Linear Algebra (SM), Kuttler
Linear Algebra (SM), Kuttler
Linear Algebra (SM), Kuttler
Exercises
Exercises
1.6
1. Let z = 5 + i9. Find z 1 .
1
(5 + i9)
5
106
9
106 i
37
53 i.
x4 + 16 = 0, Solution is: (1 i) 2, (1 + i) 2, (1 i) 2, (1 + i) 2.
4. Graph the complex cube roots of 8 in the complex plane. Do the same for the four
fourth roots of 16.
(1 i) 2, (1 + i) 2, (1 i) 2, (1 + i)
2. When you graph these, you will have three equally spaced points on the circle of
radius 2 for the cube roots and you will have four equally spaced points on the circle
of radius 2 for the fourth roots. Here are pictures which should result.
6. De Moivres theorem says [r (cos t + i sin t)] = rn (cos nt + i sin nt) for n a positive
integer. Does this formula continue to hold for all integers, n, even negative integers?
Explain.
Yes, it holds for all integers. First of all, it clearly holds if n = 0. Suppose now that
n is a negative integer. Then n > 0 and so
n
1
1
n = n
r
(cos
(nt)
+ i sin (nt))
[r (cos t + i sin t)]
Exercises
=
rn
rn (cos (nt) + i sin (nt))
=
(cos (nt) i sin (nt))
(cos (nt) i sin (nt)) (cos (nt) + i sin (nt))
n
r (cos (nt) + i sin (nt))
=
=
|z| |w| (cos cos sin sin + i (cos sin + cos sin ))
10. Write x3 + 27 in the form (x + 3) x2 + ax + b where x2 + ax + b cannot be factored
any more using only real numbers.
x3 + 27 = (x + 3) x2 3x + 9
x4 + 16 = x2 2 2x + 4 x2 + 2 2x + 4 . You can use the information in the
preceding problem. Note that (x z) (x z) has real coefficients.
Exercises
(a ib) (c id) = ac bd i (ad + bc) which is the same thing. Thus it holds for
a product of two complex numbers. Now suppose you have that it is true for the
product of n complex numbers. Then
z1 zn+1 = z1 zn zn+1
and now, by induction this equals
z1 zn zn+1
As to sums, this is even easier.
n
X
(xj + iyj ) =
j=1
n
X
j=1
xj i
n
X
xj + i
j=1
n
X
yj =
j=1
n
X
j=1
xj iyj =
n
X
yj
j=1
n
X
(xj + iyj ).
j=1
14. Suppose p (x) = an xn + an1 xn1 + + a1 x + a0 where all the ak are real numbers.
Suppose also that p (z) = 0 for some z C. Show it follows that p (z) = 0 also.
You just use the above problem. If p (z) = 0, then you have
p (z) = 0 = an z n + an1 z n1 + + a1 z + a0
= an z n + an1 z n1 + + a1 z + a0
= an z n + an1 z n1 + + a1 z + a0
= an z n + an1 z n1 + + a1 z + a0
= p (z)
2
1 1 = (1) = 1 = 1.
This is clearly a remarkable result but is there something wrong with it? If so, what
is wrong?
1/4
It is clear that 1 > 0 because 1 = 12 . (In general a2 > 0. This is clear if a > 0. If a < 0,
then adding a to both sides, yields that 0 < a. Also recall that a = (1) a and
2
2
2
that (1) = 1. Therefore, (a) = (1) a2 = a2 > 0. ) Now it follows that if C can
be ordered, then 1 > 0, but this is a problem because it implies that 0 > 1 = 12 > 0.
Exercises
18. Say a + ib < x + iy if a < x or if a = x, then b < y. This is called the lexicographic
order. Show that any two different complex numbers can be compared with this order.
What goes wrong in terms of the other requirements for an ordered field.
From the definition of this funny order, 0 < i and so if this were an order, you would
need to have 0 < i2 = 1. Now add 1 to both sides and obtain 0 > 1 = 12 > 0, a
contradiction.
19. With the order of Problem 18, consider for n N the complex number 1 n1 . Show
that with the lexicographic order just described, each of 1 im is an upper bound to
all these numbers. Therefore, this is a set which is bounded above but has no least
upper bound with respect to the lexicographic order on C.
This follows from the definition. 1 im > 1 1/n for each m. Therefore, if you
consider the numbers 1 n1 you have a nonempty set which has an upper bound but
no least upper bound.
F.13
Exercises
1.11
1. Give the complete solution to the system of equations, 3x y + 4z = 6, y + 8z = 0,
and 2x + y = 4.
x = 2 4t, y = 8t, z = t.
=
=
5, 3x + 2y + z = 7,
7, x + 3z = 5.
x = 2, y = 0, z = 1.
6. Give the complete solution of the system of equations,
x + 2y + 3z
4x + 5y + 5z
= 5, 3x + 2y + 2z = 7
= 7, x = 5
No solution.
Exercises
7. Give the complete solution of the system of equations
x + y + 3z
4x + 9y + z
2, 3x y + 5z = 6
8, x + 5y + 7z = 2
x = 2 2t, y = t, z = t.
8. Determine a such that there are infinitely many solutions and then find them. Next
determine a such that there are no solutions. Finally determine which values of a
correspond to a unique solution. The system of equations is
3za2 3a + x + y + 1 = 0
3x a y + z a2 + 4 5 = 0
za2 a 4x + 9y + 9 = 0
If a = 1, there are infinitely many solutions of the form x = 2 2t, y = t, z = t. If
a = 1, t then there are no solutions. If a is anything else, there is a unique solution.
2a
1
1
x=
,y =
,z =
a+1
a+1
a+1
2, z + w = 0,
2x + 2y + z w
4, x + y 4z 5z = 2
F.14
Exercises
1.14
1. Verify all the properties 1.11-1.18.
2. Compute 5 (1, 2 + 3i, 3, 2) + 6 (2 i, 1, 2, 7) .
3. Draw a picture of the points in R2 which are determined by the following ordered
pairs.
(a)
(b)
(c)
(d)
(1, 2)
(2, 2)
(2, 3)
(2, 5)
Exercises
F.15
Exercises
1.17
1. Show that (a b) =
1
4
h
i
2
2
|a + b| |a b| .
2. Prove from the axioms of the inner product the parallelogram identity, |a + b| +
|a b|2 = 2 |a|2 + 2 |b|2 .
P
3. For a, b Rn , define a b nk=1 k ak bk where k > 0 for each k. Show this satisfies
the axioms of the inner product. What does the Cauchy Schwarz inequality say in
this case.
The product satisfies all axioms for the inner product so the Cauchy Schwarz inequality
holds.
4. In Problem 3 above, suppose you only know k 0. Does the Cauchy Schwarz inequality still hold? If so, prove it.
Yes, it does. You dont need the part which says that the only way a a = 0 is for
a = 0 in the argument for the Cauchy Schwarz inequality.
5. Let f, g be continuous functions and define
Z
f g
f (t) g (t)dt
show this satisfies the axioms of a inner product if you think of continuous functions
in the place of a vector in Fn . What does the Cauchy Schwarz inequality say in this
case?
The only part which is not obvious for the axioms is the one which says that if
Z
|f | = 0
f (t) dt
Z
b
f (t) dt
a
!2
(b a)
1 dt
a
1/2
= (b a)
!1/2
Z
f (t) dt.
|f (t)| dt
2
|f (t)| dt
!1/2
!1/2
which yields the desired inequality when you square both sides.
Exercises
F.16
Exercises
2.2
1. In 2.1 - 2.8 describe A and 0.
2. Let A be an nn matrix. Show A equals the sum of a symmetric and a skew symmetric
matrix.
A=
A+AT
2
AAT
2
3. Show every skew symmetric matrix has all zeros down the main diagonal. The main
diagonal consists of every entry of the matrix which is of the form aii . It runs from
the upper left down to the lower right.
You know that Aij = Aji . Let j = i to conclude that Aii = Aii and so Aii = 0.
4. Using only the properties 2.1 - 2.8 show A is unique.
Suppose that B also works. Then
A = A + (A + B) = (A + A) + B = B
5. Using only the properties 2.1 - 2.8 show 0 is unique.
If 0 is another additive identity, then 0 = 0 + 0 = 0
6. Using only the properties 2.1 - 2.8 show 0A = 0. Here the 0 on the left is the scalar 0
and the 0 on the right is the zero for m n matrices.
0A = (0 + 0) A = 0A + 0A. Now add the additive inverse of 0A to both sides.
7. Using only the properties 2.1 - 2.8 and previous problems show (1) A = A.
8. Prove 2.17.
T
(AB)ij (AB)ji =
Ajk Bki =
T T
Bik
Akj = B T AT
ij
n
10. Let A and
y Rm . Show (Ax, y)Rm =
be a real m n matrix and let x R and
T
k
x,A y Rn where (, )Rk denotes the dot product in R .
P
P P
(Ax, y) = i (Ax)i yi = i k Aik xk yi
P
P
P P
x,AT y = k xk i AT ki yi = k i xk Aik yi , the same as above. Hence the two
are equal.
T
11. Use the result of Problem 10 to verify directly that (AB) = B T AT without making
any reference to subscripts.
(AB)T x, y (x, (AB) y) = AT x,By = B T AT x, y . Since this holds for every
x, y, you have for all y
T
(AB) x B T AT x, y
T
Exercises
12. Let x = (1, 1, 1) and y = (0, 1, 2) . Find xT y and xyT if possible.
1
0 1 2
xT y = 1 0 1 2 = 0 1 2
1
0 1
2
0
xyT = 1 1 1 1 = 1
2
13. Give
1
1
1
1
an example
1
1
1
1
1
1
1
1
1
14. Let A = 2
1
if possible.
1
1
1 3
1
1
2
1 , B =
0 . Find
, and C = 1 2
2 1 2
2
3 1 0
(a) AB
(b) BA
(c) AC
(d) CA
(e) CB
(f) BC
15. Consider the following digraph.
Write the matrix associated with this digraph and find the number of ways to go from
3 to 4 in three steps.
0 1 1 0
1 0 0 1
Exercises
0
1
1
0
1
0
1
1
1
0
0
0
3
0
1
1
= 3
4
2
1
1
5
2
5
3
2
0
1
1
4
5
8
3
16. Show that if A1 exists for an n n matrix, then it is unique. That is, if BA = I and
AB = I, then B = A1 .
From the given equations, multiply on the right by A1 . Then B = A1 .
17. Show (AB)
= B 1 A1 .
ABB 1 A1 = AIA1 = I
B 1 A1 AB = B 1 IB = I
Then by the definition of the inverse and its uniqueness, it follows that (AB)
and
(AB)1 = B 1 A1
exists
1
T
18. Show that if A is an invertible n n matrix, then so is AT and AT
= A1 .
T
T
AT A1 = A1 A = I
T
T
A1 AT = AA1 = I Then from the definition of the inverse and its uniqueness,
T
1
it follows that AT
exists and equals A1 .
19. Show that if A is an n n invertible matrix and x is a n 1 matrix such that Ax = b
for b an n 1 matrix, then x = A1 b.
Multiply both sides on the left by A1 .
x1 x2 + 2x3
x1
2x3 + x1
in the form A x2 where A is an appropriate matrix.
22. Write
3x3
x3
3x4 + 3x2 + x1
x4
1 1 2 0
x1
x1 x2 + 2x3
1 0 2 0 x2
x1 + 2x3
0 0 3 0 x3 =
3x3
1 3 0 3
x4
x1 + 3x2 + 3x4
23. Give another example other than the one given in this section of two square matrices,
A and B such that AB 6= BA.
Almost anything works.
1 2
1 2
5 2
=
3 4
2 0
11 6
10
Exercises
1
2
2
0
1
3
2
4
7
2
10
4
24. Suppose A and B are square matrices of the same size. Which of the following are
correct?
2
(a) (A B) = A2 2AB + B 2
Note this.
Find
1
2
1
28. Let
Find
1
2
1
29. Let
1 2
A= 2 1
1 0
3
4 .
2
3
4 .
2
2 3
2 4 5
1 4 = 0
1 2
0 2
1 2 3
1 0
A= 2 3
1 0
0 3
2 0
3
3 4 = 0 13 32
0 2
1 0 1
1 2
A= 2 1
4 5
3
4 .
10
11
Exercises
Find
1
2
4
30. Let
Find
1
1
2
1
F.17
2 3
1 4 , row echelon form:
5 10
1 0 53
0 1 23 A has no inverse.
0 0 0
1
1
A=
2
1
2 0
1 2
1 3
2 1
2
0
2
2
1
1
1 21
2 0 2
2
2
1
1 2 0
12 25
2
= 3
1 0
1 3 2
0
1
1
9
2 1 2
2 43
4
4
Exercises
2.7
1. Show the map T : Rn Rm defined by T (x) = Ax where A is an m n matrix and
x is an m 1 column vector is a linear transformation.
This follows from matrix multiplication rules.
2. Find the matrix for the linear transformation which rotates every vector in R2 through
an angle of /3.
1
cos (/3) sin (/3)
21 3
2
=
1
1
sin (/3) cos (/3)
2 3
2
3. Find the matrix for the linear transformation which rotates every vector in R2 through
an angle of /4.
1
cos (/4) sin (/4)
2 21 2
2
=
1
1
sin (/4) cos (/4)
2 2
2 2
4. Find the matrix for the linear transformation which rotates every vector in R2 through
an angle of /3.
1
1
cos (/3) sin (/3)
2
2 3
=
1
sin (/3) cos (/3)
21 3
2
5. Find the matrix for the linear transformation which rotates every vector in R2 through
an angle of 2/3.
2 cos (/3) 2 sin (/3)
1 3
=
2 sin (/3) 2 cos (/3)
3
1
6. Find the matrix for the linear transformation which rotates every vector in R2 through
an angle of /12. Hint: Note that /12 = /3 /4.
12
Exercises
cos (/3) sin (/3)
cos (/4) sin (/4)
sin (/3) cos (/3)
sin (/4) cos (/4)
1
1
1
1
4 23
4 23 + 4 2
4 2
=
1
1
1
1
4 2 3 4 2
4 2 3+ 4 2
7. Find the matrix for the linear transformation which rotates every vector in R2 through
an angle of 2/3 and then reflects across the x axis.
1
1 0
cos (2/3) sin (2/3)
12 3
2
=
1
0 1
sin (2/3) cos (2/3)
21 3
2
8. Find the matrix for the linear transformation which rotates every vector in R2 through
an angle of /3 and then reflects across the x axis.
1
12 3
1 0
cos (/3) sin (/3)
2
=
0 1
sin (/3) cos (/3)
21 3
12
9. Find the matrix for the linear transformation which rotates every vector in R2 through
an angle of /4 and then reflects across the x axis.
1
1
1 0
cos (/4) sin (/4)
2
2
2
2
=
1
1
0 1
sin (/4) cos (/4)
2 2 2 2
10. Find the matrix for the linear transformation which rotates every vector in R2 through
an angle of /6 and then reflects across the x axis followed by a reflection across the
y axis.
1
1
1 0
1 0
cos (/6) sin (/6)
2 3
2
=
0 1
0 1
sin (/6) cos (/6)
21
21 3
11. Find the matrix for the linear transformation which reflects every vector in R2 across
the x axis and then rotates every vector through an angle of /4.
1
1
cos (/4) sin (/4)
1 0
2
2
2
2
=
1
1
sin (/4) cos (/4)
0 1
2
2
2 2
12. Find the matrix for the linear transformation which rotates every vector in R2 through
an angle of /4 and next reflects every vector across the x axis. Compare with the
above problem.
1
1
1 0
cos (/4) sin (/4)
2
2
2
2
=
1
1
0 1
sin (/4) cos (/4)
2 2 2 2
13. Find the matrix for the linear transformation which reflects every vector in R2 across
the x axis and then rotates every vector through an angle of /6.
1
1
cos (/6) sin (/6)
1 0
2 3
2
=
1
sin (/6) cos (/6)
0 1
21 3
2
14. Find the matrix for the linear transformation which reflects every vector in R2 across
the y axis and then rotates every vector through an angle of /6.
1
cos (/6) sin (/6)
1 0
2 3 12
=
1
sin (/6) cos (/6)
0 1
12
2 3
13
Exercises
15. Find the matrix for the linear transformation which rotates every vector in R2 through
an angle of 5/12. Hint: Note that 5/12 = 2/3 /4.
=
cos (2/3) sin (2/3)
cos (/4) sin (/4)
sin (2/3) cos (2/3)
sin (/4) cos (/4)
1
1
1
1
3 4 2
4 23 4 2 4 2
1
1
1
1
4 2 3+ 4 2
4 2 3 4 2
16. Find the matrix for proju (v) where u = (1, 2, 3)T .
T ei =
ei (1,2,3)
14
(1, 2, 3)
1
2
3
1
2 4 6
14
3
6
9
17. Find the matrix for proju (v) where u = (1, 5, 3)T .
1
1
5
35
3
5 3
25 15
15 9
T
1 0
1
0 0
10
3 0
3
0
9
19. Give an example of a 2 2 matrix A which has all its entries nonzero and satisfies
A2 = A. Such a matrix is called idempotent.
You know it cant be invertible. So try this.
2 2
a a
a + ba a2 + ba
=
b b
b2 + ab b2 + ab
Let a2 + ab = a, b2 + ab = b. A solution which yields a nonzero matrix is
2
2
1 1
20. Let A be an m n matrix and let B be an n m matrix where n < m. Show that
AB cannot have an inverse.
This follows right away from Theorem 2.3.8. This theorem says there exists a vector
x 6= 0 such that Bx = 0. Therefore, ABx = 0 also and so AB cannot be invertible.
21. Find ker (A) for
1
0
A=
1
0
2
2
4
2
3
1
4
1
2
1
3
1
1
2
.
3
2
14
Exercises
Recall ker (A) is
1 2 3 2
0 2 1 1
1 4 4 3
0 2 1 1
1 0
1
0
2 0
. After row operations,
0
3 0
2 0
0
0
1
0
0
1
2
1
2
0
0
0
0
1
1
0
0
0
0
0
0
22. If A is a linear transformation, and Axp = b. Show that the general solution to the
equation Ax = b is of the form xp + y where y ker (A). By this I mean to show
that whenever Az = b there exists y ker (A) such that xp + y = z.
If Az = b, Then A (z xp ) = Az Axp = b b = 0 so there exists y such that
y ker (A) and xp + y = z.
23. Using Problem 21, find the general solution to the following linear system.
1
0
1
0
2t1 t2 + t3
21 t1 12 t2 t3
t1
t2
t3
2
2
4
2
4
7/2
0
0
0
3
1
4
1
2
1
3
1
x1
1
x2
2
x3 =
3
x4
2
x5
11
7
18
7
, ti F
1
0
1
0
2t1 t2 + t3
21 t1 12 t2 t3
t1
t2
t3
2
2
4
2
1
7/2
0
0
0
3
1
4
1
2
1
3
1
x1
1
x2
2
x3 =
3
x4
2
x5
6
7
13
7
, ti F
25. Show that the function Tu defined by Tu (v) v proju (v) is also a linear transformation.
This is the sum of two linear transformations so it is obviously linear.
26. If u = (1, 2, 3)T and Tu
satisfies Au x = Tu (x).
1 0 0
1
0 1 0 1 2
14
0 0 1
3
17
5
7
37
3
14
37
5
14
15
Exercises
is linearly independent.
m
X
ci T vi = T 0 = 0
i=1
ker (T ) is also a subspace so it has a basis {z1 , , zr } for some r n. Now let the
basis for T (V ) be as above. Then for v an arbitrary vector, there exist unique scalars
ai such that
T v = a1 T v1 + + am T vm
Then it follows that
T (v (a1 v1 + + am vm )) = T v (a1 T v1 + + am T vm ) = 0
Hence the vector v(a1 v1 + + am vm ) is in ker (T ) and so there exist unique scalars
bi such that
r
X
v (a1 v1 + + am vm ) =
bi zi
i=1
29. In the situation of the above problem, show {v1 , , vm , z1 , , zr } is a basis for V
and therefore, dim (V ) = dim (ker (T )) + dim (T (V ))
The claim that {v1 , , vm , z1 , , zr } is a basis will be complete if it is shown that
these vectors are linearly independent. Suppose then that
X
X
ai zi +
bj vj = 0.
i
Then do T to both sides. Then you have j bj T vj = 0 and so each bj = 0. Then you
use the linear independence to conclude that each ai = 0.
16
Exercises
ker(B)
ker(A)
A(ker(BA))
If x ker (BA) , then BAx = 0 and so Ax ker (B) . That is, BAx = 0. It follows
that
A (ker (BA)) ker (B)
The second inclusion is obvious because if x is sent to 0 by A, then B will send Ax
to 0.
