0 6 6  0 0 0 0  0

Next apply back substitution to solve the corresponding system. The

last row just gives the consistent trivial equation 0 = 0. The second row
corresponds to the equation 3x2 3x3 = 0. Here x3 is free; set x3 = t.
Then we obtain 3x2 3t = 0; hence x2 = t. Finally, the first row corre-
sponds to x1 x3 = 0, from which we obtain x1 = t. Thus the solution is
x1 = x2 = x3 = t or in vector form,

1
x = t 1 .
1

2.1.7

 
1 0 1  1 1 0 1  1
 
2 3 1  0 0 3 3  2
 
6 6 0  0 0 6 6  6

1 0 1  1
0 3 3  2 .
0 0 0  2

The last row of the reduced matrix corresponds to the self-contradictory equa-
tion 0 = 2, and consequently the system is inconsistent and has no solution.
17
 2.1.9. First row-reduce to echelon form:


1 4 9 2  0

2 2 6 3  0

2 7 16 3  0

r1 r1 1 4 9 2  0

r2 r2 2r1 0 6 12 7  0

r3 r3 2r1 0 1 2 1  0

r1 r1 1 4 9 2  0

r2 r3 0 1 2 1  0

r3 r2 0 6 12 7  0

r1 r1 1 4 9 2  0
r2 r2 0 1 2 1  0 .
r3 r3 6r2 0 0 0 1  0

Next apply back substitution. The last row corresponds to the equation
x4 = 0 and so x4 = 0. The second row corresponds to the equation x2
2x3 x4 = 0. Here x3 is free; set x3 = t. Then we obtain x2 = 2t.
Finally, the first row corresponds to x1 + 4x2 + 9x3 + 2x4 = 0, from which
we determine x1 = t. In vector form the solution is

1
2
x=
1 t.
0

18
 2.1.11. First row-reduce to echelon form:

3 6 1 1 5  0
1 2 2 3 3  0
4 8 3 2 1  0

r1 r2 1 2 2 3 3  0
r2 r1 3 6 1 1 5  0
r3 r3 4 8 3 2 1  0

r1 r1 1 2 2 3 3  0

r2 r2 + 3r1 0 0 5 10 14  0

r3 r3 + 4r1 0 0 5 10 13  0

r1 r1 1 2 2 3 3  0

r2 r2 0 0 5 10 14  0 .

r3 r3 r2 0 0 0 0 1  0

Next apply back substitution. The last row corresponds to the equation
x5 = 0 and so x5 = 0. The second row corresponds to the equation 5x3 +
10x4 + 14x5 = 0. Here x4 is free; set x4 = t. Then we obtain x3 = 2t.
The variable x2 is free; set x2 = s. Finally, the first row corresponds to the
equation x1 + 2x2 + 2x3 + 3x4 + 3x5 = 0, and so to x1 + 2s + 2 (2t) +
3t + 3 0 = 0. Hence x1 = 2s t. In vector form the solution is

2 1
1 0

x= 0

s + 2 t.

0 1
0 0

2.1.13.
 
3 6 1 1  5 1 2 2 3  3
 
1 2 2 3  3 r1 r2 3 6 1 1  5
 
6 8 3 2  1 6 8 3 2  1

 
1 2 2 3  3 1 2 2 3  3
 
0 0 5 10  14 r2 r3 0 4 9 16  19 .
 
0 4 9 16  19 0 0 5 10  14

19
 The last matrix is in echelon form and the forward elimination is finished.
The fourth column has no pivot and so x4 is free and we set x4 = t. Then the
 14 5x3 +10t
last row corresponds to the equation  = 14, which gives x3 = 14 5
2t.
The second row yields 4x2 + 9 5 2t  + 16t = 19, whence
 x2 = 2 t 31
1
20
.
Finally, the first row gives x1 = 3 + 2 12 t 31
20
+ 2 14
5
2t + 3t = 1
2
.
In vector form the solution is

1/2 0
31/20 1/2
x =
14/5 + 2 t.

0 1

2.1.15. The line of intersection of the first two planes is obtained from the
reduction
     
1 2 0  2 1 2 0  2
.
3 6 1  8 0 0 1  2
The last matrix represents the same planes as Equations (2.25) in the text.
The line of intersection of the last two planes is obtained from the reduc-
tion
     
3 6 1  8 1 2 1  0

1 2 1  0 3 6 1  8
     
1 2 1  0 1 2 1  0

0 0 4  8 0 0 1  2
  
1 2 0  2
,
0 0 1  2
In the last step, we added the second row to the first row. Again, the last
matrix represents the same planes as Equations (2.25) in the text.
The line of intersection of the first and last planes is obtained from the
reduction
     
1 2 0  2 1 2 0  2
,
1 2 1  0 0 0 1  2
the same planes as before.
20
 2.2.1.
       
p1 p1 0 p1 0 0
, , , .
0 p2 0 0 0 0 0 0

2.2.3. Following the given hint, reduce the augmented matrix [A|b] to
echelon form:
 
1 0 1  b1 r1 r1 1 0 1  b1
2 3 1  b2 r2 r2 + 2r1 0 3 3  b2 + 2b1
6 6 0  b3 r3 r3 + 6r1 0 6 6  b3 + 6b1

r1 r1 1 0 1  b1
r2 r2 0 3 3  b2 + 2b1 .
r3 r3 2r2 0 0 0  b3 + 6b1 2 (b2 + 2b1 )

Hence the condition for consistency is b3 + 6b1 2(b2 + 2b1 ) = 0, or

equivalently, 2b1 2b2 + b3 = 0.
2.2.5. Reduce the augmented matrix [A|b] to echelon form:
     
1 2 6  b1 r1 r1 1 2 6  b1
.
2 4 12  b2 r2 r2 + 2r1 0 0 0  b2 + 2b1

Hence the condition for consistency is b2 + 2b1 = 0.

2.2.7. The rank of a matrix is the number r of non-zero rows in an echelon
matrix obtained by the forward elimination phase of the Gaussian elimination
algorithm. The same row operations that reduce the matrix A to an echelon
matrix U will also row-reduce the augmented matrix [A|b] to an augmented
matrix [U|c], with the same echelon matrix U . Thus the system Ax = b is
consistent if and only if cr+1 = cr+2 = = 0; that is, if and only if all rows
in [U|c] below the rth row are zero rows. In this case [U |c] is an echelon
matrix too, and the number of non-zero rows in it is the rank of [A|b]. Hence
the system Ax = b is consistent if and only if the rank of [A|b] equals the
number r of non-zero rows in the matrix U , which is, by definition, the rank
of A.
2.3.1..
       
1 0 1 0 1 0 0
, , , .
0 1 0 0 0 0 0 0

21
 2.3.3. To apply the method of Gauss-Jordan elimination for solving the
given system of equations, continue row-reducing the echelon matrix obtained
in the solution of Exercise 2.1.5:
 
1 0 1  0 r1 r1 1 0 1  0
0 3 3  0 r2 r2 /3 0 1 1  0 .
0 0 0  0 r3 r3 0 0 0  0

Next, apply back substitution to the last matrix, which is in row-reduced

echelon form. The third row corresponds to the equation 0 = 0, and so the
system is consistent. The second row corresponds to the equation x2 x3 =
0; x3 is free, so set x3 = t, and thus also x2 = t. Finally, the first row
corresponds to the equation x1 x3 = 0, and hence also x1 = t.
2.3.5. To apply the method of Gauss-Jordan elimination for solving the
given system of equations, continue row-reducing the echelon matrix obtained
in the solution of Exercise 2.1.11:

1 2 2 3 3  0
0 0 5 10 14  0
0 0 0 0 1  0

r1 r1 1 2 2 3 3  0
r2 r2 /5 0 0 1 2 14/5  0
r3 r3 0 0 0 0 1  0

r1 r1 + 3r3 1 2 2 3 0  0
r2 r2 (14/5) r3 0 0 1 2 0  0
r3 r3 0 0 0 0 1  0

r1 r1 + 2r2 1 2 0 1 0  0
r2 r2 0 0 1 2 0  0 .
r3 r3 0 0 0 0 1  0

The pivots correspond to the variables x1 , x3 and x5 ; the variables x2 and

x4 are free: set x2 = s and x4 = t. Then, by back substitution, we obtain
x5 = 0 from the third row, x3 = 2t from the second, and x1 = 2s t from
the first.
2.3.7. First represent the system as an augmented matrix and row-reduce
22
it to echelon form:
     
2 3 1  4 r 1 r 1 2 3 1  4
 .
3 5 2  1 r2 r2 (3/2) r1 0 1/2 7/2  5

The second row corresponds to the equation 12 x2 + 72 x3 = 5, and the

first row to the equation 2x1 + 3x2 x3 = 4. We can get a particular solution
by setting the free variable x3 = 0. Then 12 x2 = 5, andso x2 = 10,
17
and 2x1 30 = 4, and thus x1 = 17. Hence xb = 10 is a particular

0
1 7
solution of Ax = b. Similarly, setting x3 = 1 gives 2 x2 + 2
= 5, or
28
x2 = 17, and 2x1 51 1 = 4, or x1 = 28. Thus xb =  17 is
1
another particular solution of Ax = b.
For the homogeneous equation Ax = 0 we can use the same echelon
matrix as above, except that the entries of its last column must be replaced by
zeros. The general solution is obtained by setting x3 = t, and solving for the
other variables as 12 x2 + 72 = 0, x2 = 7t, 2x 1 21t
t = 0, x1 = 11t.
11
Thus the general solution of Ax = 0 is v = t 7 .
1
Hence the general solution of the inhomogeneous equation Ax = b can
be written either as

17 11 28 11
x = 10 + t 7 or as x = 17 + s 7
0 1 1 1
The two equations represent the same line, as can be seen by setting s = t1.
2.4.1 .    
13 6 1 24
C = 2A + 3B = and D = 4A 3B = .
8 2 2 14
2.4.3.
3 6 9
AB = 2 and BA = 2 4 6 .
1 2 3

23
 2.4.5.
17 17
AB = 2 17 , while BA does not exist since B is 3 2 and A
3 26
is 3 3.
2.4.7.
AB = [(3 4 + 3 + 8), (4 4 9 20)] = [10, 37], and BA is
undefined.
2.4.9.  
1 3
AB = [1 8], (AB)C = [1, 8] = [25, 3], and
   3 0 
9 9 9 9
BC = , A(BC) = [1, 2] = [25, 3].
8 6 8 6
2.4.11. There are many possible answers; for example,
   
0 0 0 1
A= and B = .
0 1 0 0
2.4.13. a. Each of the terms in the sum a1 b1 +a2 b2 + + aP bP is the
outer product of an m 1 column vector and a 1 n row vector and is
therefore of the same mxn size as AB. Now the ith element of any column
ak is the element in the ith row kth column of A, that is, aik . Similarly, the
jth element of the row bk is bkj . Thus (ak bk )ij = aik bkj . Summing over k
from 1 to p, we get

 p
 p p
  
k k
ak b = (ak b )ij = aik bkj = (AB)ij
k=1 ij k=1 k=1

and so

p

AB = ak bk .
k=1

b. Let us expand the jth sum on the right of the equation given in part (b) of
24
the exercise:

p
a1i
i=1 a1i bij
p p p
  a2i i=1 a2i bij
ai bij = . bij = .. .
..
i=1 i=1
p .
ami i=1 ami bij

The column vector on the right is exactly the jth column of AB, and so AB
is the row of such columns for j = 1, . . . , n.
c. Let us expand the ith sum on the right of the equation given in part (c) of
the exercise:
p p
 p p p

    
aij bj = aij (bj1 bj2 . . . bjn ) = aij bj1 aij bj2 . . . aij bjn .
j=1 j=1 j=1 j=1 j=1

The row vector on the right is exactly the ith row of AB, and so AB is the
column of such rows for i = 1, . . . , m.
2.4.15. There are many possible answers; for example,
   
0 1 1 1
A= or A = .
0 0 1 1
2.4.17. (M 3 )24 = (M M 2 )24 can be computed by summing the products
of the elements of the second row of M with the corresponding elements in the
fourth column of M 2 . Thus, from Equations (2.104) and (2.105), we obtain
(M 3 )24 = 1 0 + 0 1 + 0 1 + 0 2 + 0 0 = 0. Hence there are no three-leg
flights between B and D.
2.4.19. On the one hand, using the blocks:
    
1 2 0 0 1 0 3 2
+
3 4 0 0 0 1 1 1
   
1020 1020 1 3 + 0 1 1 2 + 0 (1)
= +
30+40 30+40 0 3 + 1 1 0 2 + 1 (1)
 
3 2
= ,
1 1

and on the other hand, without the blocks:

25

  0 0
1 2 1 0 0 0

3 4 0 1 3 2
1 1
 
1020+13+01 1 0 2 0 + 1 2 + 0 (1)
=
30+40+03+11 3 0 + 4 0 + 0 2 + 1 (1)
 
3 2
= .
1 1

2.4.21. Expand the given matrix using Corollary 2.4.3 and then apply the
result of Exercise 2.4.20.
2.4.23. In order that all relevant matrices be conformable, the second
matrix should be partitioned as

1 2  1 0
2 0  3 1
.
0 0  2 3
0 0  7 4

Then, applying the result of Exercise 2.4.21, we obtain:

 
1 2  1 0 1 2  1 0
 
3 4  0 1 2 0  3 1
  = X11 X12
,
1 
0  0 0 0 
0  2 3 X21 X22
0 1  0 0 0 0  7 4

where

1 2   1 0   3 2
1 2 0 0
X11 = 3 4 + 0 1 = 11 6 ,
2 0 0 0
1 0 0 0 1 2

1 2  1 0   9 1
1 0 2 3
X12 = 3 4 + 0 1 = 2 8 ,
3 1 7 4
1 0 0 0 1 0

26
    
  1 2   0 0  
X21 = 0 1 + 0 0 = 2 0 ,
2 0 0 0

   
  1 0   2 3  
X22 = 0 1 + 0 0 = 3 1 .
3 1 7 4

Thus
 
1 2  1 0 1 2  1
 0
3 4  0 1 2 0  3 1
 
1 0  0 0 0 0  2 3

0 1  0 0 0 0  7 4

3 2  9 1
11 6  2 8
=
1  .
2  1 0
2 0  3 1

2.5.1. To find A1 , form the augmented matrix [A|I] and row-reduce it to

the form [I|A1 ], if possible.
        
2 3  1 0 2 3  1 0 2 3  1 0

4 1  0 1 0 7  2 1 0 1  2/7 1/7

     
2 0  1/7 3/7 1 0  1/14 3/14
.
0 1  2/7 1/7 0 1  2/7 1/7

Hence
 
1 1 1 3
A = .
14 4 2

2.5.3. To find A1 , form the augmented matrix [A|I] and row-reduce it to

the form [I|A1 ], if possible.
27
 
2 3 5  1 0 0
4 1 1  0 1 0
3 2 2  0 0 1

r1 r1 /2 1 3/2 5/2  1/2 0 0
r2 r2 4 1 1  0 1 0
r3 r3 3 2 2  0 0 1

r1 r1 1 3/2 5/2  1/2 0 0
r2 r2 4r1 0 7 9  2 1 0

r3 r3 3r1 0 5/2 19/2  3/2 0 1

r1 r1 1 3/2 5/2  1/2 0 0
r2 r2 /7 0 1 9/7  2/7 1/7 0
r3 2r3 /5 0 1 19/5  3/5 0 2/5

r1 r1 3r2 /2 1 0 4/7  1/14 3/14 0
r2 r2 0 1 9/7  2/7 1/7 0
r3 r3 r2 0 1 88/35  11/35 1/7 2/5

r1 r1 1 0 4/7  1/14 3/14 0
r2 r2 0 1 9/7  2/7 1/7 0

r3 35r3 /88 0 0 1  1/8 5/88 7/44

r1 r1 4r3 /7 1 0 0  0 2/11 1/11
r2 r2 9r3 /7 0 1 0  1/8 19/88 9/44 .
r3 r3 0 0 1  1/8 5/88 7/44

Hence

0 16 8
1
A1 = 11 19 18 .
88 11 5 14

2.5.5. To find A1 , form the augmented matrix [A|I] and row-reduce it

to the form [I|A1 ], if possible.
28
 
2 3 5  1 0 0

4 1 3  0 1 0

3 2 5  0 0 1

r1 r1 /2 1 3/2 5/2  1/2
0 0
r2 r2 4 1 3  0 1 0
r3 r3 3 2 5  0 0 1

r1 r1 1 3/2 5/2  1/2 0 0
r2 r2 4r1 0 7 7  2 1 0
r3 r3 3r1 0 5/2 5/2  3/2 0 1

r1 r1 1 3/2 5/2  1/2 0 0
r2 r2 /7 0 1 1  2/7 1/7 0
r3 2r3 /5 0 1 1  3/5 0 2/5

r1 r1 1 3/2 5/2  1/2 0 0
r2 r2 0 1 1  2/7 1/7 0
r3 r3 r2 0 0 0  11/35 1/7 2/5

Since the last row corresponds to inconsistent equations, the matrix A

cannot be reduced to the identity matrix I, and therefore A is not invertible.
   
1 0 1 0
2.5.7. For example, A = and B = are invertible,
0 1 0 1
 
2 0
but A + B = is not.
0 0
2.5.9. a. The matrix equation XA = I can be spelled out as

  2 1  
x11 x12 x13 1 0
4 1 =
x21 x22 x23 0 1
2 2

We can write the corresponding system in augmented matrix form, and reduce
it as follows:

     
2 4 2  1 0 2 4 2  1 0

1 1 2  0 1 0 1 3  1/2 1

29
(Notice the appearance of AT in the augmented matrix. This is due to the fact
that the equation XA = I is equivalent to AT X T = I, and it is the latter
equation with the X T on the right in the product that is directly translated to
the augmented matrix form.)
The unknowns x13 and x23 are free. Choosing x13 = s and x23 = t, we
get the systems of equations
2x11 + 4x12 + 2s = 1 2x21 + 4x22 + 2t = 0
and
x12 + 3s = 1/2 x22 + 3t = 1,
1
from which x12 = 2
3s, x11 = 12 + 5s, x22 = 1 3t, and x21 = 2 + 5t.
Thus
 
12 + 5s 1
2
3s s
X=
2 + 5t 1 3t t
is a solution for any s, t, and every left inverse of A must be of this form.
b. AY = I can be written as

2 1   1 0 0
4 y11 y12 y13
1 = 0 1 0 .
y21 y22 y23
2 2 0 0 1
Hence the entries of the first column of Y must satisfy
2y11 y12 = 1
4y11 y21 = 0
2y11 + 2y21 = 0.
From the last two equations we get y11 = y21 = 0 and, substituting these
values into the first equation above, we obtain the contradiction 0 = 1. Thus
there is no solution matrix Y .
2.5.11. For the given square matrix A, we know that AX = I and Y A =
I. If we multiply the first of these equations by Y from the left, we get
Y (AX) = Y I, which is equivalent to (Y A)X = Y . Substituting Y A = I
into the latter equation, we obtain X = Y .
2.5.13. Writing ei for the ith row of I, we must have
1 1
e ce c 0 0
P = P I = P e2 = e2 = 0 1 0 .
3 3
e e 0 0 1

30
But then P also produces the desired multiplication of the first row by c for
any 3 n matrix A:
1 1
c 0 0 a ca
P A = 0 1 0 a2 = a2 .
0 0 1 a3 a3

2.5.15. Since AA1 = I, and since c 1c = 1, a reasonable guess is

(cA)1 = 1c A1 . To prove this, we need only show that (cA)( 1c A1 ) = I. If
we denote the ikth element of A by aik and the kjth element of A1 by bkj ,
then the ikth element of cA is caik , and the kjth element of 1c A1 is 1c bkj , and
hence the ijth element of (cA)( 1c A1 ) is

m m
1
(caik )( bkj ) = aik bkj = (AA1 )ij = Iij ,
c
k=1 k=1

which is what we wanted to show.

2.5.17.

1 0 0 1 0 0 0 1 0

P1 = 0 1 0 , P2 = 0 0 1 , P3 = 1 0 0
0 0 1 0 1 0 0 0 1

0 0 1 0 0 1 0 1 0
P4 = 0 1 0 , P5 = 1 0 0 , P6 = 0 0 1
1 0 0 0 1 0 1 0 0
Pi1 = Pi for i = 1, 2, 3, 4, and P51 = P6 and P61 = P5 .
Furthermore, for instance

1 1
1 0 0 a a
2

P2 A = 0 0 1 a = a3 ,

3
0 1 0 a a2
which shows that P2 permutes the second row of A with the third one. Since
P2 I = P2 , the second and third rows of P2 are those of I permuted. Analo-
gous results hold for each Pi .
If we apply any Pi to an n 3 matrix B from the right, then it permutes
the columns bi of B the same way as Pi has the columns of I permuted. For
31
instance,

1 0 0
BP2 = [b1 b2 b3 ] 0 0 1 = [b1 b3 b2 ].
0 1 0

2.5.19. The matrix (A1 )1 is defined as the solution X of the equa-

tion A1 X = I. Now, multiplying both sides by A and simplifying, we get
A (A1 X) = (AA1 ) X = IX = X on the left, and AI = A on the right.
Thus, X = A. Also, earlier we had X = (A1 )1 . Hence (A1 )1 = A.

3.1.1. The set of all polynomials of degree two and the zero polynomial
is not a vector space. For example, it is not closed under addition: the sum of
x2 + x 1 and x2 is x 1, which is neither a polynomial of degree two nor
the zero polynomial.
3.1.3. The set of all solutions (x, y) of the equation 2x + 3y = 1 is not
a vector space. For example, (1, 1) and (4, 3) are solutions, but their sum
(1, 1) + (4, 3) = (5, 4) is not a solution since 2(5) + 3(4) = 2 = 1.
Thus the set is not closed under addition.
3.1.5. The set of all twice differentiable functions f for which f  (x) +
2f(x) = 1 holds is not a vector space. For example, f (x) = 1/2 is a solution,
but f (x) = 2 (1/2) = 1 is not a solution. Thus the set is not closed under
multiplication by all scalars.
3.1.7. This set is not a vector space, because addition is not commutative:
For example, let (p1 , p2 ) = (1, 2) and (q1 , q2 ) = (1, 3). Then, by the given
addition rule, we have (1, 2) + (1, 3) = (1 + 3, 2 + 1) = (4, 3), but (1, 3) +
(1, 2) = (1 + 2, 3 + 1) = (3, 4).
3.1.9. This set is not a vector space, because Axiom 7 fails: Let a = b =
1, and p2 = 0. Then (a + b)p = (a + b)(p1 , p2 ) = 2(p1 , p2 ) = (2p1 , 2p2 ), but
ap + bp = (p1 , p2 ) + (p1 , p2 ) = (2p1 , 0).
3.1.11. Prove the last three parts of Theorem 3.1.1.
Proof of Part 6:
For any real number c we have c = (1)c, and so
(c)p = ((1)c)p
32
 = (1)(cp) by Axiom 6 of Definition 3.1.1, and
= (cp) by Part 5 of Theorem 3.1.1.
Also, ((1)c)p = (c(1))p by the commutativity of multiplication of
numbers,
= c((1)p) by Axiom 6 of Definition 3.1.1, and
= c(p) by Part 5 of Theorem 3.1.1.
Proof of Part 7:
c(p q) = c[p + (q)] by Definition 3.1.3,
= cp + c(q) by Axiom 8 of Definition 3.1.1,
= cp + ((cq)) by Part 6 above, and
= cp cq by Definition 3.1.3.
Proof of Part 8:
(c d)p = [c + (d)]p by the definition of subtraction of numbers,
= cp + (d)p by Axiom 7 of Definition 3.1.1,
= cp + ((dp)) by Part 6 above, and
= cp dp by Definition 3.1.3.
3.1.13. If p + q = p + r, then adding p to both sides we get p +
(p + q) = p + (p + r). By applying the associative rule for addition, we
may change this to (p + p) + q = (p + p) + r. By Axioms 1 and 4 of
Definition 3.1.1 we can replace the p + p terms by 0, and then Axioms 1
and 3 give q = r.

3.2.1. This U is a subspace of R3 :

a. 0 U , and so U is nonempty.
b. U is closed under addition: Let u, v U . Then u1 = u2 = u3 and
v1 = v2 = v3 . Hence u1 + v1 = u2 + v2 = u3 + v3 , which shows that
u + v U.
c. U is closed under multiplication by scalars: If u U and c R, then
cu1 = cu2 = cu3 , which shows that cu U .
3.2.3. This U is not a subspace of R3 :
Take for instance u = [1, 1, 1]T and v = [1, 1, 1]T . Then u, v U,
but u + v = [2, 0, 2]T / U, because |2| =
 |0|. Thus U is not closed under
addition.
33
 3.2.5. This U is not a subspace of R3 :
Take for instance u = [1, 1, 1]T and v = [1, 2, 0]T . Then u, v U , but
u + v = [2, 3, 1]T / U , because 2 = 3 and neither is the third component of
the sum 0. Thus U is not closed under addition.
3.2.7. This U is not a subspace of Rn :
Take for instance u = e1 = [1, 0, , 0]T and c = 1. Then u U, but
cu = e1 = [1, 0, , 0]T / U . Thus U is not closed under multiplication
by scalars.
3.2.9. If U and V are subspaces of a vector space X, then U V is a
subspace of X :
a. 0 U V , and so U V is nonempty.
b. U V is closed under addition: Let u, v U V . Then we have
both u, v U and u, v V . Since both U and V are vector spaces, they
are closed under addition. that is, u + v U and u + v V . Thus, by the
definition of intersection, u + v U V.
c. U V is closed under multiplication by scalars: If u U V and
c R, then we have both u U and u V . Since both U and V are vector
spaces, they are closed under multiplication by scalars: that is, cu U and
cu V . Thus, by the definition of intersection, cu U V .
3.2.11. No. 0 / U and so U is not a vector space.
3.3.1. Let us prove the if part first: So, assume that one of the vectors
a1 , a2 and a3 is a linear combination of the other two, say

a3 = s1 a1 + s2 a2
for some coefficients s1 and s2 . (The proof would be similar if a1 or a2 were
given as a linear combination of the other two vectors.) Then we can write
equivalently
s1 a1 + s2 a2 + (1)a3 = 0.
This equation shows that there is a nontrivial linear combination of a1 , a2 and
a3 that equals the zero vector. But the existence of such a linear combination
is precisely the definition of the dependence of the given vectors.
To prove the only if part, assume that the given vectors are dependent,
that is, that there exist coefficients s1 , s2 and s3 , not all zero, such that
34
 s1 a1 + s2 a2 + s3 a3 = 0.
Assume further, without any loss of generality, that s3 = 0. Then we can
solve the above equation for a3 as
   
s1 s2
a3 = a1 + a2 .
s3 s3
This equation exhibits a3 as a linear combination of a1 and a2 ; just what we
had to show.
3.3.3. b = (7, 32, 16, 3)T , a1 = (4, 7, 2, 1)T , a2 = (4, 0, 3, 2)T ,
a3 = (1, 6, 3, 1)T .
We have to solve

4 4 1 7
7 s1
0 6
s2 = 32 .

2 3 3 16
s3
1 2 1 3
We solve this system in augmented matrix form by row reduction as follows:
 
4 4 1  7 1 2 1  3
7 0 6  32 6  32
7 0
2 3 3  16 2 3 3  16
1 2 1  3 4 4 1  7
 
1 2 1  3 1 2 1  3
0 14 13  53 5  22
0 7
0 7 5  22 0 14 13  53
0 4 5  19 0 4 5  19
 
1 2 1  3 1 2 1  3
0 7 5  22 5  22
0 7 .
0 0 3  9 0 0 3  9
0 0 15/7  45/7 0 0 0  0
Then, by back substitution, we obtain s3 = 3, 7s2 + 5 3 = 22, and so
s2 = 1, s1 2 3 = 3, and so s1 = 2. Hence b = 2a1 a2 + 3a3 .
3.3.5. We have to solve
35

4 4 0 s1 7
2 3 5 s2 = 16 .
1 2 1 s3 3
We solve this system in augmented matrix form by row reduction as:
 
4 4 0  7 1 2 1  3
2 3 5  16 2 3 5  16
1 2 1  3 4 4 0  7
 
1 2 1  3 1 2 1  3
0 7 7  22 0 7 3  22 .
0 4 4  19 0 0 0  45/7

The last row of the reduced matrix yields the self-contradictory equation
0 = 45/7, and so the vector b cannot be written as a linear combination of
the given ai vectors.
3.3.7. These four vectors are not independent since, by the result of
Exercise 3.3.4, the equation s1 a1 + s2 a2 + s3 a3 + s4 b = 0 has the nontrivial
solution s1 = 2, s2 = 1, s3 = 3, s4 = 1.
3.3.9. The three vectors a1 = (4, 2, 1)T , a2 = (4, 3, 2)T , a3 = (0, 5, 1)T
from Exercise 3. 3. 5 are not independent: We have to solve

4 4 0 s1 0
2 3 5 s2 = 0 .

1 2 1 s3 0
By row-reduction:
 
4 4 0  0 1 2 1  0

2 3 5  0 2 3 5  0

1 2 1  0 4 4 0  0
 
1 2 1  0 1 2 1  0
0 7 7  0 0 7 7  0 .
0 4 4  0 0 0 0  0
Hence s3 is free, and there are nontrivial solutions. Consequently the ai
vectors are dependent.
36
 This result also follows from the equivalence of Parts 4 and 6 of Theorem
2.5.5, which implies that if Ax = b has no solution for some b, as happens
for the b of Exercise 3.3.5, then Ax = 0 has nontrivial solutions.
3.3.11. To test the vectors a1 = (1, 0, 0, 1)T , a2 = (0, 0, 1, 1)T , a3 =
(1, 1, 0, 0)T , a4 = (1, 0, 1, 1)T for independence, we solve As = 0 by row
reduction:
 
1 0 1 1  0 1 0 1 1  0
0 0 1 0  0 0  0
0 0 1
0 1 0 1  0 0 1 0 1  0
1 1 0 1  0 0 1 1 0  0
 
1 0 1 1  0 1 0 1 1  0
0 1 0 1  0 1  0
0 1 0
0 0 1 0  0 0 0 1 0  0
0 1 1 0  0 0 0 1 1  0

1 0 1 1  0
0 1 0 1  0
.
0 0 1 0  0
0 0 0 1  0
Back substitution yields s = 0, and so the columns of A are independent.
3.3.13.
a. The vectors a1 , a2 , . . . , an of any vector space X are linearly indepen-
dent if and only if the equation s1 a1 + s2 a2 + + sn an = 0 has only the
trivial solution s = 0.
b. The vectors a1 , a2 , . . . , an of any vector space X are linearly indepen-
dent if and only if none of them can be written as a linear combination of the
others.
c. The vectors a1 , a2 , . . . , an of any vector space X are linearly indepen-
dent if and only if the solution of the equation s1 a1 + s2 a2 + + sn an = b
is unique for any b X for which a solution exists.
3.3.15. Let a1 , a2 , . . . , an be any vectors in R3 , with n > 3. Then, to
test them for dependence, we solve s1 a1 + s2 a2 + + sn an = 0. We can
do this by reducing the augmented matrix [A|0] to echelon form, where A
is the 3 n matrix that has the given vectors for its columns. Since A is
3 n, we cannot have more than three pivots. On the other hand, since there
37
are n columns. with n > 3, there must exist at least one pivot-free column
and a corresponding free variable. Thus there are nontrivial solutions, and
consequently the vectors a1 , a2 , . . . , an are dependent.
3.3.17. Let a1 , a2 , a3 span R3 . Let A be the 3 3 matrix that has the
given vectors as columns. Then, by the definition of spanning, As = b has
a solution for every vector b in R3 . By the equivalence of Parts 4 and 6 of
Theorem 2.5.5, this implies that the equation As = 0 has only the trivial
solution. Thus a1 , a2 , a3 are independent.
3.3.19. Let A be the n m matrix with columns a1 , a2 , . . . , am . If these
vectors are independent, then, by the definition of linear independence, the
equation As = 0 has only the trivial solution. Thus, if A is row-reduced to
the echelon form U, then the corresponding equation Us = 0 cannot have any
free variables, that is, the number of rows n must be at least the number of
columns m, and there cannot be any free columns. In other words, we must
have m n and r = m. The definition of independence also requires 0 < m.
Conversely, if 0 < m n and r = m, then U s = 0 has no free variables,
and so As = 0 has only the trivial solution. Thus the columns of A are
independent.

5
3.4.1. First we apply Theorems 3.3.1 and 3.3.3:

1 3 1 1 3 1 1 3 1
3
2 4 0 7 1 0 7 1 .
2 1 3 0 7 1 0 0 0

Thus a basis for Col(A) is the set

1 3
3 , 2 ,
2 1
and a basis for Row(A) is

1 0
3 , 7 .
1 1

38
 To find a basis for Null(A) we need to solve Ax= 0. The row reduction of
[A|0] would result in the same echelon matrix as above, just augmented with
a zero column. Thus the variable x3 is free. Set x3 = s; then the second row
of the reduced matrix gives x2 = s/7, and the first row leads to x1 = 10s/7.
Hence

10
s
x= 1
7 7
is the general form of a vector in Null(A), and so the one-element set

10
1

7

is a basis for Null(A).