31. Let {x1 , , xn } be a basis of ker (A) and let {Ay1 , , Aym } be a basis of A (ker (BA)).
Let z ker (BA) . Explain why
Az span {Ay1 , , Aym }
and why there exist scalars ai such that
A (z (a1 y1 + + am ym )) = 0
and why it follows z (a1 y1 + + am ym ) span {x1 , , xn }. Now explain why
ker (BA) span {x1 , , xn , y1 , , ym }
and so
dim (ker (BA)) dim (ker (B)) + dim (ker (A)) .
This important inequality is due to Sylvester. Show that equality holds if and only if
A(ker BA) = ker(B).
Let {Ax1 , , Axr } be a basis for A (ker (BA)) . Then let y ker (BA) . Thus It
follows that
Ay A (ker (BA)) .
Then there are scalars ai such that
Ay =
r
X
ai Axi
i=1
Hence y
that
Pr
i=1
ai xi ker (A) . Let {z1 , , zm } be a basis for ker (A) . This shows
17
Exercises
Then do A to both sides and obtain each bi = 0. Then it follows that each aj is also
zero because of the independence of the zj . Thus
dim ker (BA) = dim ker (A) + dim (A ker (BA))
32. Generalize the result of the previous problem to any finite product of linear mappings.
Qs
Ps
dim (ker i=1 Ai ) i=1 dim ker (Ai ). This follows by induction.
33. If W V for W, V two subspaces of Fn and if dim (W ) = dim (V ) , show W = V .
Let a basis for W be {w1 , , wr } Then if there exists v V \ W, you could add in v
to the basis and obtain a linearly independent set of vectors of V which implies that
the dimension of V is at least r + 1 contrary to assumption.
(6.27)
cij uij = 0
and now, since these uij form a basis for Vi , it follows that each cij = 0 for each j for
each i. Thus { 1 , , m } is a basis.
18
Exercises
35. Suppose you have finitely many linear mappings L1 , L2 , , Lm which map V to V
where V is a subspace of Fn and suppose they commute. That is, Li Lj = Lj Li for all
i, j. Also suppose Lk is one to one on ker (Lj ) whenever j 6= k. Letting P denote the
product of these linear transformations, P = L1 L2 Lm , first show
ker (L1 ) + + ker (Lm ) ker (P )
Next show Lj : ker (Li ) ker (Li ) . Then show
ker (L1 ) + + ker (Lm ) = ker (L1 ) ker (Lm ) .
Using Sylvesters theorem, and the result of Problem 33, show
ker (P ) = ker (L1 ) ker (Lm )
Hint: By Sylvesters theorem and the above problem,
X
dim (ker (P ))
dim (ker (Li ))
i
vi = 0, then apply
Y
i6=k
Li vk = 0
i6=k
Since each of these Li is one to one on ker (Lk ) , it follows that vk = 0. Thus
ker (L1 ) + + ker (Lm ) = ker (L1 ) ker (Lm )
P
Now it follows that a basis for ker (L1 ) + + ker (Lm ) has i dim ker (Li ) vectors in
it.
X
dim ker (P )
dim ker (Li )
i
dim ker (P )
19
Exercises
36. Let M (Fn , Fn ) denote the set of all n n matrices having entries in F. With the usual
operations of matrix addition and scalar multiplications, explain why M (Fn , Fn ) can
2
be considered as Fn . Give a basis for M (Fn , Fn ) . If A M (Fn , Fn ) , explain why
there exists a monic polynomial of the form
k + ak k + + a1 + a0
such that
Ak + ak Ak + + a1 A + a0 I = 0
The minimal polynomial of A is the polynomial like the above, for which p (A) = 0
which has smallest degree. I will discuss the uniqueness of this polynomial later. Hint:
2
Consider the matrices I, A, A2 , , An . There are n2 + 1 of these matrices. Can they
be linearly independent? Now consider all polynomials and pick one of smallest degree
and then divide by the leading coefficient.
A basis for M (Fn , Fn ) is obviously the matricies Eij where Eij has a 1 in the ij th
place and zeros everywhere else. Thus the dimension of this vector space is n2 . It
2
follows that the list of matrices I, A, A2 , , An is linearly dependent. Hence there
exists a polynomial p () which has smallest possible degree such that p (A) = 0 by the
well ordering principle of the natural numbers. Then divide by the leading coefficient.
If you insist that its leading coefficient be 1, (monic) then the polynomial is unique
and it is called the minimal polynomial. It is unique thanks to the division algorithm,
because if q () is another one, then
q () = p () l () + r ()
where the degree of r () is less than the degree of p () or else equals 0. If it is not
zero, then r (A) = 0 and this would be a contradiction. Hence q () = p () l () where
l () must be monic. Since q () has smallest possible degree, this monic polynomial
can only be 1. Thus q () = p ().
37. Suppose the field of scalars is C and A is an n n matrix. From the preceding
problem, and the fundamental theorem of algebra, this minimal polynomial factors
( 1 )r1 ( 2 )r2 ( k )rk
where rj is the algebraic multiplicity of j . Thus
r
rk
(A 1 I) 1 (A 2 I) 2 (A k I)
=0
Let Li = (A i I) i . Then obviously these commute since they are just polynomials
in A. Is Lk one to one on ker (Li )?
rk
(A k I)
rk
= (A i I + (i k ) I)
20
Exercises
rk
X
rk
j
r j
(A i I) (i k ) k
j
j=0
rk
X
rk
rk
= (i k ) I +
(A i I)j (i k )rk j
j
j=1
rk ri
rk ri
= (i k )
ri
I + g (A) (A i I)
where g (A) is some polynomial in A. Let vi ker (Li ) . Then suppose (A k I)rk vi =
0. Then
r r
r r
0 = (A k I) k i vi = (i k ) k i vi
and since i 6= k , this requires that
Q vi = 0. Thus Lk is one to one on ker (Li ) as
hoped. Therefore, since Cn = ker i Li , it follows from the above problems that
Cn = ker (L1 ) ker (Lk )
Note that there was nothing sacred about C all you needed for the above to hold is
that the minimal polynomial factors completely into a product of linear factors. In
other words, all the above works fine for Fn provided the minimal polynomial splits.
38. In the context of Problem 37, show there exists a nonzero vector x such that
(A j I) x = 0.
This is called an eigenvector and the j is called an eigenvalue. Hint: There must
exist a vector y such that
rj 1
(A 1 I) 1 (A 2 I) 2 (A j I)
(A k I)
rk
y = z 6= 0
r 1
r
r
r
(A j I) z = (A j I) (A 1 I) 1 (A 2 I) 2 (A j I) j (A k I) k y
= (A 1 I)r1 (A 2 I)r2 (A j I)rj (A k I)rk y = 0
39. Suppose Q (t) is an orthogonal matrix. This means Q (t) is a real n n matrix which
satisfies
T
Q (t) Q (t) = I
Suppose also the entries of Q (t) are differentiable. Show QT = QT Q QT .
This is just the product rule.
21
Exercises
40. Remember the Coriolis force was 2 vB where was a particular vector which
came from the matrix Q (t) as described above. Show that
Q (t) u =
X
j
u ej (t0 ) ej (t)
= er (t0 )
X
j
= er (t0 ) es (t)
which shows the desired result.
41. An illustration used in many beginning physics books is that of firing a rifle horizontally and dropping an identical bullet from the same height above the perfectly
flat ground followed by an assertion that the two bullets will hit the ground at exactly the same time. Is this true on the rotating earth assuming the experiment
takes place over a large perfectly flat field so the curvature of the earth is not an
issue? Explain. What other irregularities will occur? Recall the Coriolis acceleration
is 2 [(y cos ) i+ (x cos + z sin ) j (y sin ) k] where k points away from the
center of the earth, j points East, and i points South.
Obviously not. Because of the Coriolis force experienced by the fired bullet which is
not experienced by the dropped bullet, it will not be as simple as in the physics books.
For example, if the bullet is fired East, then y sin > 0 and will contribute to a force
acting on the bullet which has been fired which will cause it to hit the ground faster
than the one dropped. Of course at the North pole or the South pole, things should
be closer to what is expected in the physics books because there sin = 0. Also, if
you fire it North or South, there seems to be no extra force because y = 0.
F.18
Exercises
3.2
1. Find the determinants of the following matrices.
1 2 3
(a) 3 2 2 (The answer is 31.)
0 9 8
22
Exercises
4
(b) 1
3
1
1
(c)
4
1
3 2
7 8 (The answer is 375.)
9 3
2 3 2
3 2 3
, (The answer is 2.)
1 5 0
2 1 2
2. If A1 exist, what is the relationship between det (A) and det A1 . Explain your
answer.
1 = det AA1 = det (A) det A1 .
3. Let A be an n n matrix where n is odd. Suppose also that A is skew symmetric.
This means AT = A. Show that det(A) = 0.
det (A) = det AT = det (A) = det (I) det (A) = (1)n det (A) = det (A) .
4. Is it true that det (A + B) = det (A) + det (B)? If this is so, explain why it is so and
if it is not so, give a counter example.
Almost anything shows that this is not true.
1 0
1 0
det
+
=
0 1
0 1
1 0
1 0
det
+ det
=
0 1
0 1
0
2
5. Let A be an r r matrix and suppose there are r 1 rows (columns) such that all rows
(columns) are linear combinations of these r 1 rows (columns). Show det (A) = 0.
Without loss of generality, assume the last row is a linear combination of the first r 1
rows. Then the matrix is of the form
rT1
..
rTn1
Pn1
T
a
r
i=1 i i
Then from the linear property of determinants, the determinant equals
T
T
r1
r1
n1
.. n1
..
X
X
ai det . =
ai det . = 0
T
T
rn1
rn1
i=1
i=1
T
ri
0T
Where the first equal sign in the above is obtained by taking 1 times a the ith row
from the top and adding to the last row.
6. Show det (aA) = an det (A) where here A is an n n matrix and a is a scalar.
Each time you take out an a from a row, you multiply by a the determinant of the
matrix which remains. Since there are n rows, you do this n times, hence you get an .
23
Exercises
7. Suppose A is an upper triangular matrix. Show that A1 exists if and only if all
elements of the main diagonal are non zero. Is it true that A1 will also be upper
triangular? Explain. Is everything the same for lower triangular matrices?
This is obvious because the determinant of A is the product of these diagonal entries.
When you consider the usual process of finding the inverse, you get that A1 must be
upper triangular. Everything is similar for lower triangular matrices.
8. Let A and B be two n n matrices. A B (A is similar to B) means there exists an
invertible matrix S such that A = S 1 BS. Show that if A B, then B A. Show
also that A A and that if A B and B C, then A C.
This is easy except possibly for the last claim. Say A = P 1 BP and B = Q1 CQ.
Then
A = P 1 BP = A = P 1 Q1 CQP = (QP )1 C (QP )
9. In the context of Problem 8 show that if A B, then det (A) = det (B) .
det A =
=
det P 1 BP = det P 1 det (B) det (P )
det (B) det P 1 P = det (B) .
10. Let A be an n n matrix and let x be a nonzero vector such that Ax = x for some
scalar, . When this occurs, the vector, x is called an eigenvector and the scalar,
is called an eigenvalue. It turns out that not every number is an eigenvalue. Only
certain ones are. Why? Hint: Show that if Ax = x, then (I A) x = 0. Explain
why this shows that (I A) is not one to one and not onto. Now use Theorem 3.1.15
to argue det (I A) = 0. What sort of equation is this? How many solutions does it
have?
1
If that determinant equals 0 then the matrix I A has no inverse. It is not one
to one and so there exists x 6= 0 such that (I A) x = 0. Also recall the process for
finding the inverse.
a (t) b (t)
12. Let F (t) = det
. Verify
c (t) d (t)
F (t) = det
Now suppose
a (t) b (t)
c (t) d (t)
+ det
a (t) b (t)
c (t) d (t)
24
Exercises
Use Laplace expansion and the first part to verify F (t) =
Conjecture a general result valid for n n matrices and explain why it will be true.
Can a similar thing be done with the columns?
The way to see this holds in general is to use the usual proof for the product rule and
the theorem about the determinant and row operations.
a (t + h) b (t + h) c (t + h)
F (t + h) F (t) = det d (t + h) e (t + h) f (t + h)
g (t + h) h (t + h) i (t + h)
a (t + h) b (t + h) c (t + h)
a (t)
b (t)
c (t)
det d (t + h) e (t + h) f (t + h) det d (t + h) e (t + h) f (t + h)
g (t + h) h (t + h) i (t + h)
g (t + h) h (t + h) i (t + h)
a (t)
b (t)
c (t)
a (t)
b (t)
c (t)
e (t)
f (t)
+ det d (t + h) e (t + h) f (t + h) det d (t)
g (t + h) h (t + h) i (t + h)
g (t + h) h (t + h) i (t + h)
a (t)
b (t)
c (t)
a (t) b (t) c (t)
e (t)
f (t) det d (t) e (t) f (t)
+ det d (t)
g (t + h) h (t + h) i (t + h)
g (t) h (t) i (t)
Now multiply by 1/h to obtain the following for the difference quotient
det
a(t+h)a(t)
h
d (t + h)
g (t + h)
b(t+h)b(t)
h
e (t + h)
h (t + h)
+ det
c(t+h)c(t)
h
f (t + h) +det
i (t + h)
F (t+h)F (t)
.
h
a (t)
b (t)
c (t)
d(t+h)d(t)
h
e(t+h)e(t)
h
f (t+h)f (t)
h
g (t + h)
a (t)
d (t)
b (t)
e (t)
c (t)
f (t)
g(t+h)g(t)
h
h(t+h)h(t)
h
i(t+h)i(t)
h
h (t + h)
i (t + h)
Now passing to a limit yields the desired formula. Obviously this holds for any size
determinant.
13. Use the formula for the inverse in terms of the cofactor matrix to find the inverse of
the matrix
t
e
0
0
.
et cos t
et sin t
A= 0
t
t
t
0 e cos t e sin t e cos t + et sin t
25
Exercises
et
0
0
1
0
et sin t
t
t
e cos t + e sin t
0
0
et (cos t + sin t) (sin t) et
et (cos t sin t) (cos t) et
0
et cos t
et cos t et sin t
et
0
=
0
16. Suppose Q (t) is an orthogonal matrix. This means Q (t) is a real n n matrix which
satisfies
Q (t) Q (t)T = I
Suppose Q (t) is continuous for t [a, b] , some interval. Also suppose det (Q (t)) = 1.
Show that it follows det (Q (t)) = 1 for all t [a, b].
You have from the given equation that det (Q (t)) is always either 1 or 1. Since Q (t)
is continuous, so is t det (Q (t)) and so if it starts off at 1, it cannot jump to 1
because this would violate the intermediate value theorem.
26
Exercises
F.19
Exercises
3.6
1. Let m < n and let A be an m n matrix. Show that A is not one to one. Hint:
Consider the n n matrix A1 which is of the form
A
A1
0
where the 0 denotes an (n m) n matrix of zeros. Thus det A1 = 0 and so A1 is
not one to one. Now observe that A1 x is the vector,
Ax
A1 x =
0
which equals zero if and only if Ax = 0.
The hint gives it away. You could simply consider a vector of the form
0
a
where a 6= 0.
2. Let v1 , , vn be vectors in Fn and let M (v1 , , vn ) denote the matrix whose ith
column equals vi . Define
d (v1 , , vn ) det (M (v1 , , vn )) .
Prove that d is linear in each variable, (multilinear), that
d (v1 , , vi , , vj , , vn ) = d (v1 , , vj , , vi , , vn ) ,
(6.28)
and
d (e1 , , en ) = 1
(6.29)
where here ej is the vector in Fn which has a zero in every position except the j th
position in which it has a one.
This follows from the properties of determinants which are discussed above.
3. Suppose f : Fn Fn F satisfies 6.28 and 6.29 and is linear in each variable.
Show that f = d.
Consider f (x1 , , xn ) . Then by the assumptions on f it equals
X
f (x1 , , xn ) =
xi1 xin f (ei1 ein )
i1 , ,in
i1 , ,in
i1 , ,in
i1 , ,in
27
Exercises
4. Show that if you replace a row (column) of an n n matrix A with itself added to
some multiple of another row (column) then the new matrix has the same determinant
as the original one.
This was done above.
5. Use the result of Problem 4 to evaluate by hand
1 2 3
6 3 2
det
5 2 2
3 4 6
1
6
det
5
3
2
3
2
4
3
2
2
6
2
3
=5
3
4
et
et
et
cos t
sin t
cos t
1
2
1
2
the determinant
2
3
.
3
4
1
sin t
cos t
sin t
1 t
2e
cos t + 12 sin t
sin t 12 cos t
0
sin t
cos t
sin t
cos t .
sin t
1 t
2e
1
1
2 sin t 2 cos t
1
2 cos t 12 sin t
7. Let Ly = y (n) + an1 (x) y (n1) + + a1 (x) y + a0 (x) y where the ai are given
continuous functions defined on a closed interval, (a, b) and y is some function which
has n derivatives so it makes sense to write Ly. Suppose Lyk = 0 for k = 1, 2, , n.
The Wronskian of these functions, yi is defined as
y1 (x)
yn (x)
y1 (x)
yn (x)
.
.
(n1)
y1
(x)
(n1)
yn
(x)
W (x) = det
y1 (x)
y1 (x)
..
.
(n)
y1 (x)
yn (x)
yn (x)
..
.
(n)
yn (x)
Now use the differential equation, Ly = 0 which is satisfied by each of these functions,
yi and properties of determinants presented above to verify that W + an1 (x) W = 0.
Give an explicit solution of this linear differential equation, Abels formula, and use
28
Exercises
your answer to verify that the Wronskian of these solutions to the equation, Ly = 0
either vanishes identically on (a, b) or never.
The last formula above follows because W (x) equals the sum of determinants of
matrices which have two equal rows except for the last one in the sum which is the
displayed expression. Now let
(n1)
mi (x) = an1 (x) yi
+ + a1 (x) yi + a0 (x) yi
(n)
Since each yi is a solution to Ly = 0, it follows that yi (t) = mi (t). Now from the
properties of determinants, being linear in each row,
y1 (x)
yn (x)
y1 (x)
yn (x)
.
.
(n1)
(n1)
y1
(x) yn
(x)
=
d A(x)
e
W (x) = 0
dx
and so W (x) = CeA(x) . Thus the Wronskian either vanishes for all x or for no x.
8. Two n n matrices, A and B, are similar if B = S 1 AS for some invertible n n
matrix S. Show that if two matrices are similar, they have the same characteristic
polynomials. The characteristic polynomial of A is det (I A) .
Say A = S 1 BS. Then
det (I A)
det I S 1 BS
det S 1 S S 1 BS
det S 1 (I B) S
det S 1 det (I B) det (S)
det S 1 S det (I B) = det (I B)
=
=
=
=
An + an1 An1 + + a1 A + a0 I = 0
Also the characteristic polynomial is
det (tI A)
and the constant term is (1) det (A) . Thus a0 6= 0 if and only if det (A) 6= 0 if and
only if A1 has an inverse. Thus if A1 exists, it follows that
a0 I = An + an1 An1 + + a1 A = A An1 an1 An2 a1 I
29
Exercises
and also
a0 I = An1 an1 An2 a1 I A
1
An1 an1 An2 a1 I
a0
p (t) det (tI A) . Hence p (t ) = det ((t ) I A) = det (tI (I + A)) and
by definition, this is the characteristic polynomial of I + A. Therefore, the constant
term of this is the constant term of p (t ). Say
p (t) = tn + an1 tn1 + + a1 t + a0
n1
p (t) = p (t ) = (t ) + an1 (t )
k
+ + ak (t )
then the constant term of this p (t) is ak (1) k + terms having raised to a higher
power. Hence for all small enough, this term will dominate the sum of all the others
and it follows that the constant term is nonzero for all || small enough.
11. In constitutive
modeling of the stress and strain tensors, one sometimes considers sums
P
of the form k=0 ak Ak where A is a 33 matrix. Show using the Cayley Hamilton
theorem that if such a thing makes any sense, you can always obtain it as a finite sum
having no more than n terms.
Say the characteristic polynomial is q (t) . Then if n 3,
tn = q (t) l (t) + r (t)
where the degree of r (t) is either less than 3 or it equals zero. Thus
An = q (A) l (A) + r (A) = r (A)
and so all the terms An for n 3 can be replaced with some r (A) where the degree of
r (t) is no more than 2. Thus, assuming there are no convergence issues, the infinite
P2
sum must be of the form k=0 bk Ak .
12. Recall you can find the determinant from expanding along the j th column.
X
det (A) =
Aij (cof (A))ij
i
Think of det (A) as a function of the entries, Aij . Explain why the ij th cofactor is
really just
det (A)
.
Aij
This follows from the formula for determinant. If you take the partial derivative, you
get (cof (A))ij .