3.4.3. As in Exercise 3.3.1 above,

1 3 1 1 3 1 1 3 1
3 2
4 0 7
1 0 1 1
.
2 1 3 0 7 1 0 0 8
0 1 1 0 1 1 0 0 0

Thus a basis for Col(A) is the set

1 3 1
4
3 2 , ,
,

2 1 3

0 1 1
and a basis for Row(A) is

1 0 0
3 , 1 , 0 .
1 1 8

39
For Null(A): x3 = 0, x2 = 0, x1 = 0. Hence Null(A) = {0}, which has no
basis or, expressed differently, its basis is the empty set.
3.4.5. A simple example is given by

1 1 1
A= 1 1 1 .
1 0 0
Clearly, the first two rows being the same, they span a one-dimensional space,
which does not contain the third row. Thus the first two rows transposed do
not form a basis for Row(A), since Row(A) must contain the transpose of
every row. On the other hand, the transposed first two rows of

1 1 1
E= 0 1 1
0 0 0
do form a basis for Row(A), since the third row of A can be obtained as the
difference of the first two rows of E.
3.4.7. Since a is the third column of the given matrix A, it is in Col(A).
It can be expressed as a linear combination of the first two columns a1 and a2
by Gaussian elimination as follows:

  
1 3  1 1 3  1 1 3  1
  
3 2 
4 0 7 
1 0 7  1 .
  
2 1  3 0 7  1 0 0  0

Hence s2 = 1/7, s1 3/7 = 1, and so s1 /7. Thus a = (10a1 a2 )/7.

For b, similarly:

  
1 3  10 1 3  10 1 3  10
  
3 2 
1 0 7  31 0 7  31 .
  
2 1  7 0 7  27 0 0  4

The last row shows that this is the augmented matrix of an inconsistent
system, and so b is not in Col(A).
Since c = (a + b) and a is in Col(A), but b is not, c cannot be in
40
Col(A) either. (If it were in Col(A), then b = (a + c) would have to be too,
because Col(A), being a subspace, is closed under vector addition and under
multiplication by scalars.)
For d, row-reduce again:

  
1 3  5 1 3  5 1 3  5
  
3 2  9 0 7  6 0 7  6 .
  
2 1  4 0 7  6 0 0  0

Hence s2 = 6/7, s1 + 18/7 = 5, and s1 = 17/7. Thus we can write d as

d = (17a1 + 6a2 )/7.
3.4.9. First we show that any vector x Null(B) is also in Null(AB).
Now x Null(B) means that x is a solution of Bx = 0. Multiplying both
sides by A, we get ABx = 0. Thus x is also in Null(AB).
Conversely, if x Null(AB), then ABx = 0, and multiplying both sides
by A1 (which exists by assumption), we get A1 ABx = Bx = 0. Thus x is
also in Null(B).
Taken together, the two arguments above show that Null(B) and Null(AB)
contain exactly the same vectors x. This means that Null(B) = Null(AB).
3.4.11. Let A = (a1 , a2 , . . . , am ) be the list of the given spanning vectors,
and B = (b1 , b2 , . . . , bn ) the list of independent vectors that we use for
comparison. By the Exchange Theorem, the number of independent vectors
must not exceed the number of spanning vectors. More formally: the n here
is the k of the Exchange Theorem, and the m here is the n there. Thus the
conclusion k n of the theorem becomes n m.
3.4.13. Let A be the n m matrix with columns a1 , a2 , . . . , am Rn ,
with m > n. These vectors are dependent if and only if the equation As = 0
has nontrivial solutions. Now, if A is row-reduced to the echelon form U,
then the corresponding equation U s = 0 must have free variables, because
the number of pivots cannot exceed the number n of rows, and the number of
columns m is greater than n. So there are free columns and, correspondingly,
nontrivial solutions.
For an alternative proof, assume that B = (b1 , b2 , . . . , bm ), with n <
m, is a list of independent vectors in Rn . For comparison use the standard
vectors as the spanning list, that is, let A = (e1 , e2 , . . . , en ) in the Exchange
41
Theorem.. Then the conclusion k n of the theorem becomes m n. This
inequality contradicts the previous one, and so the vectors of B cannot be
independent.
3.5.1. We reduce A to echelon form:

   
1 1 1 2 0 1 1 1 2 0
.
3 3 0 6 0 0 0 3 0 0

Hence a basis for Row(A) is given by the vectors b1 = (1, 1, 1, 2, 0)T and
b2 = (0, 0, 3, 0, 0)T . Thus dim(Row(A)) = 2. Consequently, by Theorem
3.4.2, dim(Col(A)) = 2, too, and, by Corollary 3.4.2 and Theorem 3.4.4,
dim(Null(A)) = 3, and dim(Left-null(A)) = 0.
To find a basis for Null(A), solve Ax = 0 by setting x2 = s, x4 = t, and
x5 = u. Then x3 = 0 and x1 = s 2t. Hence

1 2 0
1 0 0

x = s 0 + t 0 + u 0

0 1 0
0 0 1

is the general form of a vector in Null(A), and so

1 2 0
1 0 0

c1 =
0 , c2 = 0 , c3 = 0



0 1 0
0 0 1

form a basis for Null(A). Thus, to get the desired decomposition of x =

(2, 0, 1, 4, 1)T , we must solve (2, 0, 1, 4, 1)T = s1 b1 + s2 b2 + t1 c1 +
t2 c2 + t3 c3 . By row-reduction:

1 0 1 2 0  2
1 0 1 0 0  0

1 3 0 0 0  1

2 0 0 1 0  4
0 0 0 0 1  1

42
 
1 0 1 2 0  2

0 0 2 2 0  2

0 3 1 2 0  3

0 0 2 5 0  8

0 0 0 0 1  1


1 0 1 2 0  2

0 3 1 2 0  3

0 0 2 2 0  2

0 0 2 5 0  8

0 0 0 0 1  1


1 0 1 2 0  2
0 3 1 2 0  3

0 0 2 2 0  2
.
0 0 0 3 
0  6
0 0 0 0 1  1
From here, by back substitution, we get t3 = 1, t2 = 2, t1 = 1, 3s2 =
t1 + 2t2 3 = 1 + 4 3 = 0, and s1 = t1 + 2t2 2 = 1. Thus

1 2 0 3
1 0 0 1

x0 =
0 +2
0 +
0 =
0 ,

0 1 0 2
0 0 1 1

and

1
1

xR =
1 .

2
0

43
Indeed,

3 1 2
1 1 0

x0 + xR =
0 +
1 =
1

2 2 4
1 0 1

and

1
1
 
x0 xR = 3 1 0 2 1
1 = 0.

2
0

3.5.3.
a. dim(Row(A)) = dim(Col(A)) = 1, and, by Corollary 3.4.2 and
Theorem 3.4.4, dim(Null(A)) = 4, and dim(Left-null(A)) = 0.
b. This A is an echelon matrix, and so b1 = (3, 3, 0, 4, 4)T forms a basis
for its row space, and we must have xR = sb1 with an unknown value for s.
Let x = (1, 1, 1, 1, 1)T . We use the shortcut mentioned at the end of Example
3.4.2 to solve x = sb1 + x0 for s by left-multiplying it with bT1 . We obtain,
because of the orthogonality of the row space to the nullspace, bT1 x = sbT1 b1 ,
7
which becomes 14 = 50s. Thus s = 7/25 and xR = 25 (3, 3, 0, 4, 4)T . From
1
here we get x0 = x xR = 25 (4, 4, 25, 3, 3)T .
3.5.5. This matrix needs no reduction: it is already in echelon form. Thus
b1 = (0, 2, 0, 0, 4)T and b2 = (0, 0, 0, 2, 2)T form a basis for Row(A). Hence
dim(Row(A)) = 2. Consequently, by Theorem 3.4.2, dim(Col(A)) = 2,
too, and, by Corollary 3.4.2 and Theorem 3.4.4, dim(Null(A)) = 3, and
dim(Left-null(A)) = 0.
To find a basis for Null(A) we solve Ax = 0 by setting x1 = s, x3 = t,
and x5 = u. Then 2x4 + 2u = 0, 2x2 + 4u = 0, and so x4 = u and
x2 = 2u. Hence Null(A) consists of the vectors
44

1 0 0
0 0 2

x = s
0 + t
1 + u
0 ,

0 0 1
0 0 0

and so

1 0 0
0 0 2

c1 =
0 , c2 =
1 , c3 =
0

0 0 1
0 0 0

form a basis for Null(A). To decompose x = (1, 2, 3, 4, 5)T into the sum of an
x0 Null(A) and an xR Row(A), solve x = s1 b1 +s2 b2 +t1 c1 +t2 c2 +t3 c3
by row-reduction:
 
0 0 1 0 0  1 2 0 0 0 2  2
2 0 0 0 2  2 0 2 0 0 1  4

0 0 0 1 0  3 0  1
0 0 1 0
0 2 0 0 1  4 0 0 0 1 0  3
4 2 0 0 0  5 4 2 0 0 0  5

2 0 0 0 2  2

0 2 0 0 1  4


0 0 1 0 0  1



0 0 0 1 0  3

0 0 0 0 6  3
Thus, t3 = 1/2, t2 = 3, t1 = 1, s2 = 7/4, s1 = 1/2,

1 0 0 2
0 0 2 2
1 1
x0 =
0 +3
1
2 0 =
2 6

0 0 1 1
0 0 0 1

45
and

0 0 0
2 0 2
1 7 1
xR = 0 + 0 = 0 .
2
0
4
2
2
7


4 2 11

3.5.7. Null(A) is defined as the solution set of Ax = 0. Writing ai for the

ith row of A, we have, equivalently, ai x = 0 for all i. If x is also in the row
space of A, then by the definition of Row(A), xT is a linear combination of
the rows of A, that is, xT = xAi ai , for some coefficients
xAi . Multiplying both
sides of ai x = 0 by xAi , and summing over i, we get xAi ai x = xT x = 0.
The last equality can be written as |x|2 = 0, which is true only for x = 0.
Thus 0 is the only vector lying in both Null(A) and Row(A).
3.5.9. From Exercise 3.2.9 we know that U V is a subspace of X. Since,
by the result of Exercise 3.5.8, U + V too is a subspace of X, all we need to
prove is that U V is a subset of U + V .
Let x be any element of U V . Then x is an element of U and can
therefore be written as u + v with u = x U and v = 0 V . Thus
x U + V , and so U V is a subset, and consequently a subspace, of U + V .
3.5.11. If both A and B have n columns, then clearly each of the three
nullspaces in the problem are subspaces of Rn . Now, by definition,
     
A 
n A
Null = xR  x=0 .
B B
   
A Ax
Here the condition x = 0 is equivalent to = 0, and therefore
B Bx
to the simultaneous
  occurrence of Ax = 0 and Bx = 0. Thus a vector x Rn
A
is in Null if and only if it is both in Null(A) and in Null(B), that is, in
B
Null(A) Null(B).
3.5.13. For any subspace U of an inner product space X, we have defined

U = {v X|u v = 0 for all u U }. Hence
a. 0 U , because u 0 = 0 for all u U, and so U is nonempty.
b. U is closed under addition: Let v, w U . Then u v = 0 and
46
u w = 0 for all u U . Hence u (v + w) = 0, for all u U , which shows
that v + w U .
c. U is closed under multiplication by scalars: If v U and c R,
then u v = 0 for all u U , and therefore u (cv) = 0 for all u U , which
shows that cv U . Thus U is a subspace of X.
3.5.15. The exercises cited in the hint show that every elementary row
operation on a matrix A is equivalent to multiplication by an elementary
matrix E, and that every such E is invertible. Thus, if A and B are row-
equivalent matrices, then there exist elementary matrices Ei such that A =
E1 E2 Ek B. Since each of the Ei matrices is invertible, so too is their
product R = E1 E2 Ek . Consequently, by the result of Exercise 3.5.14, B
and A = RB have the same rank.
3.5.17. Assuming B has n columns, they are linearly
dependent, if
n
and only if there exist n constants ti , not all zero, such

nthat i=1 ti bi = 0.
Multiplying both sides of this equation by A, we get i=1 ti Abi = 0, which
shows that the vectors Abi , that is, the columns of AB, are dependent with
the same coefficients as the bi vectors.
Conversely, if
the matrix A is not invertible,

n then we cannot conclude
n
that the equation i=1 ti Abi = 0 implies i=1 ti bi = 0. Thus we cannot
expect the converse of the original statement to be true, and should look for a
counterexample.
Indeed, if
   
0 1 1 0
A = and B = ,
0 1 1 1
then
 
1 1
AB =
1 1
Here the columns of AB are linearly dependent, but the columns of B are
linearly independent.
3.5.19. The first three columns of A form an echelon matrix, and so they
are independent and form the basis {a1 , a2 , a3 } for U . The last three columns
of A can be reduced to an echelon matrix with three pivots, and so {a3 , a4 , a5 }
is a basis for V .
To find a basis for U V , we must find all vectors p that can be written both
47
as s1 a1 + s2 a2 + s3 a3 and as t3 a3 + t4 a4 + t5 a5 , that is, find all solutions of the
equation s1 a1 + s2 a2 + s3 a3 = t3 a3 + t4 a4 + t5 a5 . Writing x1 = s1 , x2 = s2 ,
x3 = s3 t3 , x4 = t4 , and x5 = t5 , the equation we wish to solve
becomes Ax = 0. Since A is an echelon matrix, we need no row-reduction,
and back substitution gives x5 = t, x4 = t, x3 = 0, x2 = t/2, x1 = 0.
Equivalently, s1 = 0, s2 = t/2, s3 t3 = 0, t4 = t, and t5 = t. Thus
U V = {p|p = ta2 /2 + s3 a3 } = {p|p = s3 a3 + t(a5 a4 )}. Hence we
observe that {a2 , a3 } is a basis for U V .
Since A is an echelon matrix, {a1 , a2 , a3 , a4 } is a basis for U + V = R4 .
To find a basis for U we need to solve xT (a1 , a2 , a3 ) = 0, that is, find a
basis for the left nullspace of (a1 , a2 , a3 ). This computation results in {e4 } as
a basis for U .
To find a basis for V we need to solve xT (a3 , a4 , a5 ) = 0, that is,
find a basis for the left nullspace of (a3 , a4 , a5 ). This computation results
in {(1, 0, 1, 1)T } as a basis for V .
3.5.21. Let {u1 , u2 , . . . , um } be a basis for U and {v1 , v2 , . . . , vn } a
basis for V . Then dim U = m
and dim V = n. Furthermore,

n any u U and
v V can be written as u = m a
i=1 i iu and v = b
j=1 j jv . Now U V con-

m

n of all sums of vectors from U and from V , that is, of all sums i=1 ai ui +
sists
j=1 bj vj . Thus the combined set C = {u1 , u2 , . . . , um , v1 , v2 , . . . , vn }
spans U
V . It is also
nan independent set, for the following reason: If
u+v = m a u
i=1 i i + j=1 bj vj = 0, then u = v, and so u is also in
V . Since, by the definition of direct sum, the only vector that
mis in both U
and V
m is the zero vector, we have u = v = 0. Thus both i=1 ai ui = 0
and j=1 bj vj = 0. Since the ui vectors form a basis for U , and the vj
vectors a basis for V , the last two equations imply that all coefficients ai and
bj must be zero. Hence C is an independent set, too, and therefore a basis for
U V . This shows that dim(U V ) = m + n and so that dim(U V ) =
dim U + dim V .
3.5.23. Let B = {w1 , w2 , . . . , wk } be a basis for U V . Extend B to a
basis B1 = {w1 , . . . , wk , u1 , . . . , ul } for U and to a basis
B2 = {w1 , . . . , wk , v1 , . . . , vm } for V . (These subspaces, together with the
symbol for the number of basis vectors in each intersection, are illustrated
schematically in a Venn diagram below.) Then
B1 B2 = {w1 , . . . , wk , u1 , . . . , ul , v1 , . . . , vm } is a basis for U + V . Hence
dim(U V ) = k , dim(U + V ) = k + l + m , dim U = k + l, and dim V =

48
k + m. Now, k + l + m = (k + l) + (k + m) k, and this proves the given
formula.

U V
U V

l k m

3.5.25
dim(U + V + W ) = dim(U) + dim(V ) + dim(W )
dim(U V ) dim(U W ) dim(V W ) + dim(U V W ).
(The relevant subspaces, together with the symbol for the number of basis
vectors in each intersection, are illustrated schematically in the Venn diagram
below.)