30
Exercises
13. Let U be an open set in Rn and let g :U Rn be such that all the first partial
derivatives of all components of g exist and are continuous. Under these conditions
form the matrix Dg (x) given by
Dg (x)ij
gi (x)
gi,j (x)
xj
The best kept secret in calculus courses is that the linear transformation determined
by this matrix Dg (x) is called the derivative of g and is the correct generalization
of the concept of derivative of a function of one variable. Suppose the second partial
derivatives also exist and are continuous. Then show that
X
(cof (Dg))ij,j = 0.
j
Next differentiate with respect to xj and sum on j using the equality of mixed partial
derivatives. Assume det (Dg) 6= 0 to prove the identity in this special case. Then
explain why there exists a sequence k 0 such that for gk (x) g (x) + k x,
det (Dgk ) 6= 0 and so the identity holds for gk . Then take a limit to get the desired result in general. This is an extremely important identity which has surprising
implications.
First assume det (Dg) 6= 0. Then from the cofactor expansion, you get
X
gi,k cof (Dg)ij = jk det (Dg)
i
If j = k you get det (Dg) on the right but if j 6= k, then the left is the expansion of a
determinant which has two equal columns. Now differentiate both sides with respect
to j using the above problem. You have to use the chain rule.
X
X
gi,kj cof (Dg)ij +
gi,k cof (Dg)ij,j
i
= jk
X (det (Dg))
r,s
gr,s
gr,sj =
r,s
Next sum on j
X
X
X
gi,kj cof (Dg)ij +
gi,k cof (Dg)ij,j =
jk cof (Dg)rs gr,sj
i,j
i,j
r,s,j
r,s
i,j
i,j
Subtract the terms at the end from each side using equality of mixed partial derivatives.
This gives.
X
X
X
0=
gi,k cof (Dg)
=
gi,k
cof (Dg)
ij,j
i,j
ij,j
31
Exercises
Since Dg is assumed invertible, this requires
X
cof (Dg)ij,j = 0
j
In case det (Dg (x)) = 0 for some x, consider k , a sequence of numbers converging to
0 with the property that det (Dg (x) + k I) 6= 0. Then let gk (x) = g (x) + k x. From
the above, you have
X
X
cof (Dg)ij,j (x) = lim
cof (Dgk )ij,j (x) = 0
k
1
a1
a21
..
.
1
an
a2n
..
.
an1
1
an1
an1
n
ann
By this is meant to take the product of all terms of the form (aj ai ) such that j > i.
Hint: Show it works if n = 1 so you are looking at
1 1
a0 a1
Then suppose it holds for n 1 and consider the
polynomial.
1
1
a0
a
1
2
a20
a
1
p (t) .
..
.
.
.
n1
n1
a
a
1
0
n
an
a
0
1
Explain why p (aj ) = 0 for i = 0, , n 1. Thus
p (t) = c
n1
Y
i=0
tn1
tn
(t ai ) .
Of course c is the coefficient of tn . Find this coefficient from the above description of
p (t) and the induction hypothesis. Then plug in t = an and observe you have the
formula valid for n.
1 1
= (a1 a0 ) so the formula holds. Now
In the case of n = 1, you have
a0 a1
suppose it holds for n and consider p (t) as above. Then it is obvious that p (aj ) = 0
32
Exercises
for i = 0, , n 1 because you have the determinant of a matrix with two equal
columns. Thus
n1
Y
p (t) = c
(t ai ) .
i=0
because p (t) has degree n and so it has no more than n roots. So what is c? The
n
coefficient
it
Q of t is the Vandermonde determinant which is n n and by induction,
equals 0i<jn1 (aj ai ) . From the above product, the coefficient of tn is c. Thus
p (t) =
0i<jn1
(aj ai )
n1
Y
i=0
(t ai )
p (an ) =
0i<jn1
0i<jn
F.20
(aj ai )
n1
Y
i=0
(an ai )
(aj ai )
Exercises
4.6
1. Let {u1 , , un } be vectors in Rn . The parallelepiped determined by these vectors
P (u1 , , un ) is defined as
( n
)
X
P (u1 , , un )
tk uk : tk [0, 1] for all k .
k=1
k=1
tk Auk : tk [0, 1]
33
Exercises
In other words, draw the result of doing E to the vectors in P (e1 , e2 ). Next draw the
results of doing the other elementary matrices to P (e1 , e2 ).
For E the given elementary matrix, the result is as follows.
P (e1 , e2 )
E(P (e1 , e2 ))
1
0
0
0
1
0
0 0
1 0 ,
0 1
1 0
0 0 ,
0 1
0 0
1 0
0 1
1 0
0 0
0 1
1
0 ,
0
0
1 ,
0
0
0
1
0
0
1
1 0
0 1
0 0
0 1
1 0
0 0
So what is the dimension of the span of these? One way to systematically accomplish
this is to unravel them and then use the row reduced echelon form. Unraveling these
yields the column vectors
34
Exercises
1
0
0
0
1
0
0
0
1
0
0
1
1
0
0
0
1
0
0
1
0
0
0
1
1
0
0
0
1
0
1
0
0
0
0
1
1
0
0
0
0
1
0
1
0
0
0
1
0
1
0
1
0
0
1 0 0 0 1 0
1
0 0 1 1 0 0
0
0 1 0 0 0 1
0
0 1 0 1 0 0
0
0 0 1 0 1 0
0
0 0 1 0 0 1
0
0 1 0 0 1 0
0
1 0 0 1 0 0
0
0
0
1
0
0
0
0
0
0
0
0
0
1
0
0
0
0
0
The dimension is 5.
0 1
0 1
0 1
0 1
1 1
0 0
0 0
0 0
0 0
1 0 0 0
(b) 0 0 1 2
0 0 0 0
This one is.
1 1 0 0 0 5
(c) 0 0 1 2 0 4
0 0 0 0 1 3
This one is.
7. Row reduce the following matrices to obtain the row reduced echelon form. List the
pivot columns in the original matrix.
1 2 0 3
1 0 0 3
(a) 2 1 2 2 , row echelon form: 0 1 0 0
1 1 0 3
0 0 1 2
1 2 3
1 0 0
2 1 2
0 1 0
(b)
3 0 0 , row echelon form: 0 0 1
3 2 1
0 0 0
1 0 0 0
1 2 1 3
(c) 3 2 1 0 , row echelon form: 0 1 12 0
3 2 1 1
0 0 0 1
Exercises
35
8. Find the rank and nullity of the following matrices. If the rank is r, identify r columns
in the original matrix which have the property that every other column may be
written as a linear combination of these.
0 1 0 2 1 2 2
0 1 0 2 0 1 1
0 3 2 12 1 6 8
0 0 1 3 0 1 2
(a)
0 1 1 5 0 2 3 , row echelon form: 0 0 0 0 1 1 1 Rank
0 2 1 7 0 3 4
0 0 0 0 0 0 0
equals 3 nullity equals 4. A basis is columns 2,3,5
0 1 0 2 0 1 0
0 1 0 2 0 1 0
0 3 2 6 0 5 4
0 0 1 0 0 1 2
(b)
0 1 1 2 0 2 2 , row echelon form: 0 0 0 0 0 0 0 Rank
0 2 1 4 0 3 2
0 0 0 0 0 0 0
is 2 nullity equals 5. A basis is columns 2,3.
0 1 0 2 1 1 2
0 1 0 2 0 1 0
0 3 2 6 1 5 1
0 0 1 0 0 1 0
(c)
0 1 1 2 0 2 1 , row echelon form: 0 0 0 0 1 0 0 Rank
0 2 1 4 0 3 1
0 0 0 0 0 0 1
is 4 and nullity is 3. A basis is columns 2,3,5,7.
9. Find the rank of the following matrices. If the rank is r, identify r columns in the
original matrix which have the property that every other column may be written
as a linear combination of these. Also find a basis for the row and column spaces of
the matrices.
1 2 0
1 0 0
3 2 1
0 1 0
(a)
2 1 0 , row echelon form: 0 0 1 . The rank is 3.
0 2 1
0 0 0
1 0 0
1 0 0
4 1 1
0 1 0
(b)
2 1 0 , row echelon form: 0 0 1 . The rank is 3
0 2 0
0 0 0
0 1 0 2 1 2 2
0 1 0 2 0 1 1
0 3 2 12 1 6 8
0 0 1 3 0 1 2
(c)
0 1 1 5 0 2 3 , row echelon form: 0 0 0 0 1 1 1 .
0 2 1 7 0 3 4
0 0 0 0 0 0 0
The rank is 3.
0 1 0 2 0 1 0
0 1 0 2 0 1 0
0 3 2 6 0 5 4
0 0 1 0 0 1 2
(d)
0 1 1 2 0 2 2 , row echelon form: 0 0 0 0 0 0 0 . The
0 2 1 4 0 3 2
0 0 0 0 0 0 0
rank is 2.
0 1 0 2 1 1 2
0 1 0 2 0 1 0
0 3 2 6 1 5 1
0 0 1 0 0 1 0
(e)
0 1 1 2 0 2 1 , row echelon form: 0 0 0 0 1 0 0 . The
0 2 1 4 0 3 1
0 0 0 0 0 0 1
rank is 4.
10. Suppose A is an m n matrix. Explain why the rank of A is always no larger than
min (m, n) .
36
Exercises
It is because you cannot have more than min (m, n) nonzero rows in the row reduced
echelon form. Recall that the number of pivot columns is the same as the number of
nonzero rows from the description of this row reduced echelon form.
11. Suppose A is an m n matrix in which m n. Suppose also that the rank of A equals
m. Show that A maps Fn onto Fm . Hint: The vectors e1 , , em occur as columns
in the row reduced echelon form for A.
It follows from the fact that e1 , , em occur as columns in row reduced echelon form
that the dimension of the column space of A is n and so, since this column space is
A (Rn ) , it follows that it equals Fm .
12. Suppose A is an m n matrix and that m > n. Show there exists b Fm such that
there is no solution to the equation
Ax = b.
Since m > n the dimension of the column space of A is no more than n and so the
columns of A cannot span Fm .
13. Suppose A is an m n matrix in which m n. Suppose also that the rank of A
equals n. Show that A is one to one. Hint: If not, there exists a vector, x 6= 0 such
that Ax = 0, and this implies at least one column of A is a linear combination of the
others. Show this would require the column rank to be less than n.
The hint gives it away. The matrix has at most n independent columns. If one is a
combination of the others, then its rank is less than n.
14. Explain why an n n matrix A is both one to one and onto if and only if its rank is
n.
If A is one to one, then its columns are linearly independent, hence a basis and so it
is also onto. Since the columns are independent, its rank is n. If its rank is n this just
says that the columns are independent.
15. Suppose A is an m n matrix and {w1 , , wk } is a linearly independent set of
vectors in A (Fn ) Fm . Suppose also that Azi wi . Show that {z1 , , zk } is also
linearly independent.
P
P
If i ci zi = 0, apply A to both sides to obtain i ci wi = 0. By assumption, each
ci = 0.
16. Let A, B be m n matrices. Show that rank (A + B) rank (A) + rank (B).
This is obvious from the following.
17. Suppose A is an m n matrix, m n and the columns of A are independent. Suppose also that {z1 , , zk } is a linearly independent set of vectors in Fn . Show that
{Az1 , , Azk } is linearly independent.
37
Exercises
then A
P
n
j=1 cj zj
= 0 and so
Pn
j=1 cj zj
= 0 therefore each cj = 0.
k
X
ai zi ker (B) .
i=1
Let {w1 , , wk } be a basis for B (Fp ) ker (A) and suppose {u1 , , ur } is a basis
for ker (B). Let Bzi = wi . Now suppose x ker (AB). Then Bx ker (A) B (Fp )
and so
k
k
X
X
Bx =
a i wi =
ai Bzi
i=1
so
i=1
k
X
i=1
Hence
x
k
X
ai zi ker (B)
ai zi =
i=1
r
X
b j uj
j=1
j=1
38
Exercises
19. Let m < n and let A be an m n matrix. Show that A is not one to one.
There are more columns than rows and at most m can be pivot columns so it follows
at least one column is a linear combination of the others hence A is not one too one.
20. Let A be an m n real matrix and let b Rm . Show there exists a solution, x to the
system
AT Ax = AT b
Next show that if x, x1 are two solutions, then Ax = Ax1 . Hint: First show that
T
AT A = AT A. Next show if x ker AT A , then Ax = 0. Finally apply the Fredholm alternative. Show AT b ker(AT A) . This will give existence of a solution.
T
T
AT A = AT AT
= AT A. Also if AT Ax = 0, then
AT Ax, x = |Ax|2 = 0
and so Ax = 0. Is AT b ker AT A ? Say AT Ax = 0. Then Ax = 0. Then
AT b, x = (b,Ax) = (b, 0) = 0
=
=
=
=
|bAx+AxAy|
1
0
0 1
T
4
22. Here is a point in R : (1, 2, 3, 4) . Find the point in span
2 , 3
3
2
is closest to the given point.
This just says to find the least squares solution to the
1 0
1
0 1 x
2
=
2 3 y
3
3 2
4
1
0
2
3
T
0
1
0
1
3 2
2
3
0
1
0
1
x
=
2
3 y
2
3
T
0
1
2
1
3 3
2
4
which
equations
39
Exercises
14 12
12 14
x
y
19
19
, Solution is:
19
0 + 19
2
26
26
3
19
26
19
26
1
=
3
2
19
26
19
26
95
26
95
26
23. Here is a point in R4 : (1, 2, 3, 4) . Find the point on the plane described by x + 2y
4z + 4w = 0 which is closest to the given point.
2y + 4z 4w
y
where y, z, w are arbitary. Thus it is
The plane is of the form
z
w
the span of the vectors
2
4
4
1 0 0
0 , 1 , 0
0
0
1
and so you are looking for a least squares solution to the system
2 4 4
1
a
1 0 0
2 4 4
2 4 4
2 4 4
1
a
1 0 0 1 0 0
1 0 0 2
b =
0 1 0 0 1 0
0 1 0 3
c
0 0 1
0 0 1
0 0 1
4
56
5
8
8
a
0
37
8 17 16 b = 7 , Solution is: 147 The closest point is
37
112
8 16 17
c
0
37
then
28
2
4
4
37
147 0 112 0 56
56 1
37
37 0 + 37 1 + 37 0 = 147
37
112
0
0
1
37
28
Check:
2 4 4
37
56
37
147
37
112
37
=0
24. Suppose A, B are two invertible n n matrices. Show there exists a sequence of row
operations which when done to A yield B. Hint: Recall that every invertible matrix
is a product of elementary matrices.
Both A, B have row reduced echelon form equal to I. Therefore, they are row equivalent.
E1 Ep A = I, F1 Fr B = I
40
Exercises
and so
E1 Ep A = F1 Fr B
Now multiply on both sides by the inverses of the Fi to obtain a product of elementary
matrices which multiplied by A give B.
25. If A is invertible and n n and B is n p, show that AB has the same null space as
B and also the same rank as B.
From the above problem, AB and B have exactly the same row reduced echelon form.
Thus the result follows.
26. Here are two matrices in row reduced echelon form
1 0 1
1
A = 0 1 1 , B = 0
0
0 0 0
0 0
1 1
0 0
Does there exist a sequence of row operations which when done to A will yield B?
Explain.
No. The row reduced echelon form is unique.
27. Is it true that an upper triagular matrix has rank equal to the number of nonzero
entries down the main diagonal?
1 0 2 0
0 1 1 7
No.
0 0 0 1
0 0 0 0
28. Let {v1 , , vn1 } be vectors in Fn . Describe a systematic way to obtain a vector vn
which is perpendicular to each of these vectors. Hint: You might consider something
like this
e1
e2
en
v11
v12
v1n
det
..
..
..
.
.
.
v(n1)1
th
v(n1)2
v(n1)n
This is a good hint. You formally expand along the top row. It will work. If you
replace ei with vki so that the top row becomes.
vk1 vk2 vkn
then the determinant will be equal to 0 because it has two equal rows. However, this
determinant is also what you get when you take the dot product of the above formal
determinant having the ei on the top row with vk .
29. Let A be an m n matrix. Then ker (A) is a subspace of Fn . Is it true that every
subspace of Fn is the kernel or null space of some matrix? Prove or disprove.
Let M be a subspace of Fn . If it equals {0} , consider the matrix I. Otherwise, it has
a basis {m1 , , mk } . Consider the matrix
m1 mk 0
41
Exercises
30. Let A be an nn matrix and let P ij be the permutation matrix which switches the ith
and j th rows of the identity. Show that P ij AP ij produces a matrix which is similar
to A which switches the ith and j th entries on the main diagonal.
This is easy to see when you consider that P ij is its own inverse and that P ij multiplied
on the right switches the ith and j th columns. Thus you switch the columns and then
you switch the rows. This has the effect of switching Aii and Ajj . For example,
a b c d
1 0 0 0
a d c b
1 0 0 0
0 0 0 1 e f z h 0 0 0 1 n g h t
0 0 1 0 j k l m 0 0 1 0 = j m l k
e h z f
0 1 0 0
n t h g
0 1 0 0
More formally, the iith entry of P ij AP ij is
X ij
ij
Pis Asp Ppi
= Pijij Ajj Pjiij = Aij
s,p
31. Recall the procedure for finding the inverse of a matrix on Page 48. It was shown that
the procedure, when it works, finds the inverse of the matrix. Show that whenever
the matrix has an inverse, the procedure works.
If A has an inverse, then it is one to one. Hence the columns are independent. Therefore, they are each pivot columns. Therefore, the row reduced echelon form of A is I.
This is what was needed for the procedure to work.
F.21
Exercises
5.8
1.
2.
3.
4.
5.
1 2 0
1 0 0
1 2
Find a LU factorization of 2 1 3 . = 2 1 0 0 3
1 2 3
1 0 1
0 0
1 2 3 2
1
0
0
1
1
0 0
Find a LU factorization of 1 3 2 1 . = 1
5 0 1 3
5 10 1
0
1 2 1
1 0 0
1 0
Find a P LU factorization of 1 2 2 . = 0 0 1 2 1
2 1 1
0 1 0
1 0
1 2 1 2 1
Find a P LU factorization of 2 4 2 4 1 .
1 2 1 3 2
1 0 0
1 2 1 2 1
1 0 0
= 0 1 0 2 1 0 0 0 0 0 1
0 0 1
1 0 1
0 0 0 1 1
1 2 1
1 2 2
Find a P LU factorization of
2 4 1 .
3 2 1
1 0 0 0
1 0 0 0
1 2
1
0 0 0 1 3 1 0 0 0 4 2
=
0 0 1 0 2 0 1 0 0 0 1
0 1 0 0
1 0 1 1
0 0
0
0
3
3
3
2
1 1
24 17
0
1 2
1
0 0 3 1
1
0 0
1
2
1
0
42
Exercises
6. Is there only one LU factorization for a given matrix? Hint: Consider the equation
0 1
1 0
0 1
=
.
0 1
1 1
0 0
Apparently, LU factorization
0 1
1 0
0
=
0 1
1 1
0
0 1
1 0
0
=
0 1
0 1
0
7. Here is a matrix and
1
A= 1
0
is not unique.
1
0
1
1
an LU factorization
2 5 0
1
1 4 9 = 1
1 2 5
0
of it.
0 0
1
1 0 0
1 1
0
2
5
0
1 1 9
0
1 14
1 0 0
a
1
1
1 1 0 b = 2 , Solution is: 1
0 1 1
c
3
4
24t4 9
x
1 2
5
0
1
y
23t4 5
1 , Solution is:
0 1 1 9
z =
4 14t4
0 0
1 14
4
w
t4
1 2 1
3 2 1
1 0 2
1 2 1
3 2 1
1 0 2
1
11 11
3
11 11
1
11 11
11
0
0
1 2 1
3 0 1
1 0 2
, t4 R
13
2 11 61 2
66
5
66 2 11 61 2
1
2
33 2 11
3 2
4
6
11
11 11
11
6
2
2 11
11 2 11
11
0
1
1
43
Exercises
1 2 1 0
3 0 1 1
1 0 2 1
1
1
10
11
0
11 11
11
3
3
1
= 11
11 110
1011 310 2 5
1
1
11 110 10 11 10 2 5
11
2
6
4
11
11
11
11
11
11
11
2
1
2
0
11 10 11
22 10 11 55 10
11
1
1
0
0
2
5
2
5
2
5
10. If you had a QR factorization, A = QR, describe how you could use it to solve the
equation Ax = b. Note, this is not how people usually solve systems of equations.
You could first solve Qy = b to obtain y = QT b. Then you would solve Rx = y which
would be easy because R is upper triangular.
11. If Q is an orthogonal matrix, show the columns are an orthonormal set. That is show
that for
Q = q1 qn
it follows that qi qj = ij . Also show that any orthonormal set of vectors is linearly
independent.
Pn
Say i=1 ci qi = 0. Then take the inner product of both sides with qk to obtain
X
ci ij = 0
0 0 0
0 0 1
0 0 1
and verify this equals
and also
1 2
2
1
2 2
1
0
0 0
1
0
2 0 0
2
0 0
0 12 2
1 0
0 1
0 0
0
0
0 0
1
0
0 0
0 1 .
0 1
0
1
0
0 0
0 0
2
1
1 2 0
0 0 0 = 0 0
2
2
2
1
1
0 0
0 0 0
2 2 0 2 2
1 0 0
0 0 0
0 0 0
0 1 0 0 0 1 = 0 0 1
0 0 1
0 0 1
0 0 1
2
0
0
0
1
1
44
Exercises
Now if A1 exists and A = QR = Q R where R, R have positive entries down
1
the main diagonal, then you get R (R )
= QT Q U, where U is an orthogonal
matrix. Thus you have an upper triangular matrix which has positive entries down
the diagonal equal to an orthogonal matrix. Obviously, this requires that the upper
triangular matrix is the identity. Hence R = R and Q = Q . Why is this so obvious?
Consider the following case of a 33 which is orthogonal.
a b c
0 d e
0 0 f
Then this matrix
a b
0 d
0 0
a 0
b d
c e
2
c
a 0 0
a + b2 + c2 ce + bd cf
e
b d 0
ce + bd
d2 + e2 f e
=
f
c e f
cf
fe
f2
2
0
a b c
a
ab
ac
0 0 d e = ab b2 + d2
de + bc
f
0 0 f
ac de + bc c2 + f 2 + e2
Thus both of these on the right end are the identity and so b = c = 0, a = 1 since
the diagonal entries are all positive. Then also d = 1 and c, e = 0 and f = 1. Now it
follows that all the off diagonal entries must equal 0 since otherwise the magnitude of
a column would not equal 1.
a1
an
q1
qn
r11
0
..
.
r12
r22
..
.
r1n
r2n
rnn
45
Exercises
If you have a subspace, let {a1 , , ak } be a basis and then extend to a basis of
Rn , {a1 , , an }. Then form the above QR factorization and you will have that
{q1 , , qk } is an orthonormal basis for the subspace.
14. Suppose Qn Rn converges to an orthogonal matrix Q where Qn is orthogonal and Rn
is upper triangular having all positive entries on the diagonal. Show that then Qn
converges to Q and Rn converges to the identity.
Let
Q = (q1 , , qn ) , Qk = qk1 , , qkn
k
where the q are the columns. Also denote by rij
the ij th entry of Rk . Thus
It follows
Qk Rk = qk1 , , qkn
k
r11
..
k
rnn
k k
r11
q1 q1
and so
k k
k
r11
= r11
q1 1
Therefore,
qk1 q1 .
Therefore,
k k
lim r22
q2 = q2
k
k
and since r22
> 0, it follows, as in the first part that r22
1. Hence
lim qk2 = q2 .
k
lim rjj
= 1, lim qkj = qj .
Thus Rk I and Qk Q.
46
Exercises
F.22
Exercises
6.6
1. Maximize and minimize z = x1 2x2 + x3 subject to the constraints x1 + x2 + x3
10, x1 + x2 + x3 2, and x1 + 2x2 + x3 7 if possible. All variables are nonnegative.
The constraints lead to the augmented
1 1 1
1 1 1
1 2 1
The obvious solution is not feasible.
1 1
0 0
0 1
matrix
1
0
0
0 0 10
1 0 2
0 1 7
Do a row operation.
1 1 0 0 10
0 1 1 0 8
0 1 0 1 3
An obvious solution is still not feasible. Do another operation couple of row operations.
1 1 1 0 1 0 2
0 0 0 1 1 0 8
0 1 0 0 1 1 5
1
0
0
1
1 1 0 1 0 0 2
0 0 1 1 0 0 8
1 0 0 1 1 0 5
2 1 0 0 0 1 0
1 1
0 0
0 1
0 2
1
0
0
0
0
1
0
0
1
1
1
1
0
0
1
0
0
0
0
1
2
8
5
0
First lets work on minimizing this. There is a +2. The ratios are then 5, 2 so the pivot
is the 1 on the top of the second column. The next tableau is
1 1 1 0 1 0 0 2
0 0 0 1 1 0 0 8
1 0 1 0 2 1 0 3
2 0 2 0 1 0 1 4
There is a 1 on the bottom. The ratios of interest for that column are 3/2, 8, and so
the pivot is the 2 in that column. Then the next tableau is
1
1
1
7
1
0 0
0
2
2
2
2
1
13
1
0
1 0 12 0
2
2
2
1 0 1 0 2 1 0
3
23 0 23 0 0 12 1 11
2
Now you stop because there are no more positive numbers to the left of 1 on the
bottom row. The minimum is 11/2 and it occurs when x1 = x3 = x6 = 0 and
x2 = 7/2, x4 = 13/2, x6 = 11/2.
47
Exercises
Next consider maximization. The simplex tableau was
1 1 1 0 1 0 0
0 0 0 1 1 0 0
0 1 0 0 1 1 0
1 2 1 0 0 0 1
2
8
5
0
This time you work on getting rid of the negative entries. Consider the 1 in the first
column. There is only one ratio to consider so 1 is the pivot.
1 1 1 0 1 0 0 2
0 0 0 1 1 0 0 8
0 1 0 0 1 1 0 5
0 3 0 0 1 0 1 2
There remains a 1. The ratios are 5 and 8 so the next pivot is the 1 in the third row
and column 5.
1 2 1 0 0 1 0 7
0 1 0 1 0 1 0 3
0 1 0 0 1 1 0 5
0 4 0 0 0 1 1 7
1 1 1 1 0 0 10
1 1 1 0 1 0 1
1 2 1 0 0 1 7
The obvious solution is not feasible. Do some row operations.
0 0 0 1 1 0 9
1 1 1 0 1 0 1
1 0 1 0 2 1 5
Now the obvious solution is feasible.
0 0 0
1 1 1
1 0 1
1 2 0
First preserve the simple columns.
0 0 0
1 1 1
1 0 1
3 0 2
1 1 0 0 9
0 1 0 0 1
0 2 1 0 5
0 0 0 1 0
1
0
0
0
1
1
2
2
0
0
1
0
0
0
0
1
9
1
5
2
48
Exercises
Lets try to maximize first. Begin with the first
Use it.
0 0 0 1 1 0
1 1 1 0 1 0
0 1 0 0 1 1
0 3 1 0 1 0
0 1 0 1 0
1 2 1 0 0
0 1 0 0 1
0 4 1 0 0
0 9
0 1
0 6
1 1
0 0 0 1 1 0
1 1 1 0 1 0
1 0 1 0 2 1
3 0 2 0 2 0
There is a positive 2 in the bottom row left of 1.
2.
1
0 21 1 0 12
2
1
1 1 1 0 0
2
2
2
1 0 1 0 2 1
2 0 1 0 0 1
0
0
0
1
= 0.
3
7
6
7
9
1
5
2
0
0
0
1
0 13
2
0 72
0 5
1 7
1
1
1
0
1
1
1
0
1
3
1
0
1
0
0
0
1 1 1
1 1 3
1 1 1
1 1 3
Now use the third column.
2 2
3
1
2
3
0
1
2
3
0
3
0
0
0
1
0
0
0 0
0 1
1 0
0 1
0
0
0
1
8
1
7
0
1 0 0 0 0 8
0 1 0 1 0 1
0 0 1 0 0 7
0 1 0 0 1 1
1 13 0
0 1 0
0 13 1
0 0 0
31
1
31
1
0
0
0
1
23
3
20
3
49
Exercises
It follows that a basic solution is feasible if
2 2
0 1 13 0
3
3
1 1 3 0 1 0
2
2
0 0 13 1
3
3
0
3
3
1 2 3
1 13 0
0 1 0
0 13 1
0 0 0
23
3
20
3
0
0
0
1
23
3
20
3
0 0 13 1 0 20
3
3
3
2 1 0 0 1 0 1 1
Lets do minimization first.
0
1
0
3
0 2 1 1 0 0 7
1 3 0 1 0 0 1
0 2 0 1 1 0 6
0 3 0 2 0 1 2
0 0 0 1 0 1
1 1 1 0 0 1
0 0 2 0 1 1
3 0 1 0 0 2
0
1
0
7
0
6
1 14
0
0 0 1 0 1 0
1
1
1 1 0 0 1 0
7
2
2 0 0 1 3 0 20
4 1 0 0 0 3 1 21
2
0 1 13 0 0 23
3
3
3
1 1 3 0 1 0 0 1
2
0 0 13 1 0 20
3
3
3
2 1 0 0 1 0 1 1
Use the first column.
0
1
0
0
0
1
0
3
2
3
2
6
1
0
0
0
1
1
1
1
0
0
1
0
0
0
0
1
7
1
6
1
50
Exercises
There is still a negative on the bottom row to the left of 1.
0 0 0 1 0 1 0 1
1 1 1 0 0 1 0 7
0 0 2 0 1 1 0 6
0 3 4 0 0 1 1 7
There are no more negatives on the bottom row left of 1 so stop. The maximum
is 7 and it occurs when x1 = 7, x2 = 0, x3 = 0.
1 1 1 1 0 0 10
1 1 1 0 1 0 1
1 2 1 0 0 1 7
The basic solution is not feasible because of that 1. Lets do a row operation to
change this. I used the 1 in the second column as a pivot and zeroed out what
was above and below it. Now it seems that the basic solution is feasible.
2 0 2 1 1 0 11
1 1 1 0 1 0 1
1 0 1 0 2 1 5
Assemble the simplex tableau.
2
0
2 1 1
1
1
1 0 1
1 0 1 0 2
2 1 0 0 0
0
0
1
0
0
0
0
1
11
1
5
0
2
1
1
1
0 2 1 1 0 0 11
1 1 0 1 0 0 1
0 1 0 2 1 0 5
0 1 0 1 0 1 1
2
1
1
1
Lets do minimization first. Work with the third column because there is a positive
entry on the bottom.
0 2 0 1 1 0 0 9
1
1 1 0 1 0 0 1
0
1 0 0 1 1 0 6
2 1 0 0 0 0 1 0
0
1
0
0
2
1
1
1
1 1 0
0 1 0
0 2 1
0 1 0
0 11
0 1
0 5
1 1
51
Exercises
Lets begin with the first column.
0 2
1 1
0 1
0 1
There is still a 2 to the left of
0 3
1 2
0 1
0 3
0
1
0
2
1
0
0
0
1
1
1
2
0
0
1
0
0
0
0
1
1
0
0
0
9
1
6
2
3
7
6
14
0 1 0
0 1 0
1 1 0
0 2 1
1 1 1 1 0 0 10
1 1 1 0 1 0 1
1 2 1 0 0 1 7
Of course the obvious or basic solution is not feasible. Do a row operation involving a pivot in the second row to try and fix this.
2 0 2 1 1 0 11
1 1 1 0 1 0 1
1 0 1 0 2 1 5
Now all is well. Begin to assemble the
2
0
2
1
1
1
1 0 1
1 2 0
simplex tableau.
1 1
0 1
0 2
0 0
0
0
1
0
0
0
0
1
11
1
5
0
2 0 2 1 1 0 0 11
1 1 1 0 1 0 0 1
1 0 1 0 2 1 0 5
1 0 2 0 2 0 1 2
Next lets maximize. There is only one negative
3
0 23 1 0 12
2
1 1 1 0 0 1
2
2
2
1 0 1 0 2 1
0 0 1 0 0 1
0 27
2
0 72
0 5
1 7
52
Exercises
Next lets find the minimum.
2
1
1
1
0
1
0
1
0
1
0
0
2
1
1
2
1 1 0
0 1 0
0 2 1
0 2 0
0 11
0 1
0 5
1 2
1
0
0
0
0
0
0
1
has a 2.
2
1
1
2
0
1
0
0
1
1
1
0
0
0
1
0
9
1
6
0
1 1 1 0 1
1 2 0 1 0
0 0 0 0 1
Thus
0 12
0 5
1 0
1 1 1 0 1 0 12
1 2 0 1 0 0 5
1 1 1 0 0 1 12
Next start with the 1 in the first column.
0 1 1 1 1
1 2
0
1 0
0 1 1 1 0
0 7
0 5
1 7
Thus the minimum value of z = x5 is 7 but, for there to be a feasible solution, you
would need to have this minimum value be 0.
4. Find a solution to the following inequalities for x, y 0 if it is possible to do so. If it
is not possible, prove it is not possible.
(a)
6x + 3y 4
8x + 4y 5
Use an artificial variable. Let x1 = x, x2
artificial
variable x5 . Then minimize
x5 as
6 3 1 0 1 0 4
8 4 0 1 0 0 5
0 0 0 0 1 1 0
Keep the simple columns.
6 3 1 0
8 4 0 1
6 3 1 0
1
0
0
0 4
0 5
1 4
53
Exercises
Now proceed to minimize.
0 0
8 4
0 0
1 43
0
1
1 43
6 0 4
5 4 4
6 6 5
1
0
0
1
4
0
0
1
5
1
4
1
0
0
0 0 11
1 0 8
0 1 11
It is not clear whether there is a solution which has all variables nonnegative.
However, if you do a row operation using 5 in the first column as a pivot, you get
0 24
45 1 65 0 75
5
5
4
4 0 1 0 8
6
1
0
0 65 1 75
5
5
6 0 4 1 0 0 11
5 4 4 0 1 0 9
6 6 5 0 0 1 9
Lets include an artificial variable
6 0 4
5 4 4
6 6 5
0 0 0
Preserving the simple columns,
6 0 4
5 4 4
6 6 5
5 4 4
Use the first column.
0 6 1
0 1 1
6
6 6
5
0 1 61
1 0 0 0 0 11
0 1 0 1 0 9
0 0 1 0 0 9
0 0 0 1 1 0
0
1
0
0
0
0
0
1
11
9
9
9
0 1
1 65
0
1
1 65
0
1
0
0
0
0
0
1
1 0 0
0 1 0
0 0 1
0 1 0
1
0
0
0
2
3
2
9
3
2
It appears that the minimum value for x7 is 3/2 and so there will be no solution
to these inequalities for which all the variables are nonnegative.
54
Exercises
(d)
x1 x2 + x3 2
x1 + 2x2 4
3x1 + 2x3 7
The augmented matrix is
1
1
3
1 1 1
2 0 0
0 2 0
0 0
1 0
0 1
2
4
7
Lets add in an artificial variable and set things up to minimize this artificial
variable.
1 1 1 1 0 0 0 0 2
1 2 0 0 1 0 1 0 4
3 0 2 0 0 1 0 0 7
0 0 0 0 0 0 1 1 0
Then
1 1 1
1 2 0
3 0 2
1 2 0
3
0
0
2
0
0
1
0
2
0
1 0 0
0 1 0
0 0 1
0 1 0
1
0
0
0
21
1
0
0
0
0
1
0
0
0
0
1
2
4
7
4
0
1 0
0 0
1 1
4
4
7
0
0
1
0
0
1
2
It appears the minimum value of x7 is 0 and so this will mean there is a solution
when x2 = 2, x3 = 0, x1 = 0.
(e)
5
6
5
5
6
5
0
2 4 1
3 5 0
2 4 0
0 0
1 0
0 1
1
2
5
2 4 1 0 0 0 0 1
3 5 0 1 0 1 0 2
2 4 0 0 1 0 0 5
0 0 0 0 0 1 1 0
5 2 4 1 0 0
6 3 5 0 1 0
5 2 4 0 0 1
6 3 5 0 1 0
0
1
0
0
0
0
0
1
1
2
5
2
55
Exercises
work with the first column.
5 2
0 3
5
0 0
0 53
4
1
5
0
1
5
5
2
1 1
2
4
0
0
41 21
1
56
1
56
4
0
0
0
0
1
0
1
1
45
1
45
0
0
1
0
0
1
0
1
0
0
1
0
0
1
0
0
0
1
0
0
0
0
0
1
0
0
0
1
1
4
5
4
4
5
1
3
4
4
3
4
M y1 + 3 (M y2 )
M y1 + M y2
20
18
y1 + y2
4M 20 y1 + 3y2
2M 18 y1 + y2
You could simply regard M as large enough that yi 0 and use the techniques just
developed. The augmented matrix for the constraints is then
1 1 1 0
0
2M 2
1 3 0 1 0 4M 20
1 1 0 0 1 2M 18
Here M is large. Use the 3 as a pivot to zero out above and below it.
2
3
1
2
3
0
3
0
0 0
0 3
2
0
3
1 13
0
0 1 0
0 13 1
+ 14
3
4M 20
2
34
3M 3
2
3M
1
3
2
16
3M 3
2
34
3M 3
56
Exercises
Now it appears that the basic solution is feasible provided M
the simplex tableau.
0
0 1 0
1 0
16
3
3
0
3
0
0
3M
3
2
2
2
1
2
34
0
0
1
0
M
3
3
3
3
1 1 0 0
0 1
0
0 0
0 3
2
0
3
0 0
0
3
0
2
1 0
1 1
16
3M 3
2
34
3M 3
2M 18
There is a negative number to the left of the 1 on the bottom row and we want to
maximize so work with this column. Assume M is very large. Then the pivot should
be the top entry in this column.
0 0 1
0 1 0
16
0 3 3 3 0 0 3M 27
2
2
2
0 1
0 0 23 M + 14
3
3
3
0 0 1
0 0 1 2M 2
It follows that the maximum of y1 +y2 is 2M 2 and it happens when y1 = M +7, y2 =
M 9, y3 = 0. Thus the minimum of x1 + x2 is
M y1 + M y2 = M (M + 7) + M (M 9) = 2
F.23
Exercises
7.3
1. If A is the matrix of a linear transformation which rotates all vectors in R2 through
30 , explain why A cannot have any real eigenvalues.
Because the vectors which result are not parallel to the vector you begin with.
2. If A is an n n matrix and c is a nonzero constant, compare the eigenvalues of A and
cA.
c
3. If A is an invertible n n matrix, compare the eigenvalues of A and A1 . More
generally, for m an arbitrary integer, compare the eigenvalues of A and Am .
1 and m .
4. Let A, B be invertible n n matrices which commute. That is, AB = BA. Suppose
x is an eigenvector of B. Show that then Ax must also be an eigenvector for B.
Say Bx = x. Then
B (Ax) = A (Bx) = Ax = (Ax)
57
Exercises
m1 = 1
and so || = 1.
6. Show that if Ax = x and Ay = y, then whenever a, b are scalars,
A (ax + by) = (ax + by) .
Does this imply that ax + by is an eigenvector? Explain.
The formula is obvious from properties of matrix multiplications. However, this vector
might not be an eigenvector because it might equal 0 and
1
1
1
the matrix
1 7
0 4 .
1 5
Determine
whether
defective.
the matrix is
1 1 7
3
2
1 0 4 , eigenvectors:
1
1
1,
2. This is a defective matrix.
1 1 5
1
1
3 7 19
2 1 8 .
2 3 10
Determine
whether
1
1
2 3 10
1
7 12 30
3 7 15 .
3 6 14
Determine
whetherthe matrix is defective.
7 12 30
5
2
2
3 7 15 , eigenvectors: 1 , 0 1, 1 2
3 6 14
0
1
1
58
Exercises
7
8
2
the matrix
2 0
1 0 .
4 6
7 2 0
21
0
8 1 0 , eigenvectors: 1 3, 0 6
2 4 6
1
1
3 2
0 5
0 2
matrix
1
1 .
4
0
3 2 1
1
1
0 5
1 , eigenvectors: 0 , 12 3, 1 6
0
0 2
4
1
1
This matrix is not defective.
6 8 23
4 5 16
3 4 12
Determine whether the matrix is defective.
5 2 5
12 3 10 .
12 4 11
Determine
5 2
12 3
12 4
5
5
31
6
10 , eigenvectors: 1 , 0 1
11
0
1
This matrix is defective. In this case, there is only one eigenvalue, 1 of multiplicity
3 but the dimension of the eigenspace is only 2.
14. Find the eigenvalues and eigenvectors of the matrix
20 9 18
6 5 6 .
30 14 27
Determine whether the matrix is defective.
20 9 18
6
5 6 , eigenvectors:
30 14 27
59
Exercises
1
9
2
13
1, 1 2, 3 3
13
1
1
1
3
4
1
4
Not defective.
1
26 17
4
4
4 .
9 18
9
Determine whether the matrix is defective.
3 1
11 3
8
0
matrix
2
9 .
6
3 1 2
4
11 3 9 , eigenvectors: 1 0 This one is defective.
4
8
0 6
1
2
1 2
11 2 9 .
8
0 7
Determine whether the matrix is defective.
3
2
1 2
4
11 2 9 , eigenvectors: 1 1
4
8
0 7
1
This is defective.
2 1 1
2 3 2 .
2 2 1
2 1 1
1
1
2
2 3 2 , eigenvectors: 1 , 0 1, 1 2
2 2 1
0
1
1
This is non defective.