U p j q V
U V
U V W
k i l
U W V W

Proof : Consider a basis B for U V W . Say, it has i vectors. Extend B

to a basis for U V , so that dim(U V ) = i + j. Next, extend B to a basis
for U W , so that dim(U W ) = i + k, and to a basis for V W , so that
dim(V W ) = i + l. Since U V and U W are subspaces of U , the union
of their bases can be extended to a basis for U, and so dim(U ) = i+ j + k + p,
for some p. Similarly, dim(V ) = i + j + l + q and dim(W ) = i + k + l + r
for some q and r. By the result of Exercise 3.5.20, U + V + W is generated
by the union of U , V and W , and so also by the union of their bases. Thus
49
dim(U + V + W ) = i + j + k + l + p + q + r. On the other hand, the other
dimension formulas above yield dim(U )+dim(V )+dim(W )dim(U V )
dim(U W ) dim(V W ) + dim(U V W )
= (i+j +k +p)+(i+j +l+q)+(i+k+l+r)(i+j)(i+k)(i+l)+i
= i + j + k + l + p +q + r, the same as above.
3.5.27. Let A be a basis for U V, B a basis for U V , C a basis for
U V , and D a basis for U V . Then, clearly, A B C D is a basis
for Rn , C D for the subspace U , B D for V , and B C D for (U
V ) . However, by the result of Exercise 3.5.20, the set ( (B D) (C D) )
= B C D generates U +V , and so we must have (U V ) = U +V .
3.5.29. U V means that every u U is also in V . Thus if v V ,
then vT u = 0 for every u U. On the other hand, the set U was defined
exactly as the set of those vectors that are orthogonal to every u U . Hence
v V implies that v U , and this means that V U holds.
3.5.31. We want to find a basis for Left-null(B), as in Example 3.5.6,
because the vectors of such a basis are orthogonal to the columns of B, and
the transposes of those basis vectors can therefore be taken as the rows of the
desired matrix A. Since Left-null(B) = Null(B T ) we solve B T x = 0 by
reducing B T as follows:

1 0 0 2 1 0 0 0 1 0 0 0
1 2 0
1 1 2 0
1 0 2 0 1 .
1 0 0 0 1 0 0 2 0 0 0 2

Thus x3 is free and x1 = x2 = x4 = 0, and so A = [0, 0, 1, 0] is a matrix

such that Null(A) = Col(B).
3.5.33. For any A as stated, the equation Ax = b is consistent if and only
if b Col(A). For any b Col(A) write the above equation as Ax = Ib
and reduce
 the latter, by elementary row operations, until A is in an echelon
U
form with U having no zero rows. On the right-hand side denote the
O  
L
result of this reduction of the matrix I by Thus we get the equations
M
Ux = Lb and 0 = M b. The last equation shows that the rows of M must
be orthogonal to any vector in the column space of A, and so their transposes
50
  
L
are in the left null- space of A. Furthermore, the matrix has full rank,
M
since it is obtained from I by elementary row operations, which are invertible.
Consequently the rows of M are independent. On the other hand, since the
dimension of the left nullspace of A is m r and M has m r independent
rows, their transposes span the left nullspace of A.
3.5.35. The construction of Exercise 3.5.33 applied to the transpose AT
of A in place of A yields a basis for the nullspace of A.
3.5.37. By the algorithm of Exercise 3.5.36 we do the following reduction

1 0 1  1 0 0 0
2 3 1  0 1 0 0

3 3 0  0 0 1 0

2 0 2  0 0 0 1

5 6 1  0 0 0 0

1 0 1  1 0 0 0
0 3 3  2 1 0 0

0 3 3  3 0 1 0

0 0 0  2 0 0 1
0 6 6  5 0 0 0

1 0 1  1 0 0 0
0 3 3  2 1 0 0

0 0 3  1 1 1 0 .

0 0 0  2 0 0 1
0 0 0  1 2 0 0

Thus x = (1, 2, 0, 0)T is a particular solution, and the general solution

is

1 1 2
2 1 0
x=
0 + s 1 + t 0 ,
0 0 1
where s and t are arbitrary parameters.
3.5.39. Applying the algorithm of Exercise 3.5.38, we have:
51
  
1 2  1 0 1 2  1 0
 
3 4  0 1 0 10  3 1
 
1 0  0 0 0 2  1 0
 
0 1  0 0 0 1  0 0

1 2  1 0

0 10  3 1



 4/10 2/10
.

0 0 
0 0  3/10 1/10
Thus
 
1 1 4 2
A = .
10 3 1
3.5.41. Let A = (a1 , a2 , . . . , am ) be a list of spanning vectors. By
the definition of dimension, any basis of an n-dimensional vector space X
consists of n vectors. Thus, let B = (b1 , b2 , . . . , bn ) be the list of the
vectors of a basis and use this list as the list of independent vectors in the
Exchange Theorem. According to the theorem, the number of independent
vectors cannot exceed the number of spanning vectors, and so we must have
n m.
3.5.43. Assume that B = (b1 , b2 , . . . , bm ), with m > n, is a list of
independent vectors in X. For comparison use the vectors ai of any basis as
the spanning list, that is, let A = (a1 , a2 , . . . , an ) in the Exchange Theorem.
Then the conclusion k n of the theorem becomes m n. This inequality
contradicts the assumption of m > n, and so the vectors of B cannot be
independent.

3.6.1. a. By Corollary 3.6.1,

 
1 2
S=A= .
2 1
b. Compute S 1 :
     
1 2  1 0 1 2  1
 0
  0
2 1 0 1 2 1 1
     
1 2  1 0 1 0  1/5 2/5
 .
0 1 2/5 1/5 0 1  2/5 1/5

52
 Thus
   
1 1/5 2/5 1 1 2
S = =
2/5 1/5 5 2 1
and
    
1
1 1 2 3 1 13
xA = S x = = .
5 2 1 5 5 1
3.6.3. a. The matrix A is the same as in Exercise 3. 6. 1, and so
   
1 2 1 1 1 2
A= and A = .
2 1 5 2 1
Now
 
3 1
B=
2 1
and
    
1 1 1 2 3 1 1 7 3
S=A B= = .
5 2 1 2 1 5 4 1

b. By Theorem 3.6.1,
    
1 7 3 3 3
xA = SxB = = .
5 4 1 2 2
Thus x = 3b1 2b2 = 3a1 2a2 . (It is only a coincidence that the coordinates
of this x are the same in both bases.)
c. From Part (a) above,
 
1 1 3
S = .
4 7
Hence
    
1 1 3 2 14
xB = S xA = = ,
4 7 4 36
and so
x = 2a1 + 4a2 = 14b1 + 36b2 .
53
 3.6.5. From the given data

x2 x3 0x1 + 1x2 1x3
xA = x3 x1 = 1x1 + 0x2 + 1x3
x1 + x2 1x1 + 1x2 + 0x3

0 1 1 x1
= 1 0 1 x2 .
1 1 0 x3

By Corollary 3.6.1, with A in place of B, we have xA = A1 x, and so

0 1 1
A1 = 1 0 1 .
1 1 0
Hence

1 1 1
1
A= 1 1 1 .
2 1 1 1
The new basis vectors a1 , a2 , a3 are the columns of this matrix (including the
factor 1/2).
3.6.7. a. From the statement of the problem,

0 1 0
A = 0 0 1 .
1 0 0
By Corollary 3.6.1, with A in place of B, we have S = A.
b. From A we compute

0 0 1
A1 = 1 0 0
0 1 0
and

0 0 1 3 5
xA = A1 x = 1 0 0 4 = 3 .
0 1 0 5 4

54
 c.

0 0 1 1 2 3 0 1 0
MA = A1 M A = 1 0 0 1 2 0 0 0 1
0 1 0 1 0 0 1 0 0

0 1 0

= 3 1 2 .
0 1 2
3.6.9. a. We need to solve AS = B. As in Example 3.6.3, we reduce the
augmented matrix [A|B] :
 
1 0  2 3 1 0  2 3
2 4  0 2 1 2  0 1
3 1 7  11 3 1  7 11
 
1 0  2 3 1 0  2 3
 
0 2  2
4 0 1  1 2 .
 
0 1  1 2 0 0  1 2

Thus
 
2 3
S= .
1 2

b. The transition matrix in the reverse direction is S 1 , which happens to

equal S.
3.6.11. a. As in Example 3.6.4, in the space

P3 = {p = P : P (x) = p0 + p1 x + p2 x2 + p3 x3 ; p0 , p1 , p2 , p3 R}

we choose the basis A to consist of the monomials, that is, ai = xi for i =

0, . . . , 3, and the basis B to consist of the first four Legendre polynomials, that
is, bi = Li for i = 0, . . . , 3. Then, according to Theorem 3.6.2, the columns
of the matrix S are given by the coordinates of the bi vectors relative to A.
These can be read off the definitions of the Legendre polynomials, to give (in
55
ascending order of degrees)

1 0 1/2 0
0 1 0 3/2
S=
0
.
0 3/2 0
0 0 0 5/2
b. The coordinate vector of any p relative to A is pA = (p0 , p1 , p2 , p3 )T and its
coordinate vector relative to B is given by pB = S 1 pA . Thus multiplication
of pA by

1 0 1/3 0
0 1 0 3/5
S 1 = 0 0 2/3 0

0 0 0 2/5
will give the coordinates of any p relative to B. For the polynomial P (x) =
1 2x + 3x2 4x3 we have

1 0 1/3 0 1 2
0 1 0 3/5
2 = 22/5 .
0 0 2/3 0 3 2
0 0 0 2/5 4 8/5
Thus P (x) = 2L0 (x) 22L1 (x)/5 + 2L2 (x) 8L3 (x)/5.
3.6.13. For any x Rn , let y = Mx be the corresponding vector of
Rm . We can rewrite this equation, by using x = AxA and y = ByB , as
AxA = M ByB . Hence xA = A1 M ByB and so MA,B = A1 M B.
3.6.15. The second one of the given matrices is the 2 2 identity matrix
I. Now, for any invertible matrix S, SIS 1 = I, and so there is no matrix
other than I itself that is similar to the identity matrix.
3.6.17. a. B is similar to A if there exists an invertible matrix S such
that B = S 1 AS, or, equivalently, SB = AS. Taking the transpose of both
sides, we get B T S T = S T AT and from this B T = S T AT (S T )1 . Thus B T is
similar to AT with (S T )1 in place of S if B is similar to A.
b. If B = S 1 AS, then B k = S 1 ASS 1 AS S 1 AS = S 1 Ak S.
c. For any invertible matrices A and S, we have (S 1 AS)1 = SA1 S 1 ,
and so, if B = S 1 AS, then B 1 exists and equals SA1 S 1 , and B 1 is
similar to A1 with S 1 in place of S.
56
 3.6.19. An nn matrix B is similar to an nn matrix A if there exists an
invertible n n matrix S such that B = S 1 AS, or, equivalently, SB = AS.
Furthermore, by the result of Exercise 3.4.9, Null(B) = Null(SB).
To see that nullity(A) = nullity(AS), consider the definitions of these
nullspaces: Null(AS) is the solution set of the equation ASx = 0, and
Null(A) that of Ay = 0. Since S is invertible, the equation y = Sx estab-
lishes a one-to-one linear correspondence between the elements of Null(AS)
and Null(A). In particular, any basis of Null(AS) is mapped by S onto a basis
of Null(A). Thus the dimensions of the two nullspaces must be equal.
Coupling the statements in the two paragraphs above, we find that nullity(A)
= nullity(B). Thus, from this equation and Corollary 3.5.1 we get the desired
result: rank(A) = rank(B).

4.1.1. a. We just prove Corollary 4.1.1, because Lemma 4.1.1 is a special

case of it with n = 2.
Assume first that Equation 4.4 holds for some n 2 and for all vectors
and all scalars in it. Then, choosing c1 = c2 = 1 and ci = 0 for all i > 2, we
get Equation 4.1. On the other hand, choosing c1 = c and ci = 0 for all i > 1,
we get Equation 4.2. Thus, if T preserves all linear combinations with n 2
terms, then it is linear.
Assume conversely that T is linear, that is, that Equations 4.1 and 4.2 hold
for all choices of x1 , x2 , and c. Then, for any n 2,

 n
  n
  n 
  
T ci xi =T c1 x1 + ci xi = T (c1 x1 ) + T ci xi
i=1 i=2 i=2

 n


= c1 T (x1 ) + T ci xi .
i=2

We can similarly reduce the remaining sum by repeated applications of Equa-

tions 4.1 and 4.2 until we get
 n  n
 
T ci xi = c1 T (x1 ) + c2 T (x2 ) + + cn T (xn ) = ci T (xi ).
i=1 i=1

57
b. For any x we have T (0) = T (x x) = T (x + (1)x) = T (x) +
T ((1)x) = T (x) + (1)T (x) = T (x) T (x) = 0.
4.1.3. a. It is certainly true that if a mapping T from Rn to Rm is linear,
then it preserves straight lines: If T is any such mapping, then, applying T to
x = x0 + ta with any x, x0 , a Rn and scalar t, we must have y = T (x) =
T (x0 + ta) = T (x0 ) + tT (a). Thus the line given by x = x0 + ta is mapped
into the line given by y = y0 + tb with y 0 = T (x0 ) and b = T (a).
b. The converse statement is not true in general, that is, not every trans-
formation that preserves straight lines is linear. For example a shift within
Rn , defined by y = S(x) = x + c with fixed nonzero c, preserves straight
lines but is nonlinear, since S(0) = 0 + c = c and not 0 as required for linear
transformations (see Exercise 4.1.1).
Let us try to find out what additional conditions will make such transfor-
mations linear. So, let us assume that T maps every x = x0 + ta to some
y = y0 + tb for every scalar t. Setting t = 0 yields y 0 = T (x0 ), and so we
must have T (x0 + ta) = T (x0 ) + tb. For T to be linear we must also have
b = T (a). These two conditions are also sufficient, that is, if T maps every
x = x0 + ta to y = y0 + tb with y 0 = T (x0 ) and b =T (a), then T is linear.
Indeed, in that case, T (x0 + ta) = T (x0 ) + tT (a) with t = 1 shows the addi-
tivity of T . Setting x0 = a = 0, from here we obtain T (0) = 2T (0), and so
T (0) = 0. Hence T (ta) = T (0 + ta) = T (0) + tT (a) = 0 + tT (a) = tT (a),
which shows the homogeneity of T .
4.1.5. We can write

x1 x2 x1 x2
T (x) = 2x1 + 3x2 = 2x1 + 3x2
3x1 + 2x2 3x1 2x2

1 1 1 1  
x1
= x1 2 + x2 3 = 2 3 ,
x2
3 2 3 2
and so

1 1
[T ] = 2 3 .
3 2

4.1.7. In the formula T (x) = (aT x)b the product on the right is under-
58
stood as the scalar aT x multiplying the vector b, that is, multiplying each
component of b. Thus the ith component of T (x) is

 n
 n
 
[T (x)]i = af xj bi = (bi aj )xj ,
j=1 j=1

and so [T ]ij = bi aj .
Alternatively, we may write T (x) as the matrix product b(aT x), in which
T
a x is considered to be a 11 matrix. Thus, by the associativity of matrix
multiplication, T (x) = (baT )x. Hence

b1 b1 a1 b1 a2 b1 an
b 2 b a b2 a2 b2 an
[T ] = baT = . (a1 , a2 , . . . , an ) = 2. 1 .
.. ..
bm bm a1 bm a2 bm an
This matrix is called the tensor product of b and a and also the outer product
of b and aT .
4.1.9. a1 = (1, 1, 1)T , a2 = (1, 1, 1)T and a3 = (1, 1, 0)T are the
columns of the basis-matrix

1 1 1
A= 1 1 1 .
1 1 0
Similarly, the vectors b1 = (1, 1)T , b2 = (1, 1)T and b3 = (1, 0)T may be
combined into the (nonbasis) matrix
 
1 1 1
B=
1 1 0
By the definition of the matrix [T ],
[T ]A = B
and so
[T ] = BA1 .
59
Now A1 can be obtained in the usual way:

1 1 1  1 0 0
1 1 1  0 1 0
1 1 0  0 0 1

1 1 1  1 0 0
0 2 0  1 1 0
0 2 1  1 0 1

1 1 1  1 0 0
0 2 0  1 1 0
0 0 1  0 1 1

1 1 1  1 0 0
0 1 0  1/2 1/2 0
0 0 1  0 1 1

1 0 0  1/2 1/2 1
0 1 0  1/2 1/2 0 .
0 0 1  0 1 1

Thus,

1/2 1/2 1
A1
= 1/2 1/2 0 .
0 1 1

and

  1/2 1/2 1  
1 1 1 1 1/2 1 0 0
[T ] = BA = 1/2 0 = .
1 1 0 0 0 1
0 1 1

Hence

  x1  
1 0 0 x2 = x1
T (x) = .
0 0 1 x3
x3

4.1.11. A rotation by 45 takes the y = x line into the x-axis. Then

60
stretch by the matrix S of Example 4.1.11 and rotate back by 45 :
 2      
2 1 1 2 0 1 1 1 3 1
[T ] = = .
2 1 1 0 1 1 1 2 1 3

It is easy to check that this matrix doubles the vector (1, 1)T and leaves
the vector (1, 1)T unchanged, as it should.
4.1.13. The matrix [T ] given by

 
1 b2 a2 2ab
[T ] = 2
a + b2 2ab a2 b2
represents a reflection across the line ax + by = 0 (see Exercise 4.1.12).
If denotes the inclination of that line, with 0 , then cos 2 =
(b2 a2 )/(a2 + b2 ) and sin 2 = (2ab)/(a2 + b2 ). Thus we can rewrite [T ]
as
 
cos 2 sin 2
[T ] = .
sin 2 cos 2
Similarly, the matrix [S] given by
 
cos 2 sin 2
[S] =
sin 2 cos 2
represents a reflection across a line cx + dy = 0 whose inclination is . The
matrix that represents the result of a reflection across the line ax + by = 0
followed by a reflection across the line cx + dy = 0 is given by
  
cos 2 sin 2 cos 2 sin 2
[S][T ] =
sin 2 cos 2 sin 2 cos 2
 
cos 2 ( ) sin 2 ( )
= .
sin 2 ( ) cos 2 ( )
The resultant matrix corresponds to a rotation about the origin through the
angle 2( ) , which is twice the angle between the lines, measured from
ax + by = 0 to cx + dy = 0.
4.1.15. A linear transformation T from R2 to R2 will map a line p = p1 +
tv1 onto the line T (p) = T (p1 + tv1 ) = T (p1 ) + tT (v1 ). Similarly T
61
maps the line p = p2 + tv2 onto the line T (p) =T (p2 ) + tT (v2 ) Thus T
has the property that it maps perpendicular lines onto perpendicular lines if
and only if it maps orthogonal vectors onto orthogonal vectors. The zero
transformation satisfies these conditions trivially. Suppose that T = 0, and
that the matrix of T relative to the standard basis for R2 is given by
 
a b
[T ] = .
c d

Then T (e1 ) = (a, c)T , and T (e2 ) = (b, d)T . Since e1 and e2 are orthogo-
nal, their images are orthogonal; that is, (a, c)T (b, d)T = 0, or, (1) ab +
cd = 0. Also, the vectors e1 e2 and e1 +e2 are orthogonal, and thus so
are their images; that is, (a b, c d)T (a + b, c + d)T = 0, or, (2)
a2 + c2 = b2 + d2 . The assumption that T = 0, together with Equations
(1) and (2) above, imply that (a, c)T = 0 and (b, d)T = 0. In addition, we
may write (a, c)T = k(cos , sin )T and (b, d)T = k(cos , sin )T , where
k = a2 + c2 = b2 + d2 > 0, and = = /2. There are two cases to
consider:
(1) = + /2.
In this case, sin = cos and cos = sin , and so
 
cos sin
[T ] = k .
sin cos
Hence T is a rotation through an angle , followed by a dilation or contraction
by a factor k.
(2) = /2.
In this case, sin = cos and cos = sin , and so
 
cos sin
[T ] = k .
sin cos

Hence T is a reflection across the line through the origin with inclination /2,
followed by a dilation or contraction by a factor k.
Since a pure dilation or contraction is the special case of (1) when = 0,
and the zero transformation is obtained when k = 0, we can summarize the
possibilities as follows. A linear transformation T preserves orthogonality
if and only if it is a rotation, a reflection, a dilation or a contraction, or a
62
composition of such transformations.
4.1.17. For the transformation T of Exercise 4.1.6, we have

x1 x2 1 1 0 x1

T (x) = x2 x3 = 0 1 1 x2 ,
x3 x1 1 0 1 x3
and thus

1 1 0
[T ] = 0 1 1 .
1 0 1
We may apply Corollary 4.1.2 to determine the matrix TA,B . To do so, we
need to compute B 1 , which we find to be

1 1 1
1
B 1 = 1 1 1 .
2 1 1 1

Then matrix multiplication gives

TA,B = B 1 [T ]A

1 1 1 1 1 0 1 0 1
1
= 1 1 1 0 1 1 1 1 0
2 1 1 1 1 0 1 0 1 1

1 1 0
= 0 1 1 .
1 0 1

4.1.19. To determine the matrix TA,B that represents the integration map
T relative to the ordered bases A = (1, x, . . . , xn ) and B = (1, x, . . . , xn+1 ) ,
we can proceed as in Example 4.1.13 in the text. Using the notation aj = xj1
and bi = xi1 , we have

1
T (aj ) = bj+1 for j = 1, 2, . . . , n.
j
63
Thus
T (a1 ) = 0b1 +1b2 + + 0bn+1 ,

T (a2 ) = 0b1 + 0b2 + (1/2)b3 + + 0bn+1 ,

T (a3 ) = 0b1 + 0b2 + 0b3 + (1/3)b4 + + 0bn+1 ,

etc.
Therefore

0 0 0 0
1 0 0 0

0 1/2 0 0
TA,B =
0 0 1/3 0
.

.. .. .. ..
. . . .
0 0 0 1/(n + 1)

4.2.1. a. W must be closed under addition: Let x1 , x2 W . Then

T (x1 ) + T (x2 ) is well-defined, but if x1 + x2 were not in the domain W of
T , then T (x1 + x2 ) would not be defined, and T (x1 ) + T (x2 ) = T (x1 + x2 )
could not hold, in contradiction to our assumption.
b. W must be closed under multiplication by scalars: Let x1 W , and
let c be any scalar. Then cT (x1 ) is well-defined, but if cx1 were not in the
domain W of T , then T (cx1 ) would not be defined, and cT (x1 ) = T (cx1 )
could not hold, in contradiction to our assumption.
4.2.3. a. T being one-to-one implies Ker(T ) = {0}: By definition, T is
one-to-one if T (x) = T (y) implies x = y. Choose y = 0. By the result of
Exercise 4.1.1b, T (0) = 0, and so T (x) = 0 implies T (x) = T (0), which, by
the one-to-oneness assumption, implies x = 0. Thus Ker(T ) = {x : T (x) =
0} = {0}.
b. Ker(T ) = {0} implies T is one-to-one: If Ker(T ) = {0}, then T (x
y) = 0 implies xy = 0. But, by the linearity of T , the equation T (xy) =
0 is equivalent to T (x) T (y) = 0, that is, to T (x) = T (y). Also, x y = 0
is equivalent to x = y, and so, if Ker(T ) = {0}, then T (x) = T (y) implies
x = y. This is precisely the definition of T being one-to-one.
64
 4.2.5. If N = M T , then Row(N ) = Col(M ) and Col(N ) = Row(M).
Thus we can use the same bases as in Example 4.2.5, just switched, and so

1 1

A= 0 1
1 2
is a basis matrix for Row(N ), and

1 0
0 1
B=
1

0
1 1
is a basis matrix for Col (N ).
If N : Row(N ) Col(N ) is the isomorphism given by y = Nx, then
this equation can be written, in terms of the coordinate vectors, as

ByB = NAxA ,
that is, as

1 0   1 0 1
 
0 1 1
1 yB1 1 1 2 xA1
1 = 0 1
0 yB2 1 0 1 xA2
1 2
1 1 2 1 3

2 3  
3 6
= xA1 .
2 3 xA2
5 9
This equation can be reduced to

1 0   2 3  
0 1 yB1 3 6 xA1

0 0 yB2 = 0 0 xA2
.
0 0 0 0

65
Hence
 
2 3
NA,B = .
3 6

4.2.7. a. Ker(T ) is not empty, since 0 is always a member of it.

b. It is closed under vector addition, since, if T (x) = 0 and T (y) = 0,
then, by the linearity of T, T (x + y) = T (x) + T (y) = 0.
c. It is closed under multiplication by scalars: If T (x) = 0 and c is any
scalar, then, by the linearity of T , we have T (cx) = cT (x) = 0.


4.2.9. Let us first test the bi vectors for independence: Assume ni=1 ci bi
= 0. Then, by the linearity of T , we may rewrite this as

n
 n

ci T (ai ) = T ( ci ai ) = 0.
i=1 i=1

Also, as for any linear
ntransformation, T (0) = 0. Thus, because T is one-to-

one, we must have i=1 ci ai = 0. Since the ai vectors form a basis for U,
they are independent, and so, ci = 0 for each i. Hence the bi vectors are also
independent, and consequently, by the result of Exercise 3.4.40, they form a
basis for V .
All that remains to show is that T is a mapping onto V . But this is
now easy: Let b V . Since the bi vectors form a
basis for V , we can
n
write
b as a linear combination
of these, say, b = i=1 ci bi . But then
n n
b = c
i=1 i T (a i ) = T ( c a
i=1 i i ), which shows that any b V is the
image of some a U under T .
4.2.11. Consider the ordered basis A = B = (1, x, . . . , xn ) of Pn , that
is, take the vectors ai = xi1 , i = 1, 2, . . . , n + 1, for a basis. Then, as in
Example 4.1.13, Dai = (i 1)xi2 for i = 2, . . . , n + 1,.and Da1 = 0.
Hence, X Dai = (i 1)xi1 = (i 1)ai for i = 1, 2, . . . , n + 1, and,
similarly, D Xai = ixi1 = iai for all i.
Thus, T ai = (X D D X)ai = ai . Consequently, we find that
TA,B = I, Range(T ) = Pn , Ker(T ) = {0}, and T is both one-to-one and
onto.
66
 4.3.1. As in Example 4.3.1, we have

R(30 ) = T (1, 2)R(30 )T (1, 2)

1 0 1 3
1 0 1 0 1
1
= 0 1 2 1 3 0 0 1 2
2 0 0 1 0 0 1
0 0 2
1 1 1

2
3 2 2 3
= 1 1
3 3 52
2 2
0 0 1

1 0 1 3 1
3 2
1
= 0 1 2 1 3 1 + 2 3
2 0 0 1 0 0 2

3 1 3
1 .
= 1 3 2 35
2
0 0 2

4.3.3.

3 0 1
L1 = 0 4 2 .
0 0 1
This matrix represents the inverse of the mapping of Exercise 4.3.2, that is, it
maps the unit square onto the given rectangle without any rotation. Indeed,
L1 (0, 0, 1)T = (1, 2, 1)T , L1 (1, 0, 1)T = (4, 2, 1)T , L1 (0, 1, 1)T =
(1, 2, 1)T , L1 (1, 1, 1)T = (4, 2, 1)T .
4.3.5. First we use the matrix

pp2 pp1 0
1
R= p1 p3 p2 p3 p212
pp12 p p p2 p12 p3 p12
1 12

resulting from Exercise

 4.3.4, to rotate the vector p into the z-axis. Here
p = |p| and p12 = p1 + p22 . Next, we rotate by the angle about the z-axis,
2

using the matrix of Equation (4.107) and, finally, we rotate back by R1 . Thus
67

1 pp2 p1 p3 p1 p12 cos sin 0 pp2 pp1 0
2
R = pp1 p2 p3 p12 sin cos 0 p1 p3 p2 p3 p212
p2 p212 0 p2 p12 p3 p12 0 0 1 p1 p12 p2 p12 p3 p12

the solution of Exercise 4.3.5, with p1 = p2 = 1, p3 = 0

4.3.7. From
and p = p12 = 2, we obtain

2 cos + 2 2 cos + 2 22 sin 0
2 cos + 2 2cos + 2 2 2 sin 0
R =
2 2 sin
.
2 2 sin 4 cos 0
0 0 0 4

4.3.9. Let u = 1 (2, 1, 0)T and v = (0, 0, 1)T . Thus

5

 
2/ 5 1/ 5 0
PV (u, v) =
0 0 1

and
0   6  
3/ 5 9/ 5
PV 3 = , PV 3 = ,
4 4
4 4

0   6   0  
3/ 5 9/ 5 0
PV 3 = , PV 3 = , PV 0 = ,
0 0 4
0 0 4

0   6  
0 12/ 5
PV 0 = , PV 0 = ,
0 4
0 4

0   6  
3 5/10 21 5/10
PV 3/2 = , PV 3/2 = .
5 5
5 5
Hence we get the following picture:

68
 5.1.1. To find a matrix A with independent columns with a given column
space, is just another way of saying that we want to find a basis for the given
space.
In the present case the x-axis is one-dimensional and is spanned by any
non-zero multiple of the vector i. Thus A = (1, 0, 0)T is a matrix with
independent columns whose column space is the x-axis.
5.1.3. Since the given column space is a plane, it is two-dimensional and
we just have to find two independent vectors whose components satisfy the
condition x = y. Clearly the columns of the matrix

1 0
A= 1 0
0 1
will do.
Alternatively, using the formalism of Chapter 2, we may write the equa-
tion x = y in the form 1x 1y + 0z = 0, whose augmented matrix is the
echelon matrix B = [1, 1, 0| 0]. Columns 2 and 3 are free, and so we set y
and z equal to parameters s and t. Then we get the general solution as

x 1 0
y = s 1 + t 0
z 0 1
from which we see that the columns of the matrix A above form a basis for
Null(B), that is for the x = y plane.
69
 5.1.5. In one direction the proof is easy: If x is a solution of Ax = p, then
left-multiplying both sides by AT shows that x is also a solution of AT Ax =
AT p.
In the other direction, the assumption that p is in Col(A) must be used.
This condition means that there exists a vector pA such that p = ApA . Sub-
stituting this expression for p into AT Ax = AT p, we get AT Ax = AT ApA .
By Lemma 5.1.3 the matrix AT A is invertible for any A with independent
columns, and so the last equation yields x = pA . Thus if x satisfies AT Ax =
AT p, then it also satisfies p = ApA = Ax.
5.1.7. Since such a matrix has to map any vector (x, y, z)T to (x, y, 0)T ,
it must be the matrix

1 0 0

P = 0 1 0 .
0 0 0
5.1.9. The solution of Exercise 5.1.3 gives a matrix A with independent
columns, whose column space is the plane x = y. Thus, by Theorem 5.1.1, a
matrix of the projection onto this plane is given by P = A(AT A)1 AT Now

  1 0  
T 1 1 0 2 0
A A= 1 0 =
0 0 1 0 1
0 1

and
 
 T
1 1/2 0
A A = .
0 1
Hence

1 0    1/2 1/2 0
1/2 0 1 1 0
P = 1 0
= 1/2 1/2 0 .
0 1 0 0 1
0 1 0 0 1

Using this P , we obtain the projections of the given vectors as

1/2 1/2 0 1 0
q1 = 1/2 1/2 0 1 = 0 ,
0 0 1 2 2

70

1/2 1/2 0 1 3/2

q2 = 1/2 1/2 0 2 = 3/2 ,
0 0 1 3 3
and

1/2 1/2 0 2 1/2
q3 = 1/2 1/2 0 1 = 1/2 .
0 0 1 2 2
5.1.11.
 2
A(AT A)1 AT = A(AT A)1 AT A(AT A)1 AT = A(AT A)1 AT ,

because the AT A in the middle can be cancelled with one of the (AT A)1
factors. Also,
(A(AT A)1 AT )T = AT T ((AT A)1 )T AT = A((AT A)T )1 AT = A(AT A)1 AT .

5.1.13. This proof makes use of P 1 , which does not exist for any
projection matrix other than I. Indeed, if P 1 exists, then multiplying both
sides of the idempotency relation P 2 = P by it results in P = I.
5.1.15. We have Null(A) = Left-null(AT ) = Col(AT ) , and so, by the
result of Exercise 5.1.12, with AT in place of A, the matrix of the projection
onto Null(A) is P = I AT (AAT )1 A.
5.1.17. The normal system can be written as
    
x2i xi a xi yi
= .
xi m b yi

The second row corresponds to the equation

 
a xi + mb = yi .



Division by m gives y = ax + b with x = m1 xi and y = 1
m
yi . The latter
are the coordinates of the centroid of the given points.
5.1.19. The third equation of the normal system is

  
a xi + b yi + mc = zi .