4 2 2
0 2 2 .
2 0
2
60
Exercises
4 2 2
0 2 2 , eigenvectors:
2 0
2
i
i
1 4, i 2 2i, i 2 + 2i
1
1
1
9
6 3
0
6
0 .
3 6 9
1
9
6 3
2
1
0
6
0 , eigenvectors: 1 , 0 6, 0 12
1
1
0
3 6 9
This is nondefective.
4 2 2
21. Find the complex eigenvalues and eigenvectors of the matrix 0 2 2 . De2 0
2
termine whether the matrix is defective.
4 2 2
0 2 2 , eigenvectors:
2 0
2
i
i
1 4, i 2 2i, i 2 + 2i
1
1
1
4 2
0
22. Find the complex eigenvalues and eigenvectors of the matrix 2 4 0 .
2 2 2
Determine whether the matrix is defective.
4 2
0
2 4 0 , eigenvectors:
2 2 2
1
1
0
0 , 1 2, 1 6
1
0
1
This is not defective.
1
1 6
23. Find the complex eigenvalues and eigenvectors of the matrix 7 5 6 .
1 7
2
Determine whether the matrix is defective.
1
1 6
7 5 6 , eigenvectors:
1 7
2
i
i
1 6, i 2 6i, i 2 + 6i
1
1
1
61
Exercises
This is not defective.
4 2 0
24. Find the complex eigenvalues and eigenvectors of the matrix 2 4 0 . Deter2 2 6
mine whether the matrix is defective.
4 2 0
2 4 0 , eigenvectors:
2 2 6
1
0
1
0 6, i 4 2i, i 4 + 2i
1
1
1
This is not defective.
1
0
0
0
0
0
c
2
a 0
1 b
0 2
0 0
Find values of a, b, c for which the matrix is defective and values of a, b, c for which it
is nondefective.
First consider the eigenvalue = 1
0 a
0 0
0 0
0 0
0
b
1
0
1
b
0
0
0
c
1
0
0
0
0
Then you have ax2 = 0, bx3 = 0. If neither a nor b = 0 then = 1 would be a defective
eigenvalue and the matrix would be defective. If a = 0, then the dimension of the
eigenspace is clearly 2 and so the matrix would be nondefective. If b = 0 but a 6= 0,
then you would have a defective matrix because the eigenspace would have dimension
less than 2. If c 6= 0, then the matrix is defective. If c = 0 and a = 0, then it is non
defective. Basically, if a, c 6= 0, then the matrix is defective.
26. Here is a matrix.
a
0
0
0
1
c
where a, b, c are numbers. Show this is sometimes defective depending on the choice
of a, b, c. What is an easy case which will ensure it is not defective?
An easy case which will ensure that it is not defective is for a, b, c to be distinct. If
you have any repeats, then this will be defective.
27. Suppose A is an n n matrix consisting entirely of real entries but a + ib is a complex
eigenvalue having the eigenvector, x + iy. Here x and y are real vectors. Show that
then a ib is also an eigenvalue with the eigenvector, x iy. Hint: You should
remember that the conjugate of a product of complex numbers equals the product of
the conjugates. Here a + ib is a complex number whose conjugate equals a ib.
A (x + iy) = (a + ib) (x + iy) . Now just take complex conjugates of both sides.
62
Exercises
28. Recall an n n matrix is said to be symmetric if it has all real entries and if A = AT .
Show the eigenvalues of a real symmetric matrix are real and for each eigenvalue, it
has a real eigenvector.
Let be an eigenvalue which corresponds to x 6= 0. Then xT = xT AT . Then
xT x
= xT AT x
= xT A
x = xT x
= . We have A
which shows that
x =
x and Ax = x so
A (
x + x) = (
x + x)
Hence it has a real eigenvector.
29. Recall an n n matrix is said to be skew symmetric if it has all real entries and if
A = AT . Show that any nonzero eigenvalues must be of the form ib where i2 = 1.
In words, the eigenvalues are either 0 or pure imaginary.
Let A be skew symmetric. Then if x is an eigenvector for ,
xT x
= xT AT x
= xT A
x = xT x
32. Suppose A is an n n matrix having all real eigenvalues which are distinct. Show
there exists S such that S 1 AS = D, a diagonal matrix. If
1
0
..
D=
.
0
define eD by
and define
eD
e1
0
..
.
en
eA SeD S 1 .
Next show that if A is as just described, so is tA where t is a real number and the
eigenvalues of At are tk . If you differentiate a matrix of functions entry by entry so
that for the ij th entry of A (t) you get aij (t) where aij (t) is the ij th entry of A (t) ,
show
d At
e
= AeAt
dt
Next show det eAt 6= 0. This is called the matrix exponential. Note I have only
defined it for the case where the eigenvalues of A are real, but the same procedure will
work even for complex eigenvalues. All you have to do is to define what is meant by
63
Exercises
ea+ib .
e 1
0
1
..
S
S
.
0
etn
And so
d At
e
=
dt
= S
0
1
..
0
..
1
.
= AetA
0
..
1 et1
1
S
n etn
t1
0
e
n
0
1
S S
..
.
etn
et1
0
..
etn
1
1
2
1
1 , 1 , 1 , 1 , 1 , 1
2
2
3
1
0
1
7
12
41
1
6
1
S
1
S
14
7
12
16
1
6
61
2
3
. The
31 31
3
7
1
1
The eigenvalues are 1, 2, and 1. What is the physical interpreta3
6
6
1
1
7
3
6
6
tion of the repeated eigenvalue?
There are apparently two directions in which there is no stretching.
1 1
2
2
2
1 , 0 1, 1 2
0
1
1
3 1 1
1 2 0 The eigenvalues are 1, 4, and 2.
1 0 2
64
Exercises
The eigenvalues are given and so there are oscillatory solutions of the form
1
1 (a cos (t) + b sin (t)) ,
1
0
1 c sin
2t + d cos
2t
1
2
1 (e cos (2t) + f sin (2t))
1
|x| = x x = (x) x = (M x) x = x M x
= x M x = x x = |x|
Hence = .
38. Let M be an n n matrix and suppose x1 , , xn are n eigenvectors which form a
linearly independent set. Form the matrix S by making the columns these vectors.
Show that S 1 exists and that S 1 M S is a diagonal matrix (one having zeros everywhere except on the main diagonal) having the eigenvalues of M on the main diagonal.
When this can be done the matrix is said to be diagonalizable.
65
Exercises
Since the vectors are linearly independent, the matrix S has an inverse. Denoting this
inverse by
T
w1
..
1
S = .
wnT
wiT xj = ij .
Therefore,
w1T
S 1 M S = S 1 (M x1 , , M xn ) = ... (1 x1 , , n xn )
wnT
1
..
0
.
39. Show that a n n matrix M is diagonalizable if and only if Fn has a basis of eigenvectors. Hint: The first part is done in Problem 38. It only remains to show that if the
matrix can be diagonalized by some matrix S giving D = S 1 M S for D a diagonal
matrix, then it has a basis of eigenvectors. Try using the columns of the matrix S.
The formula says that
M S = SD
Letting xk denote the k th column of S, it follows from the way we multiply matrices
that
M xk = k xk
where k is the k th diagonal entry on D.
40. Let
and let
1
3
A=
0
B=
2
4
1
0 1
1 1
2 1
2
0
1 2 2
0 1
6 5
AB = 3 4 0 1 1 = 4 7
0 1 3
2 1
7 4
1
3
2
4
, A12 =
66
Exercises
Now consider doing it by block multiplication. This leads to
1 2
0 1
2
2
1
+
3 4
0
1 1
0 1
0 1
+3 2 1
1 1
6 5
6 5
4 7 = 4 7
=
7 4
7 4
41. Suppose A, B are n n matrices and is a nonzero eigenvalue of AB. Show that then
it is also an eigenvalue of BA. Hint: Use the definition of what it means for to be
an eigenvalue. That is,
ABx = x
where x 6= 0. Maybe you should multiply both sides by B.
0
1
C =
0
..
.
0
..
a0
a1
..
.
an1
..
.
..
.
1
+ an1
( + an1 ) det
a0
a1
..
.
..
.
..
.
1
67
Exercises
0
1
1
+1 det
det
..
.
..
.
1
a0
a1
a2
..
.
an2
form
..
.
a0
a1
a2
..
.
..
.
1
+ (an2 )
n1 + (an2 ) n2 + + a1 + a0
and so, the sum of the two is of the form
n1 ( + an1 ) + n1 + (an2 ) n2 + + a1 + a0
= n + an1 n1 + an2 n2 + + a1 + a0
It is routine to verify that when n = 2 this determinant gives the right polynomial.
44. A discreet dynamical system is of the form
x (k + 1) = Ax (k) , x (0) = x0
where A is an n n matrix and x (k) is a vector in Rn . Show first that
x (k) = Ak x0
for all k 1. If A is nondefective so that it has a basis of eigenvectors, {v1 , , vn }
where
Avj = j vj
you can write the initial condition x0 in a unique way as a linear combination of these
eigenvectors. Thus
n
X
x0 =
aj vj
j=1
n
X
aj Ak vj =
j=1
n
X
aj kj vj
j=1
which gives a formula for x (k) , the solution of the dynamical system.
The first formula is obvious from induction. Thus
Ak x0 =
n
X
j=1
aj Ak vj =
n
X
aj kj vj
j=1
68
Exercises
An + an1 An1 + + a1 A + a0 I v = 0
If A is nondefective, give a very easy proof of the Cayley Hamilton theorem based on
this. Recall this theorem says A satisfies its characteristic equation,
An + an1 An1 + + a1 A + a0 I = 0.
Let v be any vector in a basis of eigenvectors of A. Since A is nondefective, such a
basis exists. Then
An + an1 An1 + + a1 A + a0 I v = n + an1 n1 + + a1 + a0 v = 0
n
X
aj 12
j vj =
j=1
n
X
aj vj = x
j=1
n
X
aj vj
j=1
then
nm
nm
x=A
n
X
aj vj =
j=1
so Anm = I.
n
X
j=1
nm
aj A
vj =
n
X
aj vj = x
j=1
48. Sometimes sequences come in terms of a recursion formula. An example is the Fibonacci sequence.
x0 = 1 = x1 , xn+1 = xn + xn1
Show this can be considered as a discreet dynamical system as follows.
xn+1
1 1
xn
x1
1
=
,
=
xn
1 0
xn1
x0
1
69
Exercises
Now use the technique of Problem 44 to find a formula for xn .
What are the eigenvalues and eigenvectors of this matrix?
1 1
, eigenvectors:
1 0
1 1
1
1
1
1
1 1
2 2 5
2 5+ 2
5,
5+
1
1
2 2
2
2
Now also
1
1
5
2 10
1
2
1
2
5
1
1
5+
10
2
xn+1
xn
1
2
5+
1
1
2
1
1
n 1 1
1
1 1
1
2 2 5
5
1
2 10
2 2
n 1
1
1
1
1
1
2 5+ 2
+
5+
5+
1
10
2
2
2
In particular,
xn =
n
n
1
1 1
1
1
1
1
1
5 +
5+
5+
2 10
2 2
10
2
2
2
F.24
Exercises
7.10
1. Explain why it is typically impossible to compute the upper triangular matrix whose
existence is guaranteed by Schurs theorem.
To get it, you must be able to get the eigenvalues and this is typically not possible.
2. Now recall the QR factorization of Theorem 5.7.5 on Page 133. The QR algorithm
is a technique which does compute the upper triangular matrix in Schurs theorem.
There is much more to the QR algorithm than will be presented here. In fact, what
I am about to show you is not the way it is done in practice. One first obtains what
is called a Hessenburg matrix for which the algorithm will work better. However,
the idea is as follows. Start with A an n n matrix having real eigenvalues. Form
A = QR where Q is orthogonal and R is upper triangular. (Right triangular.) This
70
Exercises
can be done using the technique of Theorem 5.7.5 using Householder matrices. Next
take A1 RQ. Show that A = QA1 QT . In other words these two matrices, A, A1 are
similar. Explain why they have the same eigenvalues. Continue by letting A1 play the
role of A. Thus the algorithm is of the form An = QRn and An+1 = Rn+1 Q. Explain
why A = Qn An QTn for some Qn orthogonal. Thus An is a sequence of matrices each
similar to A. The remarkable thing is that often these matrices converge to an upper
triangular matrix T and A = QT QT for some orthogonal matrix, the limit of the Qn
where the limit means the entries converge. Then the process computes the upper
triangular Schur form of the matrix A. Thus the eigenvalues of A appear on the
diagonal of T. You will see approximately what these are as the process continues.
Suppose you have found An . Then An = Qn Rn and then
An+1 Rn Qn = QTn An Qn = QTn QTn1 An1 Qn1 Qn
Now the product of orthogonal matrices is orthogonal and so this shows by induction
that each An is orthogonally similar to A.
3. Try the QR algorithm on
1 2
6
6
which has eigenvalues 3 and 2. I suggest you use a computer algebra system to do the
computations.
1
6
1 2
37
37 37
37
37 38
37
37
=
6
6
1
6
6
37 37
0
37 37
37 37
191
1
38
6
37 37 37 37
260
37 37 37
37
37
A1 =
=
36
6
6
6
1
37
0
37
37
37
37 37
37 37
191
260
37
37
=
36
6
37
37
191
1
36
1348
1021
37
1021
37 777 371021 37 777 37
37 37 1021 37 777
36
191
6
0
37 777 37 1021
37 777 37 1021
1021 37 1021
3959
7952
1021
1021
A2 =
216
1146
1021
1021
3959
7952
1021
1021
=
0.211 56 1146
1021
0.998 51
5. 447 9 102
3. 883 3 7. 715 7
5. 447 9 102
0.998 51
0
1. 545 1
3. 883 3 7. 715 7
0.998 51
5. 447 9 102
A3 =
=
0
1. 545 1
5. 447 9 102
0.998 51
3. 457 2
7. 915 8
8. 417 6 102 1. 542 8
You can keep on going. You see that it appears to be converging to an upper triangular
matrix having 3, 1 down the diagonal.
0
2
1
0
Show that the algorithm cannot converge for this example. Hint: Try a few iterations
of the algorithm.
Exercises
71
1
0 1
2 0
=
0
1 0
0 1
2 0
0 1
0 2
A1 =
=
0 1
1 0
1 0
0 2
0 1
1 0
=
1 0
1 0
0 2
1 0
0 1
0 1
A2 =
=
. Now it is back to where you started.
0 2
1 0
2 0
0 1
0 2
Thus the algorithm merely bounces between the two matrices
and
2 0
1 0
and so it cant possibly converge.
0 1
0 2
5. Show the two matrices A
and B
are similar; that is
4 0
2 0
there exists a matrix S such that A = S 1 BS but there is no orthogonal matrix
Q such that QT BQ = A. Show the QR algorithm does converge for the matrix B
although it fails to do so for A.
0
2
Both matrices have distinct eigenvalues 2 and so they are both similar to a diagonal
matrix having these eigenvalues on the diagonal. Therefore, the matrices are indeed
similar. Lets try the QR algorithm on the second.
0 2
0 1
2 0
=
2 0
1 0
0 2
2 0
0 1
0 2
A1 =
=
and so the algorithm keeps on returning
0 2
1 0
2 0
the same matrix in the case of B. Now consider the matrix A.
0 1
0 1
4 0
=
4 0
1 0
0 1
4 0
0 1
0 4
A1 =
=
0 1
1 0
1 0
0 4
0 1
1 0
=
1 0
1 0
0 4
1 0
0 1
0 1
A2 =
=
0 4
1 0
4 0
0 1
0 4
In this case, the algorithm bounces between
and
.
4 0
1 0
6. Let F be an m n matrix. Show that F F has all real eigenvalues and furthermore,
they are all nonnegative.
The eigenvalues are real because the matrix F F is Hermitian. If is one of them
with eigenvector x,
(x, x) = (F F x, x) = (F x,F x) 0
7. If A is a real n n matrix and is a complex eigenvalue of A having eigenvector
z + iw, show that w 6= 0.
A (z + iw) = (a + ib) (z + iw)
72
Exercises
Now subtract to obtain A (2iw) = 2iaw + 2ibz. Now if w = 0, then you would have
0 =ibz and this requires b = 0 since otherwise, the eigenvector was equal to 0 which
it isnt. Hence was not complex after all.
8. Suppose A = QT DQ where Q is an orthogonal matrix and all the matrices are real.
Also D is a diagonal matrix. Show that A must be symmetric.
AT = QT DT Q = QT DQ = A.
9. Suppose A is an n n matrix and there exists a unitary matrix U such that
A = U DU
where D is a diagonal matrix. Explain why A must be normal.
A A = U D U U DU = U D DU
AA = U DU U D U = U DD U. But D D = DD and both are equal to the
diagonal matrix which has the squares of the absolute values of the diagonal entries
of D down the diagonal.
11. Show that every unitary matrix preserves distance. That is, if U is unitary,
|U x| = |x| .
|U x|2 = (U x, U x) = (U U x, x) = (x, x) = |x|2 .
12. Show that if a matrix does preserve distances, then it must be unitary.
Let U preserve distances. Then
2
|U (x + y)| = (U x + U y,U x + U y)
2
= |U x| + |U y| + 2 Re (U x, U y)
= |x|2 + |y|2 + 2 Re (U x, U y)
Therefore,
Re (U x, U y) Re (x, y)
Re (U U x x, y)
= 0
= 0
(U U x x, y) = 0
73
Exercises
13. Show that a complex normal matrix A is unitary if and only if its eigenvalues have
magnitude equal to 1.
Since A is normal, it follows that there is a unitary matrix U such that
AU = U D
where D is a diagonal matrix. Then from the above problems, A is unitary if and only
if it preserves distances. Let the columns of U be the orthonormal set {u1 , , un } .
Then we have Auk = k uk . For A to preserve distances, you must have |k | = 1.
Conversely, if |k | = 1 for all k, then {k uk } is also an orthonormal set and so A
preserves distances and is therefore normal.
14. Suppose A is an n n matrix which is diagonally dominant. Recall this means
X
|aij | < |aii |
j6=i
1 + 2i 4 2
0
i 3
5
6 7
B (1 + 2i, 6) , B (i, 3) , B (7, 11)
16. Show a square matrix is invertible if and only if it has no zero eigenvalues.
To say the matrix has no zero eigenvalues is to say that it is one to one because it
maps no nonzero vector to 0.
17. Using Schurs theorem, show the trace of an n n matrix equals the sum of the
eigenvalues and the determinant of an n n matrix is the product of the eigenvalues.
Let A be an n n matrix. By Schur, there exists a unitary matrix U and an upper
triangular matrix T such that
A = U T U
Then det (A) = det (T ) = the product of the diagonal entries of T. These are
eigenvalues of T and both A and T have the same eigenvalues because they have
same characteristic polynomial. Thus the determinant of A equals the product of
eigenvalues. Since A and T are similar, they have the same trace also. However,
trace of T is just the sum of the eigenvalues of A.
the
the
the
the
74
Exercises
= trace (U T T U ) = trace (T T )
n
X
i=1
|i |
Equality holds if and only if there are no nonzero off diagonal terms. This happens if
and only if T is a diagonal matrix. But if this is so, then A must be normal because
U AU = D, a diagonal matrix and this implies that A is normal.
19. Here is a matrix.
1234
6
5
3
0
654
9
123
98
123 10, 000 11
56
78
98
400
I know this matrix has an inverse before doing any computations. How do I know?
Gerschgorins theorem shows that there are no zero eigenvalues and so the matrix is
invertible.
20. Show the critical points of the following function are
1
(0, 3, 0) , (2, 3, 0) , and 1, 3,
3
18x2 + 36x + 2z 12 0 2x 2
0
4
0
Hessian is
2x 2
0
3
(0, 3, 0) ,
12 0
0
4
2
0
2
0
3
2
0
3
The eigenvalues are all negative and so this point is a local maximum.
(2, 3, 0)
12 0
0
4
2
0
0
0
3
0 4 0
0
0 3
75
Exercises
21. Here is a function of three variables.
f (x, y, z) = 13x2 + 2xy + 8xz + 13y 2 + 8yz + 10z 2
change the variables so that in the new variables there are no mixed terms, terms
involving xy, yz etc. Two eigenvalues are 12 and 18.
The function is
x
13 1 4
x
f (x, y, z) = y 1 13 4 y
z
4 4 10
z
13 1 4
1 13 4 , eigenvectors:
10
4 1 4
2
1
1
1 6, 1 12, 1 18
2
1
0
1
Then in terms of the new variables, the quadratic form is 6x2 + 12y 2 + 18z 2. Note
how I only needed to find the eigenvalues. This changes in the next problems.
T
x
x
2
9/2 3/2
x
y 9/2
5
9/2 y + 4 9 3 y + 2
z
z
3/2 9/2 7
z
2
9/2 3/2
9/2
5
9/2 , eigenvectors:
7
3/2 9/2
0
2
3 1, 1 17 , 3 19
2
2
1
1
1
35
0
71 14
7
3
1
3
14
1
3
1
35
10
14
35
10
14
1
0
0
1
0
0
, A = Q 0 17 0 QT Thus you need to have
QT AQ = 0 17
0
2
2
19
19
0
0
0
0
2
2
the new variables be
1
T
35
0
71 14
x
x
7
3 35 1 10 3 14 y = y
35
10
14
1
3
1
z
z
35 35
10 10
14 14
Then in terms of the new variables, the quadratic form is
76
Exercises
T
1
0
0
x
x
y 0 17 0 y +
2
19
z
0
0
z
2
1
35
0
17 14
x
7
3
1
3
y + 2
4 9 3 35
35
10
14
10
14
1
3
1
z
35 35
10 10
14 14
19 2
19
2
=(x )2 + 72 5 7x 17
2 7z + 2
2 (y ) + 2 (z ) + 7
T
x
1
y 1
z
1
1 1
1
1 1 1
1
1 1
1
1
x
x
1 1 y + 1 0 0 y
1 1
z
z
1
1
1
, eigenvectors: 1 1, 1 , 1 2
0
2
1
1
1
1
3 3 2 2 6 6
Appropriate orthogonal matrix: Q = 31 3 21 2 16 6 .
1
1
0
3 3
3 6
New variables,
x
y =
z
1
3 3
1
3 3
1
3 3
12 2
1
2 2
0
61 6
x
61 6 y
1
z
3 6
1
1
x
3 3 2 2 6 6
1
+ 1 0 0 13 3
61 6 y
2 2
1
1
z
0
3 3
3 6
for t R and classify them as local minima, local maxima or saddle points.
77
Exercises
4y 8z 4x 6 8x 12
4x 6
2
2
8x 12
2
0
Plug in the critical points.
0
4t 6 8t 12
4t 6
You need to consider the sign of the eigenvalues.
2
2
8t 12
2
0
4t 6 8t 12
= 80t2 + 240t 3 + 22 + 184. Thus you
2
2
det 4t 6
8t 12
2
184 + 80t2 + 240t has a minimum value when t = 23 and at this point, the value of
2
this function is 184 + 80 23 + 240 32 = 4. Therefore, from the quadratic formula,
there is a positive eigenvalue and a negative eigenvalue for any value of t. Thus this
function always has a direction in which it appears to have a local min. and a direction
in which it appears to have a local max. for each of the critical points.
6x2 24x 6z + 16 0 12 6x
0
4
0
Hessian:
12 6x
0
1
f (x, y, z) =
(0, 1, 0)
16 0
0 4
12 0
12
0
1
This is a saddle point because it has a negative and two positive eigenvalues.
(4, 1, 0)
22 0
0
4
12 0
12
0
1
78
Exercises
This is also a saddle point because it has a negative and two positive eigenvalues.
(2, 1, 12)
64 0 0
0 4 0
0 0 1
36x2 2y 48 2x
2x
0
96 4
4 0
Clearly it has eigenvalues of different sign because the determinant is negative. Therefore, this is a saddle point. (2, 0) works out the same way. To have some fun, graph
this function of two variables and see if you can see the critical points are saddle points
from the picture.
27. Let f (x, y) = 3x4 5x2 + 2 y 2 x2 + y 2 . Find and classify the critical points using the
second derivative test.
12x3 2xy 2 10x
4
2
2 2
2
3x 5x + 2 y x + y =
2y 2x2 y
Critical points: (1, 1) , (1, 1) , (1, 1) , (1, 1) , (0, 0) , 61 5 6, 0 , 16 5 6, 0
36x2 2y 2 10 4xy
3x4 5x2 + 2 y 2 x2 + y 2 , Hessian is
4xy
2 2x2
(1, 1)
24 4
4 0
28. Let f (x, y) = 5x4 7x2 2 3y 2 x2 + 11y 2 4y 4 . Find and classify the critical points
using the second derivative test.
20x3 6xy 2 14x
5x4 7x2 2 3y 2 x2 + 11y 2 4y 4 =
6x2 y 16y 3 + 22y
1
1
7 10, 0 , 10
7 10, 0
Critical points (1, 1) , (1, 1) , (0, 0) , 10
5x4 7x2 2 3y 2 x2 + 11y 2 4y 4 ,
79
Exercises
Hessian is
(1, 1)
60x2 6y 2 14
12xy
12xy
6x2 48y 2 + 22
40 12
12 32
40 12
12 32
14 0
0
22
28 0
0 89
5
1
This is a local minimum. So is 10 7 10, 0 .
29. Let f (x, y, z) = 2x4 3yx2 + 3x2 + 5x2 z + 3y 2 6y + 3 3zy + 3z + z 2 . Find and
classify the critical points using the second derivative test.
2x4 3yx2 + 3x2 + 5x2 z + 3y 2 6y + 3 3zy + 3z + z 2
54 0
0
0
6 3
0
3 2
It has two positive eigenvalues and one negative eigenvalue so this is a saddle point.
30. Let f (x, y, z) = 3yx2 3x2 x2 z y 2 + 2y 1 + 3zy 3z 3z 2 . Find and classify
the critical points using the second derivative test.
3yx2 3x2 x2 z y 2 + 2y 1 + 3zy 3z 3z 2
6xy 6x 2xz
= 3x2 2y + 3z + 2
x2 + 3y 6z 3
Critical points: (0, 1, 0)
80
Exercises
6y 2z 6
6x
Hessian is
2x
eigenvalues:
6x 2x
2
3
3
6
0
0
0
13 4, 13 4, 0
0
3
6
This has two directions in which it appears to have a local maximum. However, the
test fails because of the zero eigenvalue.
31. Let Q be orthogonal. Find the possible values of det (Q) .
1
32. Let U be unitary. Find the possible values of det (U ) .
U is unitary which means it preserves length. Therefore, if U x = x, it must be the
case that || = 1. Hence = ei for some . In other words, is a point on the unit
circle.
33. If a matrix is nonzero can it have only zero for eigenvalues?
1 1
Definitely yes. Consider this one.
1 1
34. A matrix A is called nilpotent if Ak = 0 for some positive integer k. Suppose A is a
nilpotent matrix. Show it has only 0 for an eigenvalue.
Say is an eigenvalue. Then for some x 6= 0,
Ax = x
Then k is an eigenvalue for Ak and so k = 0. Hence = 0.
35. If A is a nonzero nilpotent matrix, show it must be defective.
If it is not defective, then there exists S such that
S 1 AS = 0
But then A = 0. Hence, if A 6= 0 is nilpotent, then A is defective.
36. Suppose A is a nondefective n n matrix and its eigenvalues are all either 0 or 1.
Show A2 = A. Could you say anything interesting if the eigenvalues were all either
0,1,or 1? By DeMoivres theorem, an nth root of unity is of the form
2k
2k
cos
+ i sin
n
n
Could you generalize the sort of thing just described to get An = A? Hint: Since A
is nondefective, there exists S such that S 1 AS = D where D is a diagonal matrix.
S 1 AS = D and so S 1 A2 S = D2 = D = S 1 AS and so, in the first case, A2 = A.
If the eigenvaluse are 0,1,1, then all even powers of A are equal and all odd powers
are equal.
In the last case, you could have the eigenvalues of A equal to cos 2k
+ i sin 2k
.
n
n
Then
S 1 An S = I
and so An = I.
81
Exercises
37. This and the following problems will present most of a differential equations course.
To begin with, consider the scalar initial value problem
y = ay, y (t0 ) = y0
When a is real, show the unique solution to this problem is y = y0 ea(tt0 ) . Next
suppose
y = (a + ib) y, y (t0 ) = y0
(6.30)
where y (t) = u (t) + iv (t) . Show there exists a unique solution and it is
y (t) = y0 ea(tt0 ) (cos b (t t0 ) + i sin b (t t0 ))
e(a+ib)(tt0 ) y0 .
(6.31)
Next show that for a real or complex there exists a unique solution to the initial value
problem
y = ay + f, y (t0 ) = y0
and it is given by
y (t) = ea(tt0 ) y0 + eat
t0
Hint: For the first part write as y ay = 0 and multiply both sides by eat . Then
explain why you get
d at
e y (t) = 0, y (t0 ) = 0.
dt
Now you finish the argument. To show uniqueness in the second part, suppose
y = (a + ib) y, y (0) = 0
and verify this requires y (t) = 0. To do this, note
y = (a ib) y, y (0) = 0
and that
d
2
|y (t)|
dt
Thus from the first part |y (t)| = 0e2at = 0. Finally observe by a simple computation
that 6.30 is solved by 6.31. For the last part, write the equation as
y ay = f
and multiply both sides by eat and then integrate from t0 to t using the initial
condition.
38. Now consider A an n n matrix. By Schurs theorem there exists unitary Q such that
Q1 AQ = T
where T is upper triangular. Now consider the first order initial value problem
x = Ax, x (t0 ) = x0 .
82
Exercises
Show there exists a unique solution to this first order system. Hint: Let y = Q1 x
and so the system becomes
y = T y, y (t0 ) = Q1 x0
(6.32)
Then use the solution you get in this to get the solution to the initial value problem
which occurs one level up, namely
yn1
= t(n1)(n1) yn1 + t(n1)n yn , yn1 (t0 ) = Q1 x0 n1
Continue doing this to obtain a unique solution to 6.32.
39. Now suppose (t) is an n n matrix of the form
(t) =
x1 (t)
xn (t)
where
(6.33)
(6.34)
are obtained in the form (t) c for some choice of c Fn . In other words, is the
general solution to this equation (t) c for c Fn ? Prove the following theorem using
linear algebra.
Theorem F.24.1 Suppose (t) is an n n matrix which satisfies
(t) = A (t) .
1
83
Exercises
1
c = (t0 ) x
By the existence part of Problem 38 there exists a solution to
x = Ax, x (t0 ) = c
but this cannot be in the form (t) c. Thus for every t, (t)1 exists. Next suppose
1
for some t0 , (t0 ) exists. Let z = Az and choose c such that
z (t0 ) = (t0 ) c
Then both z (t) , (t) c solve
x = Ax, x (t0 ) = z (t0 )
Apply uniqueness to conclude z = (t) c. Finally, consider that (t) c for c Fn
1
either is the general solution or it is not the general solution. If it is, then (t)
1
exists for all t. If it is not, then (t) cannot exist for any t from what was just
shown.
1
41. Let (t) = A (t) . Then (t) is called a fundamental matrix if (t)
t. Show there exists a unique solution to the equation
x = Ax + f , x (t0 ) = x0
x0 + (t)
(s)
f (s) ds
t0
Now these few problems have done virtually everything of significance in an entire undergraduate differential equations course, illustrating the superiority of linear algebra.
The above formula is called the variation of constants formula.
Hint: Uniquenss is easy. If x1 , x2 are two solutions then let u (t) = x1 (t) x2 (t) and
argue u = Au, u (t0 ) = 0. Then use Problem 38. To verify there exists a solution, you
could just differentiate the above formula using the fundamental theorem of calculus
and verify it works. Another way is to assume the solution in the form
x (t) = (t) c (t)
and find c (t) to make it all work out. This is called the method of variation of
parameters.
For the method of variation of parameters,
x (t) =
x (t) =
and so
c (t) =
(s)
f (s) ds + k
t0
84
Exercises
Then the solution desired is
x (t) = (t) k + (t)
(s)
f (s) ds
t0
1
Note that by approximating with Riemann sums and passing to a limit, it follows that
the constant (t) can be taken into the integral to write
Z t
1
x (t) = (t) (t0 ) x0 +
(t) (s)1 f (s) ds
t0
42. Show there exists a special such that (t) = A (t) , (0) = I, and suppose
1
(t) exists for all t. Show using uniqueness that
1
(t) = (t)
and that for all t, s R
(t + s) = (t) (s)
Explain why with this special , the solution to 6.35 can be written as
Z t
x (t) = (t t0 ) x0 +
(t s) f (s) ds.
t0
A (t) (t) A
AA (t) A (t) A
A ( (t))
and also (0) = 0. By uniqueness given above, (t) = 0 for all t. Therefore,
( (t) (t))
85
Exercises
(t)
= (t + s) (t) (s)
= A (t + s) A (t) (s)
= A (t)
and (0) = 0. Therefore, (t) = 0 for all t by uniqueness. It follows that for all t, s,
(t + s) = (t) (s)
Therefore, in the variation of constants formula, it reduces to
Z t
x (t) = (t t0 ) x0 +
(t s) f (s) ds
t0
One thing might not be clear. If (t) = 0, why is (t) a constant? This works for
scalar valued functions by a use of the mean value theorem. However, this is a matrix
valued function. Nevertheless this is true. You just consider the ij th entry which is
eTi (t) ej
Its derivative equals 0 and so it is constant.
43. You can see more on this problem and the next one in the latest version of Horn
and Johnson, [16]. Two n n matrices A, B are said to be congruent if there is an
invertible P such that
B = P AP
Let A be a Hermitian matrix. Thus it has all real eigenvalues. Let n+ be the number
of positive eigenvalues, n , the number of negative eigenvalues and n0 the number of
zero eigenvalues. For k a positive integer, let Ik denote the k k identity matrix and
Ok the k k zero matrix. Then the inertia matrix of A is the following block diagonal
n n matrix.
In+
In
On0
Show that A is congruent to its inertia matrix. Next show that congruence is an
equivalence relation. Finally, show that if two Hermitian matrices have the same
inertia matrix, then they must be congruent. Hint: First recall that there is a
unitary matrix, U such that
Dn+
Dn
U AU =
On0
1/2
Dn+
Dn 1/2
D = D =
In0
Now consider D U AU D.
86
Exercises
The hint gives it away. When you block multiply, the thing which results is the inertia
matrix. Congruance is obviously an equivalence relation. Consider the transitive law
44. Show that if A, B are two congruent Hermitian matrices, then they have the same
inertia matrix. Hint: Let A = SBS where S is invertible. Show that A, B have the
same rank and this implies that they are each unitarily similar to a diagonal matrix
which has the same number of zero entries on the main diagonal. Therefore, letting
VA be the span of the eigenvectors associated with positive eigenvalues of A and VB
being defined similarly, it suffices to show that these have the same dimensions. Show
that (Ax, x) > 0 for all x VA . Next consider S VA . For x VA , explain why
1
(BS x,S x) =
S 1 A (S ) S x,S x
= S 1 Ax,S x = Ax, S 1 S x = (Ax, x) > 0
Next explain why this shows that S VA is a subspace of VB and so the dimension of VB
is at least as large as the dimension of VA . Hence there are at least as many positive
eigenvalues for B as there are for A. Switching A, B you can turn the inequality
around. Thus the two have the same inertia matrix.
The above manipulation is straight forward. Now an orthonormal basis exists for Fn
which consists of eigenvectors of B. Say you have {v1 , , vq , w1 , , wr , z1 , , zs }
where the vi correspond to the positive eigenvalues, the wi to the negative eigenvalues
and the zi to the eigenvalue 0. We know that B is positive on span (v1 , , vq ) . Could
it be positive on any larger subspace? Could it be positive on span (v1 , , vq , u)
where
u span (w1 , , wr , z1 , , zs )?
No, it couldnt because you could then consider (Bu, u) and it would end up being no
larger than 0. It follows that S VA VV = span (v1 , , vq ) and so the dimension
of S VA , which is the same as the dimension of VA which is the same as the number
of positive eigenvalues, counted according to multiplicity, is no larger than q. Now
reverse the argument letting A B. Hence the two have the same number of
positive eigenvalues. Since they have the same number of zero eigenvalues, this implies
they have the same inertia matrix.
45. Let A be an m n matrix. Then if you unraveled it, you could consider it as a vector
in Cnm . The Frobenius inner product on the vector space of m n matrices is defined
as
(A, B) trace (AB )
Show that this really does satisfy the axioms of an inner product space and that it
also amounts to nothing more than considering m n matrices as vectors in Cnm .
(A, B) =
Aij Bij =
i,j
(A, A) =
i,j
X
i,j
Aij Aij =
X
i,j
|Aij |
87
Exercises
and so this does satisfy the axioms of the inner product. Also, this inner product
coincides with the standard inner product on Cnm .
46. Consider the n n unitary matrices. Show that whenever U is such a matrix, it
follows that
|U |Cnn = n
Next explain why if {Uk } is any sequence of unitary matrices, there exists a subsequence {Ukm }
m=1 such that limm Ukm = U where U is unitary. Here the limit
takes place in the sense that the entries of Ukm converge to the corresponding entries
of U .
From the above and the fact that U is unitary,
2
Explain why dist ( (A) , (B)) is small if and only if every eigenvalue of A is close
to some eigenvalue of B. Now prove the following theorem using the above problem
and Schurs theorem. This theorem says roughly that if A is close to B then the
eigenvalues of A are close to those of B in the sense that every eigenvalue of A is close
to an eigenvalue of B.
Theorem F.24.2 Suppose limk Ak = A. Then
lim dist ( (Ak ) , (A)) = 0
88
Exercises
Since Ukm Akm Ukm is upper triangular, it follows that so is T . This shows the claim.
Next suppose that the conclusion does not hold. Then there exists > 0 and a
subsequence, still called k such that
dist ( (Ak ) , (A))
It follows that for every choice of Uk such that Uk Ak Uk = Tk , a triangular matrix and
for every choice of unitary U such that U AU = T, a triangular matrix,
|Tk T |Cn2 dist ( (Ak ) , (A))
Recall from the proof of Schurs theorem, you change the order of the eigenvalues
on the diagonal by adjusting the unitary matrix U . No matter how you permute
the entries on the diagonals, the two are always far apart. Now pick Uk for each k,
such that Uk Ak Uk = Tk a triangular matrix. By the first part of the proof, there
exists a subsequence {km } such that Tkm T where T is unitarily similar to A. This
contradicts the above.
a b
48. Let A =
be a 2 2 matrix which is not a multiple of the identity. Show
c d
that A is similar to a 2 2 matrix which has one entry equal to 0. Hint: First note
that there exists a vector a such that Aa is not a multiple of a. Then consider
B=
Aa
1
a Aa
and since the trace equals 0 the term in the bottom right is also equal to 0.
49. Let A be a complex n n matrix which has trace equal to 0. Show that A is similar
to a matrix which has all zeros on the main diagonal. Hint: Use Problem 30 on
Page 122 to argue that you can say that a given matrix is similar to one which has
the diagonal entries permuted in any order desired. Then use the above problem and
block multiplication to show that if the A has k nonzero entries, then it is similar to
a matrix which has k 1 nonzero entries. Finally, when A is similar to one which has
at most one nonzero entry, this one must also be zero because of the condition on the
trace.
Suppose A has k nonzero diagonal entries. First of all, we can assume k 2 because
the trace equals 0. Then from Problem 30 on Page 122 it is similar to a block matrix
of the form
P Q
R S
89
Exercises
where the diagonal entries are just re ordered and P is a 2 2 matrix which which is
not a multiple of the identity and has both diagonal entries nonzero. Otherwise, the
trace could not equal zero if all the diagonal entries were equal. Then from the above
problem, there exists L such that L1 P L has at least one zero on the diagonal.
1
1
L
0
P Q
L 0
L P L L1 Q
=
0
I
R S
0 I
RL
S
Thus this new matrix is similar to the one above and it has at least one fewer nonzero
entries on the main diagonal. Continue this process obtaining a sequence of similar
matrices, each having fewer nonzero diagonal entries than the preceding one till there
is at most one nonzero entry. It follows that this one must be zero also because the
trace equals 0.
50. An n n matrix X is a comutator if there are n n matrices A, B such that X =
AB BA. Show that the trace of any comutator is 0. Next show that if a complex
matrix X has trace equal to 0, then it is in fact a comutator. Hint: Use the above
problem to show that it suffices to consider X having all zero entries on the main
diagonal. Then define
1
0
(
Xij
2
ij if i 6= j
A=
, Bij =
.
..
0 if i = j
0
n
The first part is easy because the trace of a product is the same in either order.
Suppose then that X has zero trace. Then by the above problem, it is similar to a
matrix Y which has a zero diagonal. Suppose you can show that Y = AB BA. Then
if Y = S 1 XS, you could write
S 1 XS = AB BA,
X = S (AB BA) S 1 = SAS 1 SBS 1 SBS 1 SAS 1
Thus X will be a comutator. It suffices then to show that whenever a matrix has all
zero diagonal entries, it must be a comutator. Now use the definition of A, B in the
hint.
X
X
(AB BA)ij
Ais Bsj
Bir Arj
s
iXij
Xij
j
= Xij
ij
ij
F.25
Exercises
8.4
90
Exercises
1
1
1
0
1. Let H denote span 2 , 4 , 3 , 1 . Find the dimension of H
0
0
1
1
and determine a basis.