71



Division by m gives z = ax + by + c with x = m1 xi , y = m1 yi , and
1


z = m zi . The latter are the coordinates of the centroid of the given points.
5.2.1. Normalizing the given vectors to a1 = 13 (2, 1, 2)T and a2 =
1 (1, 0, 1)T , we can apply Equation 5.49, since these vectors are obvi-
2
ously orthogonal. Now a1 x = 13 (2, 1, 2)T (2, 3, 4) = 15 3
and a2 x =
1 (1, 0, 1)T (2, 3, 4) = 2 . Thus the required projection is
2 2

2 1 7
15 2 1
xC = 1 + 0 = 5 .
9 2 2 1 3 13

5.2.3. a. The subspace U is spanned by the orthonormal vectors q1 , q2 ,

and q3 given by the Equations 5.56. Thus we may use Equation 5.49 with
these vectors in place of the ai there. Now

1 2
q1 x = (2, 0, 1, 1)T (1, 2, 3, 1) = ,
6 6

1
q2 x = (0, 2, 1, 1)T (1, 2, 3, 1) = 6,
6

and
1 15
q3 x = (3, 5, 2, 8)T (1, 2, 3, 1) = .
102 102

Thus, substituting into Equation 5.49, we get

2 0 3 23
2
0 2
+

15 5 = 1 279
.
xU =
6 1 1 102 2 102 106
1 1 8 52

b. The projection xU of x into U can be obtained simply by subtracting

72
xU from x:

1 23 5
2 279
xU = x xU = 1 = 25 3 .
3 102 106 102 8
1 52 2

c. Since U is one-dimensional, the vector xU forms a basis for it.

d. To obtain an appropriate q4 , we just have to normalize xU . Actually,
we can ignore the factor 25/102, and normalize without it. 52 + 32 + 82 + 22 =
1
102, and so q4 = 102 (5, 3, 8, 2)T .
5. 2. 5. Apply first |Qx| = |x| to x = ei . Now

1
0
Qe1 = (q1 q2 qn )
... = 1q1 + 0q2 + + 0qn = q1
0
and similarly Qei = qi for all i. Thus |Qei | = |ei | = 1 implies |qi | = 1.
Next, apply |Qx| = |x| to x = ei + ej with i = j:

|qi + qj |2 = |Qei + Qej |2 = |Q(ei + ej )|2 = |ei + ej |2

= |ei |2 + 2eTi ej + |ej |2 = 1 + 0 + 1 = 2.
On the other hand,
|qi + qj |2 = |qi |2 + 2qTi qj + |qj |2 = 2 + 2qTi qj .
Comparing this result with the previous equation, we see that qTi qj = 0 must
hold. Thus the columns of Q are orthonormal, which means that Q is an
orthogonal matrix.
5.2.7. The vectors q1 = 13 (1, 2, 2)T and q2 = 13 (2, 1, 2)T are or-
thonormal. Thus, we need to find a vector x of length 1 such that qT1 x = 0
and qT2 x = 0 hold. With x = (x, y, z)T , the above equations reduce to
x + 2y + 2z = 0 and 2x y + 2z = 0. Subtracting the first equation from
the second one, we get 3x 3y = 0 or x = y. Letting x = y = 2, we obtain
73
z = 1. Thus x = (2, 2, 1)T is a vector orthogonal to q1 and q2 . Its length
is 3, and so q3 = 13 (2, 2, 1)T is a unit vector in the same direction. Thus
q1 , q2 , and q3 form an orthonormal basis for R3 .

n T T
5.2.9. By Exercise 2.4.13a,

i=1 qi qi = QQ and, by Theorem 5.2.2,
T 1 n T 1
Q = Q . Consequently, i=1 qi qi = QQ = I.

6.1.1.
   
 2 3 2   1 4 0 
   
 1 4 0  = (1)  2 3 2  =
  
 0 1 5   0 1 5 
   
 1 4 0   1 4 0 
 
(1)  0 11 2  = (1)2  0 1 5 
 0 1 5   0 11 2 
 
 1 4 0 
2

= (1)  0 1 5  = 53.
 0 0 53 

6.1.3.
   
 1 1 2 3   1 1 2 3 
   
 2 0 5 0   
  =  0 2 1 6 
 0 0 0 1   0 0 0 1 
   
 3 3 1 4   0 0 7 13 
 
 1 1 2 3 
 
 0 2 1 6 
= (1)   = 14.

 0 0 7 13 
 0 0 0 1 

6.1.5. Axiom 2 does not apply, because it relates two distinct columns
of a matrix to each other, and when n = 1, there is only one column, since
A = [a11 ] .
Axiom 1 becomes det (sa1 + ta1 ) = s det (a1 ) + t det (a1 ). Choosing
s = a11 , t = 0, and a1 = e1 = [1], we can thus write det(A) = det(a11 ) =
det(a11 e1 ) = a11 det(e1 ), which, by Axiom 3, equals a11 .
6.1.7. This result could be proved much the same way as Theorem 6.1.7
was in the text. Alternatively, it follows from Theorems 6.1.6 and 6.1.7, since
74
AT is upper triangular:

T
a11 0 0 a11 a21 a31
a21 a22 0 = 0 a22 a32 .
a31 a32 a33 0 0 a33

6.1.9. If A invertible, then AA1 = I. Hence, by Theorem 6.1.9 and Ax-

iom 3 of Definition 6.1.1, we have |A||A1 | = |AA1 | = |I| = 1. Dividing
by |A|, we get |A1 | = 1/|A|.
6.1.11. If AB is invertible, then det(AB) = det(A) det(B) = 0, and so
neither det(A) nor det(B) is zero, which implies that A and B are invertible.
Conversely, if A and B are invertible, then neither det(A) nor det(B) is zero,
and so det(AB) = det(A) det(B) = 0, which implies that AB is invertible.
6.1.13. By the definition of an orthogonal matrix, QT Q = I. Then
Theorems 6.1.9 and 6.1.6 give |QT Q| = |QT ||Q| = |Q|2 . On the other hand,
by Axiom 3 of Definition 6.1.1, we have |QT Q| = |I| = 1, and so |Q|2 = 1,
from which we get the desired result: det(Q) = 1.
6.1.15. The formula for the Vandermonde determinant of order n > 3
can be expressed as

 
 1 a1 a21 a1n1 
 
 1 a2 a22 a2n1  
 
 1 a3 a23 a3n1 = (aj ai ).
 .. .. .. .. .. 
 . . . . .  1i<jn
 
 1 an a2n ann1 

The notation on the right-hand side represents the product of all factors of the
form aj ai , where i and j are subscripts from 1 to n, with i < j.
To prove the formula, we will show how to express it for the case of an
order n determinant in terms of the corresponding one of order n 1. Starting
with the Vandermonde determinant of order n, subtracting the first row from
75
each of the succeeding rows results in the equivalent determinant
 
 1 a1 a21 a1n1 
 2 2 n1 n1 
 0 a2 a1 a2 a1 a2 a1 
 
 0 a3 a1 a23 a21 a3n1 a1n1  ,
 . .. .. .. .. 
 .. . . . . 
 
 0 a a a2n a21 ann1 a1n1 
n 1

which is equal to
 
 a2 a1 a22 a21 a2n1 a1n1 
 
 a3 a1 a23 a21 a3n1 a1n1 
 .. .. .. .. .
 . . . . 
 
 a a a2n a21 ann1 a1n1 
n 1

Each of the terms in the first column is a factor of all of the terms in its row
and thus may be factored out of the determinant, resulting in the equivalent
expression

 
 1 a2 +a1 a22 +a2 a1 + a21 a2n1 + + a1n1 
n  
  1 a3 +a1 a23 +a3 a1 + a21 a3n1 + + a1n1 
(aj a1 )  .. .. .. .. .. .

j=2  . . . . . 
 1 an +a1 a2n +ana1 + a21 ann1 + + a1n1 

The last determinant can be simplified by multiplying the first column by

a1 and subtracting the product from the second column, then multiplying the
first column by a21 and subtracting the product from the third column, etc.,
resulting in the equivalent expression
 
 1 a2 a22 +a2 a1 a2n2 + + a1n3 a2 
n  
  1 a3 a23 +a3 a1 a3n2 + + a1n3 a3 
(aj a1 )  .. .. .. .. .. .

j=2  . . . . . 
 1 a a2n+an a1 ann2 + + a1n3 an 
n

The last expression can be further simplified by multiplying the second

column by a1 and subtracting the product from the third column, then multi-
plying the second column by a21 and subtracting the product from the fourth
76
column, etc., resulting in the equivalent expression
 
 1 a2 a22 a2n2 + + a1n4 a22 
n  
  1 a3 a23 a3n2 + + a1n4 a23 
(aj a1 )  .. .. .. .. .. .

j=2  . . . . . 
 1 a an an + + a1n4 a2n
2 n2 
n

Continuing to simplify in a similar fashion leads to the following expres-

sion, which is equal to the original Vandermonde determinant of order n :
 
 1 a2 a22 a2n2 
n  
  1 a3 a23 a3n2 
(aj a1 )  .. .. .. .. ..  .
j=2  . . . . . 
 1 a 2
an an n2 
n

Note that the determinant in this expression is another Vandermonde de-

terminant, but of order n 1. Evaluating it in the same way indicated above,
results in the expression
 
 1 a3 a23 a3n3 
n n  
   1 a4 a24 a4n3 

(aj a1 ) (ak a2 )  .. .. .. .. ..  ,
j=2 k=3  . . . . . 
 1 a a 2
an3 
n n n

and then similarly evaluating the resulting Vandermonde determinants of lower

order leads eventually to the desired formula. (A more rigorous proof would
employ mathematical induction. The argument above provides the crucial
inductive step in such a proof.)

6.2.1. Expanding along the first column, we get

 
 2 3 2       
   4 0   1 0   
 1 4 0  = 2   (3)  +2 1 4 
 1 5   0 5   0 1 
 0 1 5 
= 2 (20 + 0) + 3 (5 + 0) + 2 (1 0) = 53.

6.2.3. Expansion along the third row, and then along the second row,
77
gives:
 
 1 1 2 3   
   1 1 2 
 2 0 5 0   
  = (1)  2 0 5  =
 0 0 0 1  
   3 3 1 
 3 3 1 4 
   
 1 2   1 1 
2   (5)  = 2 (1 + 6) + 5 (3 3) = 14.
3 1 3 3 

6.2.5. Expanding the determinant in the numerator of Cramers rule

along the column b, we obtain

1 
xi = bj Aji .
|A| j

Substituting this expression for xi into Ax results in

 1  1  
(Ax)k = aki xi = bj aki Aji = (bj aki Aji ).
i
|A| i j |A| j i

By Lemma 6.2.1, the inside sum on the right is 0 if j = k, and, by Theorem

6.2.1, it is |A| if j = k. Thus the sum over j has only one nonzero term,
which equals bk |A|. Therefore (Ax)k = bk and Ax = b, as we had to show.
6.2.7. Using Cramer s Rule and the expansion of Theorem 6.1.3, we get

   
 1 1 2   1 1 2 
   
x1 =  2 0 4   2 0 4 
 3 3 1   0 3 1 
(0 + 12 + 12) (12 + 2 + 0) 10
= = = 5,
(0 + 12 + 0) (12 + 2 + 0) 2

and similarly, x2 = 0, x3 = 3.
 
1 3
6.2.9. A = implies A11 = 1, A12 = 2, A21 = 3, A22 = 1,
2 1

78
  
1 3
and so adj(A) = . Hence
2 1
 
1 1 1 1 3
A = adj(A) = .
|A| 5 2 1

6.2.11. From Theorem 6.2.3, adj(A) = |A|A1 . Apply the result det(cA)
= cn det(A) of Exercise 6.1.6 here, with c = |A|, and A1 in place of A, to
get det(adj(A)) = det(|A|A1 ) = |A|n det(A1 ). Now, by Corollary 6.1.2,
det(A1 ) = 1/|A|, and so

det(adj(A)) = (det(A))n1 .
6.2.13. The area A of the triangle with vertices (a1 , a2 ), (b1 , b2 ), (c1 , c2 )
is half that of the parallelogram spanned by the vectors b a and c a,
where a = (a1 , a2 ), b = (b1 , b2 ), and c = (c1 , c2 ). Thus, by Theorem 6.2.4,
   
1 1  b1 a1 c1 a1  1  b1 a1 b2 a2 

A = |b a, c a| =  =
2 2 b2 a2 c2 a2  2  c1 a1 c2 a2 
   
 0 0 1   a1 a 2 1 
1   
1 
= b1 a 1 b 2 a2 0  = b 1 a 1 b2 a2 0 
2  c a c a  2
0   c1 a1 c2 a2 0 

1 1 2 2
 
 a1 a2 1 
1  
= b 1 b2 1 .
2  c c 1 

1 2

6.2.15. If we expand the determinant along the first row, then we obtain
a linear combination of the elements of that row, which can be rearranged
into the form y = ax2 + bx + c, since the coefficient of y is a nonzero deter-
minant (see Exercise 6.1.14). Note that, by the result in Exercise 6.2.13, the
coefficient of x2 is also a nonzero determinant if the points (x1 , y1 ), (x2 , y2 ),
and (x3 , y3 ) are noncollinear. If they are collinear, then we get an equation of
a straight line, which may be considered to be a degenerate parabola. If we
substitute the coordinates of any of the given points for x, y, then two rows
become equal, which makes the determinant vanish. Thus the given points lie
on the parabola.
6.2.17. Applying the formula in Exercise 6.2.16, the equation of the circle
79
can be expressed in the form

 
 x2 + y 2 x y 1 
 
 0 0 0 1 
  = 0.
 5 2 1 1 
 
 16 4 0 1 
Expanding along the second row yields
 2 
 x + y2 x y 

 5 2 1  = 0.

 16 4 0 
Then expanding along the top row results in
     
 2 1  2    2 
  (x + y )  5 1
2 x+  5 y = 0.
 4 0   16 0   16 4 
From here, evaluating the coefficient determinants, we obtain
4(x2 + y 2 ) 16x 12y = 0, or, x2 + y 2 4x 3y = 0.
Completing the square of the terms in x and y results in
(x 2)2 + (y 3/2)2 = 25/4,
and so the circle has center (2, 3/2) and radius 5/2.
6.2.19. If we expand the determinant along the first row, then we obtain
a linear combination of the elements of that row. That combination is a linear
equation in x, y, and z, which is the equation of a plane. If we substitute the
coordinates of any of the given points for x, y, and z, then two rows become
equal, which makes the determinant vanish. Thus the given points lie on the
plane.

6.3.1. If u = (1, 1, 0)T and v = (1, 2, 0)T , then

 
 i j k   
   1 1 
u v =  1 1 0 =
  1 =3k.
 1  2 
2 0

80
 The triangle spanned by u and v has an altitude, parallel to the x-axis, of
length 1 and a corresponding base of length 3. Its area is 3/2, and thus the
area of the parallelogram spanned by the two vectors is 3, which is equal to
|u v| = |3k|.
6.3.3. First, we will verify Statement 11 of Theorem 6.3.1:
If u = (1, 1, 0)T , v = (1, 2, 0)T , and w = (1, 0, 3)T , then
 
 i j k 
 
v w =  1 2 0  = 6i 3j 2k
 1 0 3 

 
 i j k 
 
w u =  1 0 3  = 3i + 3j k

 1 1 0 

and, as shown in Exercise 6.3.1,

u v = 3k.

Thus

u (v w) = (1, 1, 0)T (6, 3, 2)T = 6 + 3 + 0 = 9,

v (w u) = (1, 2, 0)T (3, 3, 1)T = 3 + 6 + 0 = 9,

w (u v) = (1, 0, 3)T (0, 0, 3)T = 0 + 0 + 9 = 9,

and
 
 1 1 1 
 
det(u, v, w) =  1 2 0  = 9.

 0 0 3 

Next we will verify Statement 12:

 
 i j k 
T T
 
(u v) w = (0, 0, 3) (1, 0, 3) =  0 0 3  = 3j,

 1 0 3 

81
and
(u w)v (v w)u
= [(1, 1, 0)T (1, 0, 3)T ](1, 2, 0)T [(1, 2, 0)T (1, 0, 3)](1, 1, 0)T
= (1, 2, 0)T (1, 1, 0)T = (0, 3, 0)T = 3j.

Finally, we will verify Statement 13:

 
 i j k 
 
u (v w) = (1, 1, 0)T (6, 3, 2)T =  1 1 0 

 6 3 2 
= 2i + 2j + 3k = (2, 2, 3)T ,
and
(u w)v (u v)w
= [(1, 1, 0)T (1, 0, 3)T ](1, 2, 0)T [(1, 1, 0)T (1, 2, 0)T ](1, 0, 3)T
= (1, 2, 0)T + (1, 0, 3)T = (2, 2, 3)T .