1 1 1 0
1 0 0 1
2 4 3 1 , row echelon form: 0 1 0 0 . The first three vectors form
0 0 1 1
0 0 1 1
a basis and the dimension is 3.
2. Let M = u = (u1 , u2 , u3 , u4 ) R4 : u3 = u1 = 0 . Is M a subspace? Explain.
2t + 3s
s t : s, t R .
t+s
Is this set of vectors a subspace of R3 ? If so, explain why, give a basis for the subspace
and find its dimension.
2
3
Yes. A basis is 1 , 1
1
1
91
Exercises
11. Consider the vectors of the form
2t + 3s + u
st
t+s
: s, t, u R .
Is this set of vectors a subspace of R4 ? If so, explain why, give a basis for the subspace
and find its dimension.
It is a subspace. It is spanned by
3
1
1
0
It will be
3 2
1 1
1 1
0 0
pendent.
a basis if it is a linearly
1
0
, row echelon form:
0
1
2
1
1 0
, ,
1 0
0
1
independent set.
1 0 0
0 1 0
2t + u + 1
t + 3u
t+s+v
: s, t, u, v R .
Is this set of vectors a subspace of R4 ? If so, explain why, give a basis for the subspace
and find its dimension.
Because of that 1 this is not a subspace.
13. Let V denote the set of functions defined on [0, 1]. Vector addition is defined as
(f + g) (x) f (x) + g (x) and scalar multiplication is defined as (f ) (x) (f (x)).
Verify V is a vector space. What is its dimension, finite or infinite?
Pick n points {x1 , , xn } . Then let ei (x) = 0 unless x = xi when it equals 1. Then
n
{ei }i=1 is linearly independent, this for any n.
14. Let V denote the set of polynomial functions defined on [0, 1]. Vector addition is
defined as (f + g) (x) f (x) + g (x) and scalar multiplication is defined as (f ) (x)
(f (x)). Verify V is a vector space. What is its dimension, finite or infinite?
This is clearly a subspace of the set of all functions
defined on
[0, 1] . Its dimension is
still infinite. For example, you can see that 1, x, x2 , , xn is linearly independent
for each n. You could use Theorem 8.2.9 to show this, for example.
15. Let V be the set of polynomials defined on R having degree no more than 4. Give a
basis for this vector space.
1, x, x2 , x3 , x4
92
Exercises
16. Let the vectors be of the form a + b 2 where a, b are rational numbers and let the
field of scalars be F = Q, the rational numbers. Show directly this is a vector space.
What is its dimension? What is a basis for this vector space?
A basis is 1, 2 . In fact it is actually a field.
1
2
a+b 2
= aab
2 2b2 . Since a, b are rational, it follows that the denominator is not
0. Thus every element has an inverse so it is actually a field. It is certainly a vector
space over the rational numbers.
17. Let V be a vector space with field of scalars F and suppose {v1 , , vn } is a basis for
V . Now let W also be a vector space with field of scalars F. Let L : {v1 , , vn }
W be a function such that Lvj = wj . Explain how L can be extended to a linear
transformation mapping V to W in a unique way.
Pn
Pn
L ( i=1 ci vi ) i=1 ci wi
18. If you have 5 vectors in F5 and the vectors are linearly independent, can it always be
concluded they span F5 ? Explain.
Yes. If not, you could add in a vector not in the span and obtain a linearly independent
set of 6 vectors which is not possible.
19. If you have 6 vectors in F5 , is it possible they are linearly independent? Explain.
No. There is a spanning set having 5 vectors and this would need to be as long as the
linearly independent set.
20. Suppose V, W are subspaces of Fn . Show V W defined to be all vectors which are in
both V and W is a subspace also.
If x, y V W then a + b V and is also in W because these are subspaces. Hence
V W is a subspace.
21. Suppose V and W both have dimension equal to 7 and they are subspaces of a vector
space of dimension 10. What are the possibilities for the dimension of V W ? Hint:
Remember that a linear independent set can be extended to form a basis.
See the next problem.
22. Suppose V has dimension p and W has dimension q and they are each contained in
a subspace, U which has dimension equal to n where n > max (p, q) . What are the
possibilities for the dimension of V W ? Hint: Remember that a linear independent
set can be extended to form a basis.
Let {x1 , , xk } be a basis for V W. Then there is a basis for V and W which are
respectively
{x1 , , xk , yk+1 , , yp } , {x1 , , xk , zk+1 , , zq }
It follows that you must have k + p k + q k n and so you must have
p+qnk
23. If b 6= 0, can the solution set of Ax = b be a plane through the origin? Explain.
No. It cant. It does not contain 0.
Exercises
93
24. Suppose a system of equations has fewer equations than variables and you have found
a solution to this system of equations. Is it possible that your solution is the only one?
Explain.
No. There must then be infinitely many solutions. If the system is Ax = b, then
there are infinitely many solutions to Ax = 0 and so the solutions to Ax = b are a
particular solution to Ax = b added to the solutions to Ax = 0 of which there are
infinitely many.
25. Suppose a system of linear equations has a 24 augmented matrix and the last column
is a pivot column. Could the system of linear equations be consistent? Explain.
No. This would lead to 0 = 1.
26. Suppose the coefficient matrix of a system of n equations with n variables has the
property that every column is a pivot column. Does it follow that the system of
equations must have a solution? If so, must the solution be unique? Explain.
Yes. It has a unique solution.
27. Suppose there is a unique solution to a system of linear equations. What must be true
of the pivot columns in the augmented matrix.
The last one must not be a pivot column and the ones to the left must each be pivot
columns.
28. State whether each of the following sets of data are possible for the matrix equation
Ax = b. If possible, describe the solution set. That is, tell whether there exists a
unique solution no solution or infinitely many solutions.
(a) A is a 5 6 matrix, rank (A) = 4 and rank (A|b) = 4. Hint: This says b is in
the span of four of the columns. Thus the columns are not independent.
Infinite solution set.
(b) A is a 3 4 matrix, rank (A) = 3 and rank (A|b) = 2.
This surely cant happen. If you add in another column, the rank does not get
smaller.
(c) A is a 4 2 matrix, rank (A) = 4 and rank (A|b) = 4. Hint: This says b is in
the span of the columns and the columns must be independent.
You cant have the rank equal 4 if you only have two columns.
(d) A is a 5 5 matrix, rank (A) = 4 and rank (A|b) = 5. Hint: This says b is not
in the span of the columns.
In this case, there is no solution to the system of equations represented by the
augmented matrix.
(e) A is a 4 2 matrix, rank (A) = 2 and rank (A|b) = 2.
In this case, there is a unique solution since the columns of A are independent.
29. Suppose A is an m n matrix in which m n. Suppose also that the rank of A equals
m. Show that A maps Fn onto Fm . Hint: The vectors e1 , , em occur as columns
in the row reduced echelon form for A.
This says that the columns of A have a subset of m vectors which are linearly independent. Therefore, this set of vectors is a basis for Fm . It follows that the span of
the columns is all of Fm . Thus A is onto.
94
Exercises
30. Suppose A is an m n matrix in which m n. Suppose also that the rank of A equals
n. Show that A is one to one. Hint: If not, there exists a vector, x such that Ax = 0,
and this implies at least one column of A is a linear combination of the others. Show
this would require the column rank to be less than n.
The columns are independent. Therefore, A is one to one.
31. Explain why an n n matrix A is both one to one and onto if and only if its rank is
n.
The rank is n is the same as saying the columns are independent which is the same as
saying A is one to one which is the same as saying the columns are a basis. Thus the
span of the columns of A is all of Fn and so A is onto. If A is onto, then the columns
must be linearly independent since otherwise the span of these columns would have
dimension less than n and so the dimension of Fn would be less than n.
32. If you have not done this problem already, here it is again. It is a very important
result. Suppose A is an m n matrix and B is an n p matrix. Show that
dim (ker (AB)) dim (ker (A)) + dim (ker (B)) .
Let {w1 , , wk } be a basis for B (Fp ) ker (A) and suppose {u1 , , ur } is a basis
for ker (B). Let Bzi = wi . Now suppose x ker (AB). Then Bx ker (A) B (Fp )
and so
k
k
X
X
Bx =
a i wi =
ai Bzi
i=1
so
i=1
k
X
i=1
Hence
x
k
X
ai zi ker (B)
ai zi =
i=1
r
X
b j uj
j=1
j=1
This is because B (Fp ) ker (A) ker (A) and these are equal exactly when they have
the same dimension.
95
Exercises
33. Recall that every positive integer can be factored into a product of primes in a unique
way. Show there must be infinitely many primes. Hint: Show that if you have any
finite set of primes and you multiply them and then add 1, the result cannot be
divisible by any of the primes in your finite set. This idea in the hint is due to Euclid
who lived about 300 B.C.
Consider p1 p2 pn + 1. Then if
p1 p2 pn + 1 = lpk
you would have
1
+m=l
pk
where m is an integer, the product of the other primes different than pk . This is a
contradiction. Hence there must be another prime to divide p1 p2 pn + 1 and so
there are infinitely many primes.
34. There are lots of fields. This will give an example of a finite field. Let Z denote the set
of integers. Thus Z = { , 3, 2, 1, 0, 1, 2, 3, }. Also let p be a prime number.
We will say that two integers, a, b are equivalent and write a b if a b is divisible
by p. Thus they are equivalent if a b = px for some integer x. First show that
a a. Next show that if a b then b a. Finally show that if a b and b c
then a c. For a an integer, denote by [a] the set of all integers which is equivalent
to a, the equivalence class of a. Show first that is suffices to consider only [a] for
a = 0, 1, 2, , p 1 and that for 0 a < b p 1, [a] 6= [b]. That is, [a] = [r] where
r {0, 1, 2, , p 1}. Thus there are exactly p of these equivalence classes. Hint:
Recall the Euclidean algorithm. For a > 0, a = mp + r where r < p. Next define the
following operations.
[a] + [b] [a + b]
[a] [b] [ab]
Show these operations are well defined. That is, if [a] = [a ] and [b] = [b ] , then
[a] + [b] = [a ] + [b ] with a similar conclusion holding for multiplication. Thus for
addition you need to verify [a + b] = [a + b ] and for multiplication you need to verify
[ab] = [a b ]. For example, if p = 5 you have [3] = [8] and [2] = [7] . Is [2 3] = [8 7]?
Is [2 + 3] = [8 + 7]? Clearly so in this example because when you subtract, the result
is divisible by 5. So why is this so in general? Now verify that {[0] , [1] , , [p 1]}
with these operations is a Field. This is called the integers modulo a prime and is
written Zp . Since there are infinitely many primes p, it follows there are infinitely
many of these finite fields. Hint: Most of the axioms are easy once you have shown
the operations are well defined. The only two which are tricky are the ones which
give the existence of the additive inverse and the multiplicative inverse. Of these, the
first is not hard. [x] = [x]. Since p is prime, there exist integers x, y such that
1 = px+ky and so 1ky = px which says 1 ky and so [1] = [ky] . Now you finish the
argument. What is the multiplicative identity in this collection of equivalence classes?
Of course you could now consider field extensions based on these fields.
The only substantive issue is why Zp is a field. Let [x] Zp where [x] 6= [0]. Thus x
is not a multiple of p. Then x and p are relatively prime. Hence from Theorem 1.9.3,
there exists a, b such that
1 = ap + bx
Then
[1 bx] = [ap] = 0
96
Exercises
and it follows that
[b] [x] = [1]
so [b] = [x]
=
0 0
1 0
1 0
0 0
0
0
1
Thus the identity is a comutator. Compare this with Problem 50 on Page 198.
1 0
1 0
This is easy. The left side equals
=
0 1
0 1
36. Suppose V is a vector space with field of scalars F. Let T L (V, W ) , the space of
linear transformations mapping V onto W where W is another vector space. Define
an equivalence relation on V as follows. v w means v w ker (T ) . Recall that
ker (T ) {v : T v = 0}. Show this is an equivalence relation. Now for [v] an equivalence class define T [v] T v. Show this is well defined. Also show that with the
operations
[v] + [w]
[v]
[v + w]
[v]
this set of equivalence classes, denoted by V / ker (T ) is a vector space. Show next that
T : V / ker (T ) W is one to one, linear, and onto. This new vector space, V / ker (T )
is called a quotient space. Show its dimension equals the difference between the
dimension of V and the dimension of ker (T ).
It is obviously an equivalence relation. Is the operation of addition well defined? If
[v] = [v ] , similar for w, w , is it the case that [v + w] = [v + w ]? Is
v + w (v + w ) ker (T )?
Of course so. It is just the sum of two things in ker (T ) . Similarly the operation of
scalar multiplication is well defined. Is T well defined? If v w, is it true that
T v =T w? Of course so.
T (v w) = 0
37. Let V be an n dimensional vector space and let W be a subspace. Generalize the
above problem to define and give properties of V /W . What is its dimension? What
is a basis?
Here you define an equivalence relation by v u if v u W. This is clearly
an equivalence relation. Define [v] + [u] [u + v] and [u] [u]. Then by a
repeat of the above, everything is well defined. It is clear that V /W , the set of
equivalence classes is a vector space with these operations as just defined. What is
its dimension? Let {w1 , , wr } be a basis for W . Now extend to get a basis for
V, {w1 , , wr , v1 , , vnr }. Then if
nr
X
ci [vi ] = [0] ,
i=1
97
Exercises
it follows that
Pnr
i=1
di wi +
i=1
nr
X
ci vi = 0
i=1
Now, since {w1 , , wr , v1 , , vnr } is a basis, it follows that all the di and ci are
equal to 0. Therefore, in particular the vectors [vi ] form a linearly independent set in
V /W. Do these vectors span V /W ? Suppose [v] V /W . Then
v=
r
X
di wi +
i=1
nr
X
ci vi
i=1
nr
X
ci [vi ]
i=1
40. As mentioned, for distinct algebraic numbers i , the complex numbers {ei }ni=1 are
linearly independent over the field of scalars A where A denotes the algebraic numbers,
those which are roots of a polynomial having integer (rational) coefficients. What is
the dimension of the vector space C with field of scalars A, finite or infinite? If the
field of scalars were C instead of A, would this change? What if the field of scalars
were R?
How many distinct algebraic numbers can you get? Clearly as many as desired. Just
consider the nth roots of some complex number. If you have n distinct algebraic
n
numbers i , then from the Lindemann Weierstrass theorem mentioned above, {ei }i=1
is linearly independent. Of course if the field of scalars were C this would change
completely. Then the dimension of C over C is clearly 1.
41. Suppose F is a countable field and let A be the algebraic numbers, those numbers
which are roots of a polynomial having coefficients in F which are in G, some other
field containing F. Show A is also countable.
You consider each polynomial of degree n which has coefficients in the countable field
F. There are countably many of these. Each has finitely many roots in A. Therefore,
98
Exercises
there are countably many things in A which correspond to polynomials of degree n
The totality of A is obtained then from taking a countable union of countable sets,
and this is countable.
p (x)
, degree of p (x) < degree of denominator.
q1 (x) qm (x)
where the polynomials qi (x) are relatively prime and all the polynomials p (x) and
qi (x) have coefficients in a field of scalars F. Thus there exist polynomials ai (x)
having coefficients in F such that
1=
m
X
ai (x) qi (x)
i=1
Explain why
P
m
X
p (x) m
a (x) p (x)
i=1 ai (x) qi (x)
Qi
R (x) =
=
q1 (x) qm (x)
j6=i qj (x)
i=1
Pm
p (x) i=1 ai (x) qi (x)
p (x)
R (x) =
=
q1 (x) qm (x)
q1 (x) qm (x)
Then simplifying this yields
R (x) =
m
X
a (x) p (x)
Qi
j6=i qj (x)
i=1
Now continue doing this on each term in the above sum till finally you obtain an
expression of the form
m
X
bi (x)
i=1
qi (x)
Using the Euclidean algorithm for polynomials, explain why the above is of the form
M (x) +
m
X
ri (x)
i=1
qi (x)
where the degree of each ri (x) is less than the degree of qi (x) and M (x) is a polynomial.
You just note that
bi (x) = li (x) qi (x) + ri (x)
where the degree of ri (x) is less than the degree of qi (x) . Then replacing each bi (x)
by this expression, you obtain the above expression.
Now argue that M (x) = 0.
You have
M (x) +
m
X
ri (x)
i=1
qi (x)
p (x)
q1 (x) qm (x)
99
Exercises
Now multiply both sides by q1 (x) qm (x) . If M (x) is not zero, then you would have
the degree of the left is larger than the degree on the right in the equation which
results. Hence M (x) = 0.
From this explain why the usual partial fractions expansion of calculus must be true.
You can use the fact that every polynomial having real coefficients factors into a product of irreducible quadratic polynomials and linear polynomials having real coefficients.
This follows from the fundamental theorem of algebra in the appendix.
n
What if some qi (x) = ki (x) ? In this case, you can go further. You have the partial
fractions expansion described above and you have a term of the form
a (x)
n
n , degree of a (x) < degree k (x)
k (x)
n
X
ai (x)
i=1
k (x)
where the degree of ai (x) < degree of k (x). This can be proved by induction. If
n = 1, there is nothing to show. Suppose then it is true for n 1.
!
a (x)
1
a (x)
=
k (x)n
k (x) k (x)n1
n1
n1
a (x)
b (x)
r (x)
1
+
n =
k (x) k (x)n1 k (x)
k (x)
The second term gives what is desired by induction. I claim that the degree of b (x)
n
must be less than the degree of k (x) . If not, you could multiply both sides by k (x)
and obtain
a (x) = b (x) k (x)n1 + r (x)
n
and the degree of the left is less than the degree of k (x) while the degree of the
n
right is greater than the degree of k (x) . Therefore, the desired decomposition has
been obtained. Now the usual procedure for partial fractions follows. If you have
a rational function p (x) /q (x) where the degree of q (x) is larger than the degree
of p (x) , there exist factors of q (x) into products of linear and irreducible over R
quadratic factors by using the fundamental theorem of algebra. Then you just apply
the above decomposition result.
43. Suppose {f1 , , fn } is an independent set of smooth functions defined on some interval (a, b). Now let A be an invertible n n matrix. Define new functions {g1 , , gn }
as follows.
g1
f1
..
.
. = A ..
gn
fn
100
Exercises
Is it the case that {g1 , , gn } is also independent? Explain why.
P
Suppose ni=1 ai gi = 0. Then
0=
X
i
ai
Aij fj =
It follows that
X
j
fj
Aij ai
Aij ai = 0
for each j. Therefore, since AT is invertible, it follows that each ai = 0. Hence the
functions gi are linearly independent.
F.26
Exercises
9.5
1. If A, B, and C are each n n matrices and ABC is invertible, why are each of A, B,
and C invertible?
This is because ABC is one to one.
2. Give an example of a 3 2 matrix with the property that the linear transformation
determined by this matrix is one to one but not onto.
1 0
0 1
0 0
Since the matrix (A I) is not invertible, it follows that there exists x which it sends
to 0.
5. How does the minimal polynomial of an algebraic number relate to the minimal polynomial of a linear transformation? Can an algebraic number be thought of as a linear
transformation? How?
They are pretty much the same thing. If you have an algebraic number it can
be thought of as a linear transformation which maps A to A according to the rule
() 6= .
6. Recall the fact from algebra that if p () and q () are polynomials, then there exists
l () , a polynomial such that
q () = p () l () + r ()
where the degree of r () is less than the degree of p () or else r () = 0. With this in
mind, why must the minimal polynomial always divide the characteristic polynomial?
That is, why does there always exist a polynomial l () such that p () l () = q ()?
Can you give conditions which imply the minimal polynomial equals the characteristic
polynomial? Go ahead and use the Cayley Hamilton theorem.
101
Exercises
0
2
0
2
0
2
1
1
1
1
1
1
2
+1
,1
=
,
A
2
0
0
2
0
2
1 1
A=
2 1
8. Let = {u1 , , un } be a basis for Fn and let T : Fn Fn be defined as follows.
!
n
n
X
X
T
a k uk =
a k b k uk
k=1
k=1
First show that T is a linear transformation. Next show that the matrix of T with
respect to this basis is [T ] =
b1
..
.
bn
Show that the above definition is equivalent to simply specifying T on the basis vectors
of by
T (uk ) = bk uk .
Formally,
(b1 u1 , b2 u2 , , bn un ) = (u1 , u2 , , un ) [T ]
102
Exercises
9. In the situation of the above problem, let = {e1 , , en } be the standard basis for
Fn where ek is the vector which has 1 in the k th entry and zeros elsewhere. Show that
[T ] =
1
u1 un [T ] u1 un
(6.36)
We have T ui = bi ui and so we know [T ] . The problem is to find [T ] where is the
good basis. Consider the diagram
Fn
Fn
Fn
[T ]
[T ]
Fn
Fn
Fn
1
u1 un
On the left side going down you get x x
x.