6.3.5. Geometrically, in the case of nonzero vectors, v w is a vector

whose magnitude is equal to the area of the parallelogram spanned by the
vectors v and w and whose direction is orthogonal to the plane of that parallel-
ogram. Thus, if n = (v w)/|v w| denotes the unit vector in the direction
of v w, then |u n| is equal to the length of the orthogonal projection of u
onto the direction of v w, which is the height of the parallelepiped spanned
by the vectors u, v, and w. Therefore, the volume V of that parallelepiped is
given by
V = height area of base = |u n| |v w|
 
 vw 
= u  |v w| = |u (v w)|.
|v w| 
If either v or w is 0, then the parallelepiped spanned by the three vectors
degenerates into one of zero volume, and the scalar triple product is also zero
in this case.
6.3.7. If we expand the given expression by applying Statement 11 in
Theorem 6.3.1, with u = a b, v = c, and w = d, then we obtain
(a b) (c d) = c (d (a b))
82
Evaluating the triple cross product in the expression on the right side of the
above equation with the aid of Statement 12 in Theorem 6.3.1 results in
(a b) (c d)= c [(d b)a (d a)b]
= (d b)(c a) (d a)(c b)
= (a c)(b d) (a d)(b c).

6.3.9. a. 1. T is additive: By Part 6 of Theorem 6.3.1, we have T (x + y) =

a (x + y) = a x + a y =T (x)+T (y).
2. T is homogeneous: By Part 1 of Theorem 6.3.1 we have T (cx) =
a(cx) = c(a x) = cT (x).
b. The columns of [T ] are given by the action of T on the standard vectors.
Thus
t1 = T (i) = a i = (a1 i + a2 j + a3 k) i = a2 k + a3 j = (0, a3 , a2 )T ,
t2 = T (j) = a j = (a1 i + a2 j + a3 k) j = a1 k a3 i = (a3 , 0, a1 )T ,
t3 = T (k) = a k = (a1 i+a2 j+a3 k)k = a1 j+a2 i = (a2 , a1 , 0)T .
Hence

0 a3 a2
[T ] = a3 0 a1 .
a2 a1 0
c. Solve [T ]x = 0 by row reduction, assuming without loss of generality
a3 = 0:
 
0 a3 a2  0 a3 0 a1  0
a3 0 a1  0 0 a3 a2  0
a2 a1 0  0 a2 a1 0  0
 
a3 0 a1  0 a3 0 a1  0
0 a3 a2  0 0 a3 a2  0 .
0 a1 a1 a2 /a3  0 0 0 0  0

Now x3 is free, and so, set x3 = t. Then a3 x2 + a2 t = 0 gives x2 =

a2 t/a3 and a3 x2 a1 t = 0 gives x1 = a1 t/a3 . If we set t = ca3 , then we
obtain from the above x = ca. Thus the nullspace of the matrix [T ] is the line
of a. We could have found this result alternatively by observing that [T ]x = 0
is equivalent to T (x) = a x = 0, and the cross product is zero if and only
if its factors are parallel.
83
 d. Since, for any matrix, rank + nullity = number of columns, in the
present case we have rank([T ]) = 3 1 = 2. Alternatively, the rank is
the number of non-zero rows in the echelon form, which is clearly 2 in the
matrix above, because of the assumption a3 = 0. Since a was assumed to
be nonzero, at least one of its components has to be nonzero, and we end up
with two nonzero rows in the echelon form, regardless of which component
is nonzero.

7.1.1. The characteristic equation is

 
 2 3 
|A I| =   = 0,

3 2

or equivalently, 2 4 5 = ( + 1)( 5) = 0. The eigenvalues are

1 = 1 and 2 = 5.
To find the eigenvectors corresponding to 1 = 1, we need to solve the
equation
    
3 3 s1 0
(A 1 I)s = = .
3 3 s2 0

The solutions of this equation are of the form s = s(1, 1)T . Thus the
eigenvectors belonging to the eigenvalue 1 = 1 form a one-dimensional
subspace of R2 with basis vector s1 = (1, 1)T
To find the eigenvectors corresponding to 2 = 5, we need to solve the
equation
    
3 3 s1 0
(A 2 I)s = = .
3 3 s2 0
The solutions of this equation are of the form s = s(1, 1)T . Thus the eigen-
vectors belonging to the eigenvalue 2 = 5 form a one-dimensional subspace
of R2 with basis vector s2 = (1, 1)T .
7.1.3. The characteristic equation is

 
 0 0 
|A I| =   = 0,

0 0

84
or equivalently, 2 = 0. The only solution is = 0. Thus 0 is the only
eigenvalue and it has multiplicity 2. To find the corresponding eigenvectors,
we need to solve (A I)s = 0. Clearly, every s R2 is a solution, that
is, an eigenvector. In other words, = 0 is the sole eigenvalue with the
corresponding eigenspace being the whole of R2 .
7.1.5. The characteristic equation is
 
 2 0 1 
 

|A I| =  0 2 0  = 0,
 1 0 2 

or equivalently, (2)3 (2) = 0. Factoring on the left, we may write the

equation in the form (2 )[(2 )2 1] = 0, or (2 )(1 )(3 ) = 0.
Thus the eigenvalues are 1 = 1, 2 = 2 and 3 = 3.
To find the eigenvectors corresponding to 1 = 1, we need to solve the
equation

1 0 1 s1 0
(A 1 I)s = 0 1 0 s2 = 0 .
1 0 1 s3 0

The solutions of this equation are of the form s = s(1, 0, 1)T . Thus
the eigenvectors belonging to the eigenvalue 1 = 1 form a one-dimensional
subspace of R3 with basis vector s1 = (1, 0, 1)T
Next, to find the eigenvectors corresponding to 2 = 2, we need to solve
the equation

0 0 1 s1 0

(A 2 I)s = 0 0 0 s2 = 0 .

1 0 0 s3 0

The solutions of this equation are of the form s = s(0, 1, 0)T . Thus the
eigenvectors belonging to the eigenvalue 2 = 2 form a one-dimensional
subspace of R3 with basis vector s2 = (0, 1, 0)T .
Finally, to find the eigenvectors corresponding to 3 = 3, we need to solve
85
the equation

1 0 1 s1 0
(A 3 I)s = 0 1 0 s2 = 0 .
1 0 1 s3 0

The solutions of this equation are of the form s = s(1, 0, 1)T . Thus the
eigenvectors belonging to the eigenvalue 3 = 3 form a one-dimensional
subspace of R3 with basis vector s3 = (1, 0, 1)T .
7.1.7. The characteristic equation is
 
 1 0 1 
 
|A I| =  2 3 1  = 0.
 6 6 

We can reduce this equation to (1)(3)+126(3)+6(1) = 0;

or, equivalently, to (1 )(3 ) = 0. Thus the eigenvalues are 1 =
0, 2 = 1 and 3 = 3.
To find the eigenvectors corresponding to 1 = 0, we need to solve the
equation

1 0 1 s1 0

(A 1 I)s = 2 3 1 s2 = 0 .

6 6 0 s3 0

The solutions of this equation are of the form s = s(1, 1, 1)T . Thus the
eigenvectors belonging to the eigenvalue 1 = 0 form a one-dimensional
subspace of R3 with basis vector s1 = (1, 1, 1)T
Next, to find the eigenvectors corresponding to 2 = 1, we need to solve
the equation

0 0 1 s1 0

(A 2 I)s = 2 2 1 s2 = 0 .

6 6 1 s3 0

The solutions of this equation are of the form s = s(1, 1, 0)T . Thus the
eigenvectors belonging to the eigenvalue 2 = 1 form a one-dimensional
subspace of R3 with basis vector s2 = (1, 1, 0)T
86
 Finally, to find the eigenvectors corresponding to 3 = 3, we need to solve
the equation

2 0 1 s1 0

(A 3 I)s = 2 0 1 s2 = 0 .

6 6 3 s3 0

The solutions of this equation are of the form s = s(1, 0, 2)T . Thus
the eigenvectors belonging to the eigenvalue 3 = 3 form a one-dimensional
subspace of R3 with basis vector s3 = (1, 0, 2)T .
7.1.9. If the scalar c = 0, then we have
   
  n
 

|cA I| = c(A I) = c A I  .
c c
It follows from this relation that is an eigenvalue of the matrix cA with
c = 0, if and only if /c is an eigenvalue of A, and that the eigenvectors
corresponding to those related eigenvalues will be the same. On the other
hand, if c = 0, then cA = O, and its only eigenvalue is = 0 with eigenspace
Rn . (See Exercise 7.1.3).
7.1.11. First suppose that the matrix A is singular. Then As = 0 has a
nontrivial solution s, and As = 0s for all such solutions. This equation shows
that 0 is an eigenvalue of A, with all solutions s of As = 0 as its eigenvectors.
In the other direction, suppose that 0 is an eigenvalue of A. Then for all
corresponding nonzero eigenvectors s we have As = 0s = 0. By Part 6 of
Theorem 2.5.5, it follows that A is singular.
7.1.13. Suppose that s is an eigenvector of a nonsingular matrix A,
belonging to the eigenvalue . Then As = s. Since A is nonsingular, it is
invertible, and = 0 (see Exercise 7.1.11). Therefore, A1 (As) = A1 (s),
and thus s = A1 s, or A1 s = 1 s. Hence s is an eigenvector of A1
belonging to the eigenvalue 1 .
7.1.15. Apply the matrix A to the vector u. That gives Au = uuT u = u,
because uT u = 1. This shows that u is an eigenvector of A belonging to the
eigenvalue 1. If v is any vector orthogonal to u, then Av = uuT v = 0,
because uT v = 0. Thus any such vector v is an eigenvector of A belonging
to the eigenvalue 0. So the eigenvalues are 1 and (for n > 1 only) 0. The
87
corresponding eigenspaces are the line of u and its orthogonal complement,
respectively.
Now, any vector x Rn can be decomposed uniquely as the sum of its
projection cu onto u and a vector v orthogonal to u. Thus Ax = A(cu + v) =
cu. This equation shows that A represents the projection onto the line of u.
7.1.17. According to the definition, sT is a left eigenvector of a matrix
A belonging to the eigenvalue if and only if the equation sT A = sT holds.
This last equation is true if and only if (sT A)T = (sT )T , or, AT s = s, and
thus if and only if s is an eigenvector of AT belonging to the eigenvalue .
7.1.19. Suppose that s is an eigenvector of a matrix A, belonging to two
eigenvalues 1 and 2 = 1 . Then As = 1 s = 2 s. Hence (1 2 ) s = 0
and so, since (1 2 ) = 0, we must have s = 0.

7.2.1. By Exercise 7.1.11, if A is invertible, then all of its eigenvalues are

different from zero, even if it is not diagonalizable.
Conversely, if A is diagonalizable, say as A = SS 1 , and i = 0 for all
i, then

1
1 0 0
0 1 0
1 2
= .. .. .. ..
. . . .
1
0 0 n

exists, and X = S1 S 1 is A1 . Indeed, XA = S1 S 1 SS 1 =

S1 S 1 = SS 1 = I, and similarly AX = I as well.
Furthermore, for a diagonalizable and invertible A and any positive integer
88
k,
 k
1
1 0 0
 1 k
k
 k 0 2 0
= 1 = .. .. .. ..


. . . .
 1 k
0 0 n
k
1 0 0
0 k 0
2
= .. .. .. ..
. . . .
k
0 0 n

and
 k  k
Ak = A1 = S1 S 1 = Sk S 1 .

7.2.3. By Theorem 7.3.2, any symmetric matrix can be diagonalized with

an orthogonal matrix S, so that A = SS 1 . Furthermore, if all eigenvalues
of A are nonnegative, then writing
1/2
1 0 0
0 1/2 0
2
1/2 = . . . . ,
.. .. .. ..
0 0 1/2n

we can compute A as A = S1/2 S 1 , because then
 2  2  2
A = S1/2 S 1 = S 1/2 S 1 = SS 1 = A.

Furthermore, the orthogonality of S impliesS T = S 1 , and so ( A)T =
(S 1 )T (1/2 )T S T = S1/2 S 1 = A, that is, A is also symmetric.
7.2.5. A can be diagonalized with an orthogonal matrix as A = SS 1 .
The eigenvalues and eigenvectors of A were computed in Exercise 7.1.5. The
former make up the diagonal elements of , and the latter, after normalization,
the columns of S. Thus A100 = S100 S 1 can be evaluated as
89
 100
1 0 1 1 0 0 1 0 1
1
A100 = 0 2 0 0 2100 0 0 2 0
2 0 0 3100
1 0 1 1 0 1
100 100

3 +1 0 3 1
1
= 0 2101 0 .
2 3100 1 0 3100 + 1

7.2.7. If B = SAS 1 , then

|B I| = |S 1 AS I| = |S 1 AS S 1 IS|
= |S 1 (A I)S| = |S 1 ||A I||S| = |A I|,
since |S 1 | = 1/|S|.
7.2.9. If S 1 AS = , then A = SS 1 . The orthogonality of S implies
T
S = S 1 , and so
AT = (SS 1 )T = (S 1 )T T S T = SS 1 = A.

7.2.11. In matrix form, the given equations become

 
1000
x(0) =
1000
and
x(k + 1) = Ax(k),
where
 
0.8 0.4
A= ,
0.8 2.0
and so
x(n) = An x(0)
for any positive integer n. We diagonalize A to compute An here.
The characteristic equation for this A is
 
 0.8 0.4 
|A I| =   = (0.8 )(2 ) + 0.32 = 0.
0.8 2.0 

90
The solutions are 1 = 1.6 and 2 = 1.2. The corresponding eigenvectors
can be found by substituting these eigenvalues into (A I)s = 0:
   
0.8 1.6 0.4 0.8 0.4
(A 1 I)s1 = s1 = s1 = 0.
0.8 2.0 1.6 0.8 0.4

A solution is s1 = (1, 2)T .

For the other eigenvalue we have the equation
   
0.8 1.2 0.4 0.4 0.4
(A 2 I)s2 = s2 = s2 = 0.
0.8 2.0 1.2 0.8 0.8

A solution is s2 = (1, 1)T . Thus

 
1 1
S= ,
2 1
 
1.6 0
=
0 1.2
and
 
1 1 1
S = .
2 1

According to Corollary 3.6.1, the coordinate vectors of each x(n), for

n = 0, 1, . . ., relative to the basis S, are given by
    
1 1 1 1000 0
xS (0) = S x(0) = =
2 1 1000 1000
and
xS (n) = S 1 x(n) = S 1 An x(0) = S 1 An SxS (0) = n xS (0)
    
1.6n 0 0 n 0
= = 1.2 .
0 1.2n 1000 1000

Hence the solution in the standard basis is given by

    
n 1 1 0 n 1000
x(n) = SxS (n) = 1.2 = 1.2 .
2 1 1000 1000

91
 As the last equation shows, the proportion of predators to preys does not
change over time if they were evenly split in the beginning, and the numbers
of both increase by 20% each year, approaching infinity.

7.3.1. The matrix of the given quadratic form is

 
1 13 4
A= .
30 4 7

The eigenvalues and corresponding unit eigenvectors of this matrix are 1 =

1/6, 2 = 1/2, s1 = 15 (1, 2)T , s2 = 15 (2, 1)T Hence, the given equation
represents the
ellipse below.
It is centered at the origin, its major axis has
half-length 1/ 1 = 6 and pointsin thedirection of the eigenvector s1 ,
and its minor axis has half-length 1/ 2 = 2 and points in the direction of
the eigenvector s2 .

-1 1

-1

-2

7.3.3. The matrix of the given quadratic form is

 
1 2 2
A= .
12 2 1

The eigenvalues and corresponding unit eigenvectors of this matrix are

1 = 1/4, 2 = 1/6, s1 = 15 (2, 1)T , s2 = 15 (1, 2)T . Hence, the given
equation represents a hyperbola centered at the origin, whose axis has half-
length 1/ 1 = 2 and points in the direction of the eigenvector s1 , and so its
vertices are at x = 15 (4, 2)T .

92
 4

-4 -2 2 4

-2

-4

7.3.5. a.
  xi xj
((xT Ax))k = aij xi xj = aij ( xj + xi )
xk i j i j
xk xk
  
= aij ( ik xj + xi jk ) = akj xj + aik xi
i j j i
T
= (Ax)k + (x A)k = 2(Ax)Tk .

(The last equality above follows from the symmetry of A.) Hence (xT Ax) =
2(Ax)T .
b. Since xT x = xT Ix, Part (a) with A = I shows that (xT x) = 2xT .
Thus Lagranges equation (xT Ax) = (xT x) becomes 2(Ax)T = 2xT ,
from which we see that the Lagrange multiplier is an eigenvalue of A.
7.3.7. The construction in the proof of Theorem 7.3.2 remains unchanged
, except that if A is not symmetric, then Equation 7.107 does not hold and
Equation 7.108 becomes


1


0


1

S1 AS1 = .


A ,
.

. 1


0

93
where the stars denote appropriate, possibly nonzero entries. Similarly, the
0 entries in the first two rows of the matrix in Equation 7.110 may become
nonzero; and so on, until in Equation 7.112 all the entries above the diagonal
may be different from zero.
7.4.1.
a)
xH = (2, 2i) and yH = (5i, 4 i),
b)
 
2 H 2
|x| = x x = (2, 2i) = 4 4i2 = 8 and |x| = 8,
2i
 
2 H 5i
|y| = y y = (5i, 4 i) = 25 + 16 + 1 = 42 and |y| = 42,
4+i
c)
 
H 5i
x y = (2, 2i) = 10i 8i + 2 = 2 + 2i,
4+i
 
H 2
y x = (5i, 4 i) = 10i + 8i + 2 = 2 2i.
2i

7.4.3.
a)
xH = (2ei/4 , 2i) and yH = (ei/4 , ei/4 ),
b)
 i/4 
2 2e
|x| = (2e i/4
, 2i) = 4 4i2 = 8 and |x| = 8,
2i
c)
 i/4 
H i/4 e
x y = (2e , 2i) i/4 = 2 2iei/4 = 2 2 2i,
e

yH x = xH y = 2 2 + 2i.

7.4.5.  
x1 + iy1
a) Let u2 = . Then uH
1 u2 = 0 becomes x1 +iy1 ix2 +y2 =
x2 + iy2
0, which can be written in components as x1 + y2 = 0 and y1 x2 = 0. We
94
  
1
i
may choose x1 = y2 = 0 and y1 = x2 = 1 Thus u2 = is a unit
2 1
vector orthogonal to u1 .
b) The matrix
 
1 1 i
U = (u1 , u2 ) =
2 i 1
is unitary, and so
 
1 H 1 1 i
U =U = .
2 i 1
Just as in the real case, the coordinate vector xU is given by xU = U 1 x,
which in the present case becomes
    
1 1 i x1 1 x1 ix2
xU = = .
2 i 1 x2 2 x2 ix1
c) From the last equation, with x1 = 2 + 4i and x2 = 1 2i, we get
   
1 2 + 4i i (1 2i) 1 3i
xU = = .
2 1 2i i (2 + 4i) 2 5 4i
Hence
3i 5 4i
x = u1 + u2 .
2 2
7.4.7. AHH = (ajk )HH = (akj )H = (akj )T = (akj )T = (ajk ) = A.
7.4.9. For any complex number z, z H = z if and only if z is real. Apply-
ing this observation and Theorem 7.4.1 to z = xH Ax, and using (xH Ax)H =
xH AH xHH = xH AH x, we find that xH AH x = xH Ax for every x, if and
only if xH Ax is real for every x. Thus if AH = A, then xH Ax is real for
every x.
Conversely, assume that xH Ax is real for every x. Write x + y in place of
x. Then (x + y)H AH (x + y) = (x + y)H A(x + y) for all x, y, is equivalent
to xH AH y + yH AH x = xH Ay + yH Ax for all x, y. Choose x = ej and
y = ek . Then the last equation becomes

(AH )jk + (AH )kj = ajk + akj .

95
Next, choosing x = ej and y = iek , we get

i(AH )jk i(AH )kj = iajk iakj .

Multiplying the last equation by i and subtracting it from the previous one,
we obtain
(AH )jk = ajk .
Since this equation holds for all jk, we have AH = A, as was to be shown.
7.4.11. If AH = A, then
 H  H
   (iAt) k  (iAt)k
H itA H
U = e = =
k=0
k! k=0
k!

   k

 (iAt)H  (iAt)k
= = = eAt = U 1 .
k=0
k! k=0
k!

7.4.13. Suppose A is Hermitian and s1 and s2 are nonzero eigenvectors

belonging to the distinct eigenvalues 1 and 2 respectively. Then

As1 = 1 s1 and As2 = 2 s2 .

Multiplying these equations by sH H

2 and s1 , we get

sH H H H
2 As1 = 1 s2 s1 and s1 As2 = 2 s1 s2 .

Taking the Hermitian conjugate of the last equation and using AH = A, we

can change it to
sH H
2 As1 = 2 s2 s1 .

Thus
(2 1 )sH
2 s1 = 0,

and, since 2 1 = 0, we must have sH

2 s1 = 0.
7.4.15. The characteristic equation is
 
 1 1 
|A I| =   = 0,

1 1

96
or equivalently, (1 )2 + 1 = 0. Hence, the eigenvalues are 1 = 1 + i and
2 = 1 i.
To find the eigenvectors corresponding to 1 = 1 + i, we need to solve the
equation
    
i 1 s1 0
(A 1 I)s = = .
1 i s2 0
The general solution is s1 = t(1, i)T , where t is an arbitrary complex param-
eter.
To find the eigenvectors corresponding to 2 = 1 i, we need to solve the
equation

    
i 1 s1 0
(A 2 I)s = = .
1 i s2 0
The general solution is s2 = u(i, 1)T , where u is an arbitrary complex param-
eter.
8.1.1.

1 0 0 1 0 0 1 0 0
E32 E31 E21 = 0 1 0 0 1 0 3 1 0
0 1 1 1 0 1 0 0 1

1 0 0 1 0 0
= 0 1 0 3 1 0
0 1 1 1 0 1

1 0 0
= 3 1 0
4 1 1

The last matrix is L1 , and it too is lower diagonal. Its 4 entry does not,
however, come from a single lij , but it is the combination (l21 + l31 ) of
two coefficients of the forward elimination algorithm. While the order of
the matrix multiplications in Equation 8.17 has kept the off-diagonal entries
separate, the reverse order used here mixes them up.
8.1.3. Let A = AT and A = LDU. Since D is diagonal, we then have
97
DT = D and U T DLT = LDU. As we know, L is invertible, and therefore LT
must be invertible, too. Thus, multiplying the last equation by L1 from the
left and by (LT )1 from the right, we obtain L1 U T D = DU (LT )1 . Notice
that the product on the left is lower diagonal, and the one on the right is upper
diagonal. This is possible only if both are diagonal, which, for invertible D,
implies that L1 U T = C must be diagonal. (If D is singular, that is, it has
some zeros on its main diagonal, then the columns of C corresponding to the
nonzero entries of D must still contain only diagonal nonzero elements, and
the other columns, being arbitrary, can be chosen to contain only diagonal
nonzero elements.) Since, by Corollary 8.1.1, the diagonal elements of L1
and U T are all 1, the diagonal elements of their product must be ls too. Thus
C = I and U T = L must hold.
8.1.5. In the forward phase of Gaussian elimination, assuming no row
exchanges are needed, to get a 0 in place of a21 , we compute l21 = a21 /a11 and
subtract l21 a1j from each element a2j of the second row for j = 2, 3, . . . , n.
Thus again, as in the n n case, we need n long operations for the reduction
of the second row of A.
Next, we do the same for each of the other rows below the first row. Thus
to get all the m1 zeros in the first column requires (m1)n long operations.
Now we do the same for the (m 1) (n 1) submatrix below and to
the right of a11 . For this we need (m 2)(n 1) long operations.
Continuing in this manner, we find that the total number of long operations
needed for the reduction of A, in case m n, is
m
 m1

((m 1) k)(n k) = [(m 1)n (m + n 1)k + k 2 ] =
k=1 k=1

m(m 1) m(m 1)(2m 1) m2 n m3

(m 1)2 n (m + n 1) + = .
2 6 2 6
If m > n, then the reduction must stop when k = n. So the total number of
long operations is then
n
 n

((m 1) k)(n k) = [(m 1)n (m + n 1)k + k 2 ] =
k=1 k=1

98
 n(n + 1) n(n + 1)(2n + 1) mn2 n3
(m 1)n2 (m + n 1) + = .
2 6 2 6
Since in both of the foregoing cases the reduction of A produces L and
U as well, the formulas above give also the numbers of the long operations
needed for the LU factorization of A.
8.1.7. By Equation 8.34, the forward elimination phase requires approxi-
mately n3 /3 long operations.
Assuming that none of the n pivots is zero, we need approximately n2 /2
divisions to change them to ls, because we have approximately that many
possibly nonzero elements in the echelon matrix, which must each be divided
by a pivot. Then we need n 1 multiplications and subtractions on the right
side to produce zeros in the last column above the last pivot. (The zeros
need not be computed, only the corresponding numbers on the right must be.)
Similarly, we need n 2 multiplications and subtractions on the right side to
produce zeros in the next to last column, and so on. Hence the total number
of such operations is n(n 1)/2 = n2 /2.
For large n the number n2 of second phase operations is negligible com-
pared to the number n3 /3 in the first phase, and so the latter provides the
approximate necessary total.
8.1.9. The reduction to upper triangular form requires the same operations
as the forward elimination phase of Gaussian elimination, which, by Equation
8.34, uses approximately n3 /3 long operations when n is large. To multiply
the diagonal entries we need n 1 multiplications, and this number can be
neglected next to n3 /3 when n is large.
8.2.1. The first step of Gaussian Elimination would produce
  
0.002 1  4
0 3001  11998
and our machine would round the second row to give
  
0.002 1  4
0 3000  12000

The machine would then solve this by back substitution and obtain x2 = 4
and x1 = 0 This solution is, however, wrong. The correct solution is x2 =
99
11998 6
3001
= 3.9980 . . . and x1 = 6.002 = 0.9996 . . .. Thus, while the machines
answer for x2 is close enough, for x1 it is way off.
The reason for the discrepancy is this: In the first step of the back substi-
tution, the machine rounded x2 = 3.998 . . . to 4. This, in itself, is certainly
all right, but in the next step, the machine had to divide x2 by 0.002 in solving
for x1 . Here the small roundoff error, hidden in taking x2 as 4, got magnified
five hundredfold.
8.2.3. The scale factors are s1 = 1 and s2 = 6, and the ratios r1 = 0.0002
and r2 = 1. Since r2 > r1 , we put the second row on top:

  
6 1  2
[A|b] =
0.002 1  4
Now we subtract 0.0002/6 times the first row from the second, to get
  
6 1  2
 .
0 1.00033 . . .  3.99933 . . .

The machine of Example 8.2.1 would round the above matrix to

  
6 1  2
0 1  4
and solve the system from here as x2 = 4 and x1 = 1.
The correct values are, as we have seen in the solution of Example 8.2.1,
x2 = 11998
3001
6
= 3.998 . . . and x1 = 6.002 = 0.9996 . . .. Thus, this method has
produced excellent approximations. The reason for this success is that here
we did not have to divide the roundoff error by a small pivot in the course of
the back substitution.
8.2.5. The scale factors are s1 = 5, s2 = 11 and s3 = 9, and the ratios
r1 = 1/5, r2 = 4/11 and r3 = 5/9. Since r3 > r2 > r1 , we put the third row
on top, and then proceed with the row reduction:
 
5 8 9  1 5 8 9  1
1 2 5  1 0 2 16  4
4 7 11 0 0 67 19  4

The new scale factors are s2 = 16 and s3 = 67, and the ratios r2 = 2/16 and
100
r3 = 67/67 = 1. Thus we switch the last two rows:
 
5 8 9  1 5 8 9  1

0 67 
19  4 0 67 19  4 .

0 2 16  4 0 0 1110  260

Back substitution now gives x3 = 26/111, x2 = 14/111, and x1 = 47/111.

8.3.1. a. Setting

1
x0 = x0 = 1 ,
1
we get

4.0000 1
x1 = Ax0 = 3.0000 and x1 = 0.7500 ,
1.0000 0.2500

3.2500 1
x2 = Ax1 = 2.2500 and x2 = 0.6923 ,
0.2500 0.0769

3.0769 1
x3 = Ax2 = 2.0769 and x3 = 0.6750 ,
0.0769 0.0250

3.0250 1
x4 = Ax3 = 2.0250 and x4 = 0.6694 ,
0.0250 0.0083

3.0083 1
x5 = Ax4 = 2.0083 and x5 = 0.6676 .
0.0083 0.0028
Thus 1 3 and s1 (3, 2, 0)T .
If we start with

1
x0 = x0 = 1 ,
1
then we get
101

4.0000 1
x1 = Ax0 = 3.0000 and x1 = 0.7500 ,
1.0000 0.2500

3.2500 1
x2 = Ax1 = 2.2500 and x2 = 0.6923 ,
0.2500 0.0769

3.0769 1
x3 = Ax2 = 2.0769 and x3 = 0.6750 ,
0.0769 0.0250

3.0250 1
x4 = Ax3 = 2.0250 and x4 = 0.6694 ,
0.0250 0.0083

3.0083 1
x5 = Ax4 = 2.0083 and x5 = 0.6676 .
0.0083 0.0028
Thus 1 3 and s2 (3, 2, 0)T .
b. As we can see from the computations above, if a dominant eigenvalue
has geometric multiplicity greater than 1, then different initial vectors may
lead to different eigenvectors. Clearly, any eigenvector belonging to such a
dominant eigenvalue can be so obtained by an appropriate choice of the initial
vector, since if we started with an eigenvector, the method would stay with it.
It is only this observation that needs to be added to Theorem 8.3.1.
8.3.3. The eigenvalues of a matrix A are the solutions of its charac-
teristic equation |A I| = 0. Hence if c is not an eigenvalue of A, then
|A cI| =  0. By Theorem 6.1.8, this inequality implies that B = A cI is
nonsingular.
Alternatively, by Exercises 7.1.10 and 11, the eigenvalues of B = A cI
are the c values, which are nonzero if c = , and a matrix with only
nonzero eigenvalues is nonsingular.
A2.1. Equation A.22 follows from Equations A.20, A.17, and A.18, since
| cos + i sin |2 = cos2 + sin2 = 1 for any .

n k
A2.3. If P (z) = k=0 ak z is a polynomial with real coefficients ak ,
then, by Equations A.12 and A.13, generalized to arbitrary finite sums and

102
products, it follows that
n
 n
 n

P (z) = ak zk = ak zk = ak z k = P (z).
k=0 k=0 k=0

Thus P (z0 ) = 0 is equivalent to P (z 0 ) = P (z0 ) = 0 = 0.

A2.5. Suppose that n=0 z n converges absolutely; that is, n=0 |zn | con-
verges. Then since |xn | |zn | and |y

n | |z |
n
for all indices n, it follows from

the Comparison Test that both n=0 xn and n=0 yn converge absolutely
and thus both converge by the Absolute
Convergence Test for real series. The
result of Exercise A.2.4 implies that n=0 zn also converges.
A2.7. In order to use Formula A.33 to find the roots of a given complex
number, it is first necessary to express the number in polar exponential form,
that is, in the form of Equation A.29.
a. Since |i| = 1 and the principal argument of i is /2, i = ei((/2)+2k)
and thus i1/2 = ei((/4)+k) . Distinct roots are obtained for k = 0 and 1:
k = 0 yields

i/4 2
z1 = e = (1 + i),
2
and k = 1 yields

i(5/4) 2
z2 = e = (1 + i).
2

b. Since |1 + i| = 2 and the principal argument of 1 + i is /4,

1 + i = 2ei((/4)+2k) ,
and thus

(1 + i)1/2 = 2ei((/8)+k) .
4

Distinct roots are obtained for k = 0 and 1:

k = 0 yields

z1 = 2ei/8 = 2(cos(/8) + i sin(/8))
4 4

and k = 1 yields

2ei(5/8) = 2(cos(5/8) + i sin(5/8))
4 4
z2 =
103

4
= 2(cos(/8) + i sin(/8)).
c. Since |1| = 1 and the principal argument of 1 is 0, 1 = ei2k and thus
1 = ei(2k/3) . Distinct roots are obtained for k = 0, 1, and 2:
1/3

k = 0 yields
z1 = e0 = 1,
k = 1 yields
1
z2 = ei(2/3) = (1 + 3i),
2
and k = 2 yields
1
z3 = ei(4/3) = (1 3i).
2
d. Since | 1| = 1 and the principal argument of 1 is ,
1 = ei(+2k) ,
and thus
(1)1/3 = ei(2k+1)/3 .
Distinct roots are obtained for k = 0, 1, and 2:
k = 0 yields
1
z1 = ei/3 = (1 + 3i),
2
k = 1 yields
z2 = ei = 1,
and k = 2 yields
1
z3 = ei5/3 = (1 3i).
2
e. From part (a),
i = ei((/2)+2k) ,
and thus
i1/4 = ei((/8)+(k/4)) .
104
Distinct roots are obtained for k = 0, 1, 2, and 3:
k = 0 yields
z1 = ei/8 = cos(/8) + i sin(/8),
k = 1 yields
z2 = ei5/8 = cos(5/8) + i sin(5/8),
k = 2 yields
z3 = ei9/8 = cos(9/8) + i sin(9/8) = [cos(/8) + i sin(/8)],
and k = 3 yields
z4 = ei13/8 = cos(13/8) + i sin(13/8) = [cos(5/8) + i sin(5/8)].

105
http://www.springer.com/978-0-8176-8324-5