Going up on the
right, you need to have x u1 un x u1 un x. Then you need
[T ] =
u1
un
[T ]
u1
un
1
10. Generalize the above problem to the situation where T is given by specifying its
action on the vectors of a basis = {u1 , , un } as follows.
T uk =
n
X
ajk uj .
j=1
Letting A = (aij ) , verify that for = {e1 , , en } , 6.36 still holds and that [T ] = A.
This isnt really any different. As explained earlier, the matrix of T with respect to
the given basis is just A where it is as described. Thus A = [T ] and by a repeat of
the above argument,
1
[T ] = u1 un [T ] u1 un
11. Let P3 denote the set of real polynomials of degree no more than 3, defined on an
interval [a, b]. Show that P3 is a subspace of the vector space
of all functions defined
on this interval. Show that a basis for P3 is 1, x, x2 , x3 . Now let D denote the
differentiation operator which sends a function to its derivative. Show D is a linear
transformation which sends P3 to P3 . Find the matrix of this linear transformation
with respect to the given basis.
0, 1, 2x, 3x2 = 1, x, x2 , x3 A where A is the desired matrix. Then clearly
0 1 0 0
0 0 2 0
A=
0 0 0 3
0 0 0 0
12. Generalize the above problem to Pn , the space of polynomials of degree no more than
n with basis {1, x, , xn } .
The pattern seems pretty clear. A will be an (n + 1) (n + 1) matrix which has zeros
everywhere except on the super diagonal where you have, starting at the top and going
towards the bottom, the numbers in the following order. 1, 2, 3, , n.
103
Exercises
13. In the situation of the above problem, let the linear transformation be T = D2 + 1,
defined as T f = f + f. Find the matrix of this linear transformation with respect to
the given basis {1, x, , xn }.
First consider the operator D2 .
0 0 2 6x 12x2 =
x x2
0
0
A=
0
0
0
Then the matrix will be this one added
1 0
0 1
0 0
0 0
0 0
x3
x4
would have
0
0
0
0
0
2
0
0
0
0
0
6
0
0
0
0
0
12
0
0
2 0 0
0 6 0
1 0 12
0 1 0
0 0 1
14. In calculus, the following situation is encountered. There exists a vector valued function f :U Rm where U is an open subset of Rn . Such a function is said to have
a derivative or to be differentiable at x U if there exists a linear transformation
T : Rn Rm such that
lim
v0
|f (x + v) f (x) T v|
= 0.
|v|
First show that this linear transformation, if it exists, must be unique. Next show
that for = {e1 , , en } , , the standard basis, the k th column of [T ] is
f
(x) .
xk
Actually, the result of this problem is a well kept secret. People typically dont see
this in calculus. It is seen for the first time in advanced calculus if then.
Suppose that T also works. Then letting t > 0,
|f (x + tv) f (x) tT v|
|f (x + tv) f (x) tT v|
= lim
=0
t0
t0
t
t
lim
In particular,
t (T v T v)
=0
t0
t
which says that T v = T v and since v is arbitrary, this shows T = T .
lim
15. Recall that A is similar to B if there exists a matrix P such that A = P 1 BP. Show
that if A and B are similar, then they have the same determinant. Give an example
of two matrices which are not similar but have the same determinant.
These two are not similar but have the same determinant.
1 0
1 1
,
0 1
0 1
104
Exercises
You can see these are not similar by noticing that the second has an eigenspace of
dimension equal to 1 so it is not similar to any diagonal matrix which is what the first
one is.
16. Suppose A L (V, W ) where dim (V ) > dim (W ) . Show ker (A) 6= {0}. That is, show
there exist nonzero vectors v V such that Av = 0.
If not, then A would be one to one and so it would take a basis to a linearly independent
set which is impossible because W has smaller dimension than V .
17. A vector v is in the convex hull of a nonempty set S if there are finitely many vectors
of S, {v1 , , vm } and nonnegative scalars {t1 , , tm } such that
v=
m
X
tk vk ,
k=1
m
X
tk = 1.
k=1
m1
X
tk vk .
k=1
Thus every vector in the convex hull of S can be obtained as a convex combination
of at most n + 1 points of S. This incredible result is in Rudin [23]. Hint: Consider
L : Rm V R defined by
!
m
m
X
X
L (a)
ak vk ,
ak
k=1
k=1
Explain why ker (L) 6= {0} . Next, letting a ker (L) \ {0} and R, note that
a ker (L) . Thus for all R,
v=
m
X
(tk + ak ) vk .
k=1
If any of the tk = 0, there is nothing to prove. Hence you can assume each of these
tk > 0. Suppose then that m > n + 1 and consider
!
m
m
X
X
L (a)
ak vk ,
ak
k=1
k=1
m
X
(tk + ak ) vk
k=1
P
P
P
(tk + ak ) = 1 because
ak = 0. Also, for = 0,
because m
k=1 ak vk = 0. Now
each of the tk + ak > 0. Adjust so that each of these is still 0 but one vanishes.
Then you have gotten v as a convex combination of fewer than m vectors.
105
Exercises
18. For those who know about compactness, use Problem 17 to show that if S Rn and
S is compact, then so is its convex hull.
Suppose that the convex hull of S is not compact. Then nthere exists
o
Pn+1
v1k , v2k , , v(n+1)k k=1 , each in S and sk C s : i=1 si = 1, si [0, 1]
such that
(n+1
)
X
sik vik
i=1
k=1
20. Let A be an n n matrix of elements of F. There are two cases. In the first case,
F contains a splitting field of pA () so that p () factors into a product of linear
polynomials having coefficients in F. It is the second case which of interest here where
pA () does not factor into linear factors having coefficients in F. Let G be a splitting
field of pA () and let qA () be the minimal polynomial of A with respect to the field
G. Explain why qA () must divide pA (). Now why must qA () factor completely
into linear factors?
In the field G
pA () = qA () l () + r ()
where r () either equals 0 or has degree less than the degree of qA (). Then it
follows that r (A) = 0 and so it must be the case that r () = 0. Hence qA ()
must divide pA (). Now G was the splitting field of pA () and this new polynomial
divides this one. Therefore, this new polynomial also must factor completely into
linear polynomials having coefficients in G.
pA () = qA () l ()
and all roots of pA () are in G so the same is true of the roots of qA () because these
are all roots of pA ().
21. In Lemma 9.2.2 verify that L is linear.
106
Exercises
L a
= L
n
X
i vi + b
i=1
n
X
n
X
(ai + b i ) vi
i=1
n
X
= a
i Lvi + b
i=1
n
X
i Lvi
i=1
n
X
i=1
F.27
(ai + b i ) Lvi
i=1
n
X
= aL
i vi
i=1
i vi
+ bL
n
X
i vi
i=1
Exercises
10.6
1. In the discussion of Nilpotent transformations, it was asserted that if two nn matrices
A, B are similar, then Ak is also similar to B k . Why is this so? If two matrices are
similar, why must they have the same rank?
Say A = S 1 BS. Then Ak = S 1 B k S and so these are also similar. You can write
S 1 BSS 1 BSS 1 BS S 1 BS
and note that the inside SS 1 cancels. Therefore, the result of the above is as claimed.
Hence Ak is similar to B k .
Now the dimension of A (Fn ) is the dimension of S 1 BS (Fn ) which is the same as the
dimension of S 1 B (Fn ) which is the same as the dimension of B (Fn ) because S, S 1
are one to one and onto. Hence they each take a basis to a basis.
2. If A, B are both invertible, then they are both row equivalent to the identity matrix.
Are they necessarily similar? Explain.
1 1
1 0
Obviously not. Consider
,
. These are both in Jordan form.
0 1
0 1
3. Suppose you have two nilpotent matrices A, B and Ak and B k both have the same
rank for all k. Does it follow that A, B are similar? What if it is not known that A, B
are nilpotent? Does it follow then?
Yes, this is so. You have A = S 1 JS and B = T 1 J T where J, J are both Jordan
canonical form. Then you have that J = SAS 1 and J = T BT 1 and so J k and J k
have the same rank for all k. Therefore, J = J because these matrices are nilpotent.
It follows that
SAS 1 = T BT 1
and consequently,
"
1
B = T 1 SAS 1 T = S 1 T
AS 1 T
so that A is similar to B. In case they are not nilpotent, the above problem gives a
counter example.
107
Exercises
4. When we say a polynomial equals zero, we mean that all the coefficients equal 0. If
we assign a different meaning to it which says that a polynomial
p () =
n
X
ak k = 0,
k=0
when the value of the polynomial equals zero whenever a particular value of F
is placed in the formula for p () , can the same conclusion be drawn? Is there any
difference in the two definitions for ordinary fields like Q? Hint: Consider Z2 , the
integers mod 2.
This sort of thing where it equals 0 if it sends everything to 0 works fine when we
think of polynomials as functions. However, consider 2 + where the coefficients
come from Z2 . It sends everything to 0 but we dont want to consider this polynomial
to be the zero polynomial because its coefficients are not all 0.
5. Let A L (V, V ) where V is a finite dimensional vector space with field of scalars F.
Let p () be the minimal polynomial and suppose () is any nonzero polynomial such
that (A) is not one to one and () has smallest possible degree such that (A) is
nonzero and not one to one. Show () must divide p ().
Say (A) x = 0 where x 6= 0. Consider all monic polynomials () such that (A) x = 0.
One such is the minimal polynomial p (). This condition is less severe than saying
that (A) = 0. Therefore, there are more polynomials considered. If you take the one
of least degree, its degree must be no larger than the degree of p () . By assumption
there is a polynomial () for which (A) 6= 0 but such that (A) x = 0. You can do
the usual thing.
p () = l () () + r ()
where r () = 0 or else it has smaller degree than () . Then
0 = l (A) (A) x + r (A) x
It follows that r (A) x = 0. If r () 6= 0 this would be a contradiction to the definition
of (). Therefore, () divides p () .
6. Let A L (V, V ) where V is a finite dimensional vector space with field of scalars F.
Let p () be the minimal polynomial and suppose () is an irreducible polynomial
with the property that (A) x = 0 for some specific x 6= 0. Show that () must
divide p () . Hint: First write
p () = () g () + r ()
where r () is either 0 or has degree smaller than the degree of (). If r () = 0 you
are done. Suppose it is not 0. Let () be the monic polynomial of smallest degree
with the property that (A) x = 0. Now use the Euclidean algorithm to divide ()
by () . Contradict the irreducibility of () .
Say (A) x = 0 where x 6= 0. Consider all monic polynomials () such that (A) x =
0. One such is the minimal polynomial p (). This condition is less severe than saying
that (A) = 0. Therefore, there are more polynomials considered. If you take the one
of least degree, its degree must be no larger than the degree of p () . You can do
the usual thing.
p () = g () () + r ()
108
Exercises
where r () = 0 or else it has smaller degree than () . Then
0 = g (A) (A) x + r (A) x
It follows that r (A) x = 0. If r () 6= 0 this would be a contradiction to the definition
of (). Therefore, () divides p () . Also
() = () l () + r ()
where the degree of r () is less than the degree of () or else r () = 0. If r () 6= 0,
then r (A) x = 0 and it would have degree too small. Hence r () = 0 and so ()
must divide () . But this requires that the two must be scalar multiples of each
other. Hence () must also divide p ().
q
Y
j ()nj
j=1
where the j () are irreducible. Explain using Corollary 8.3.11 why the irreducible
factors of the minimal polynomial are j () and why the minimal polynomial is of
the form
q
Y
r
j () j
j=1
where rj nj . You can use the Cayley Hamilton theorem if you like.
This follows right away from the Cayley Hamilton theorem and that corollary. Since
the minimal polynomial divides the characteristic polynomial, it must be of the form
just mentioned.
8. Let
1 0
A= 0 0
0 1
0
1
0
You know that this satisfies a polynomial of degree no more than 3. Consider
2
3
1 0 0
1 0 0
1 0 0
1 0 0
0 1 0 , 0 0 1 , 0 0 1 , 0 0 1 These are
0 0 1
0 1 0
0 1 0
0 1 0
1 0 0
1 0 0
1 0
0
1 0 0
0 1 0 , 0 0 1 , 0 1 0 , 0 0 1
0 0 1
0 1 0
0 0 1
0 1 0
Thus it is desired to find a linear combination of these which equals 0 and out of all
of them to pick the one using as few of them as possible starting with the left side.
An easy way to do this is to use the row reduced echelon from. Write them as column
vectors and then do row operations.
109
Exercises
1
0
0
0
1
0
0
0
1
1
1
1
1 1
1
1
1
0
0
0
0 0
0
0
0
0
0
0
0 0
0
0
0
0
0
0
0 0
0
0
0
0 1 0 , 1 0 1 0 , row echelon form: 0
1 0
1
0 1 0
1
0
0
0
0
0 0
0
0
0
1
0 1
0 1
0 1
0
0 1 0
1 0 1 0
0
0
1
0
0
0
0
0
0
0
0
0
1
0
0
0
0
0
0
1
1
1
0
0
0
0
0
0
1
0
0
3
0 0
0 1
1 0
0 0
1 0
0 1 0 1
1 0
0 0
1
0
0
2
0
1 +
0
0
0 0 0
0 = 0 0 0
1
0 0 0
1
0
0
0
0
1
9. SupposeA is an n n matrix
and let v be a vector. Consider the A cyclic set of
vectors v, Av, , Am1 v where this is an independent set of vectors but Am v is
a linear combination of the preceding vectors in the list. Show how to obtain a monic
polynomial of smallest degree, m, v () such that
v (A) v = 0
Now let {w1 , , wn } be a basis and let () be the least common multiple of the
wk () . Explain why this must be the minimal polynomial of A. Give a reasonably
easy algorithm for computing v ().
You could simply form the cyclic set of vectors described, make them the columns of
a matrix and then do the row reduced echelon form to reveal linear relations. This is
how you can compute v (). Then, having found this, for each vector in a basis, and
letting () be the least common multiple, it follows that (A) v = 0 for each v in a
basis and so (A) = 0. Also, if (A) = 0, then () must be divisible by each of the
v () for v in the given basis. Here is why.
() = l () v () + r () , degree of r () < degree of v () or r () = 0
Then if r () 6= 0, you could contradict the fact that v () has smallest degree out of
all those () for which (A) v = 0. Therefore, () is divisible by each v () for
each v in the given basis. The polynomial of smallest such degree is by definition the
least common multiple.
10. Here is a matrix.
7 1
21 3
70
10
1
3
10
Using the process of Problem 9 find the minimal polynomial of this matrix. It turns
out the characteristic polynomial is 3 .
110
Exercises
1 0 0
1
7 1 1
1
0 0 0
1
0 1 0 0 21 3 3 0 0 0 0 0
0 0 1
0
70
10 10
0
0 0 0
0
1
7
0
1 7 0
1 0 0
0 21 0 , 0 21 0 , row echelon form: 0 1 0 It fol0
70
0
0 70 0
0 0 0
2
lows that e1 () , = 0 + 0. The same sort of thing will happen with e2 , e3 .
Therefore, the minimal polynomial is just 2 . Note that
7 1 1
0 0 0
21 3 3 = 0 0 0
70
10 10
0 0 0
11. Find the minimal polynomial for
1 2 3
A= 2 1 4
3 2 1
by the above technique. Is what you found also the characteristic polynomial?
First here
1
0
0
1 1
0 2
0 3
1
4
26
2 8 24
3
2
6
4 26
1 0 0 14
8 24 , row echelon form: 0 1 0
0
2 6
0 0 1
3
e1 () : 14 + 32 = 3
The minimal polynomial is then 3 32 +14. We know it cant have any larger degree
and so this must be it.
2
1 2 3
1 2 3
2 1 4 3 2 1 4
3 2 1
3 2 1
0
1 2 3
0 0
+14 2 1 4 = 0 0
3 2 1
0 0
0
0
0
This is also the characteristic equation. Note that no computation involving determinants was required.
12. Let A be an n n matrix with field of scalars C. Letting be an eigenvalue, show
the dimension of the eigenspace equals the number of Jordan blocks in the Jordan
canonical form which are associated with . Recall the eigenspace is ker (I A) .
Let the blocks associated with in the Jordan form be
J1 ()
J2 ()
..
.
Jr ()
111
Exercises
Thus this is an upper triangular matrix which has down the main diagonal. If you
take away I from it, you get a matrix of the form
N1 ()
N2 ()
..
.
Nr ()
where
0
Ni () =
0
..
..
1
0
How many columns of zeros will there be? Exactly r where r is the number of those
blocks. You get one at the beginning of each block. Hence there will be exactly r free
variables and so the dimension of the eigenspace will equal to the number of blocks.
13. For any n n matrix, why is the dimension of the eigenspace always less than or
equal to the algebraic multiplicity of the eigenvalue as a root of the characteristic
equation? Hint: Note the algebraic multiplicity is the size of the appropriate block
in the Jordan form.
This is obvious from the above problem. The dimension of the eigenspace equals the
number of blocks which must be no larger than the length of the diagonal.
14. Give an example of two nilpotent
minimal polynomial if possible.
2
0 1 0 0
0 0 0
0 0 0 0
0 0 0
0 0 0 0 = 0 0 0
0 0 0 0
0 0 0
2
0 1 0 0
0 0 0
0 0 0 0
0 0 0
0 0 0 1 = 0 0 0
0 0 0 0
0 0 0
These are two which work.
0
0
,
0
0
0
0
0
0
15. Use the existence of the Jordan canonical form for a linear transformation whose
minimal polynomial factors completely to give a proof of the Cayley Hamilton theorem
which is valid for any field of scalars. Hint: First assume the minimal polynomial
factors completely into linear factors. If this does not happen, consider a splitting field
of the minimal polynomial. Then consider the minimal polynomial with respect to
this larger field. How will the two minimal polynomials be related? Show the minimal
polynomial always divides the characteristic polynomial.
First suppose the minimal polynomial factors. Then the linear transformation has a
Jordan form.
J (1 )
..
J =
.
J (r )
112
Exercises
where each block is as described in the description of the Jordan form. Say J (i )
is ri ri in size. Then the characteristic polynomial is the determinant of I J a
matrix of the form
( 1 ) I + N (1 )
..
.
( r ) I + N (r )
where the blocks N (i ) have some 1s on the super diagonal and zeros elsewhere.
Thus the characteristic polynomial is the determinant of this matrix which is
q () =
r
Y
i=1
r
Y
i=1
J (1 i )
..
ri
( i )
ri
(J i I)
J (0)
..
.
J (r i )
r
J (1 i ) i
..
..
.
ri
J (r i )
Qr
ri
Therefore, the product i=1 (J i I) = 0 from block multiplication.
ri
is of the
Now in the general case where the minimal polynomial pF () does not necessarily
factor, consider a splitting field for the minimal polynomial. The new minimal polynomial divides the old one. Therefore, the new minimal polynomial with respect to
this splitting field G can be completely factored. Hence the Jordan form exists with
respect to this new splitting field and new minimal polynomial pG () . Then from the
above, it follows that pG (J) = 0. However, pG () divides pF () and so it is also the
case that pF (J) = 0. Since J is a matrix of T with respect to a suitable basis, it
follows that pF (T ) = 0 also, where T is the original linear transformation.
16. Here is a matrix. Find its Jordan canonical form by directly finding the eigenvectors
and generalized eigenvectors based on these to find a basis which will yield the Jordan
form. The eigenvalues are 1 and 2.
3 2 5 3
1 0 1 2
4 3 6 4
1 1 1 3
113
Exercises
Why is
3
1
4
1
2 5 3
4
1
0 1 2
1
, eigenvectors: 1, 0 2
3
1
3 6 4
1 1 3
1
0
3 2 5 3
1 0 0 0
x1
4
1 0 1 2 0 1 0 0 x2 1
4 3 6 4 0 0 1 0 x3 = 3
1 1 1 3
0 0 0 1
x4
1
4t4 4
4
x1
4 2 5 3
t4 + 1
1 1 1 2 x2 1
3 2
1 0
4 3
1 1
5 2 5
1 2 1
4 3 4
1 1 1
5
1
6
1
3
2
4
1
3
1 0
0 1
2
2
0 0
4
3
0 0
x1
x2
x3 =
x4
A generalized eigenvector is
4
1
2
0
0
x1
x2
0
0 x3
1
x4
1
0
=
1
0
1
t3
1
0
, Solution is:
t3
1
0
1
0
0
1
0
4
4
1
0
1 1 0 1
3 2 1 0 ,
1
0
0
1
0
1
0
1
4 4 1
1 1 0
3 2 1
1 0 0
4 4 1
1 1 0
S=
3 2 1
1 0 0
0
1
0
1
0
1
0
1
Now to get the Jordan form you take S 1 AS, but of course we know the Jordan form
at this point. There will be two blocks, one for 1 and one for 2. Nevertheless, here it
114
Exercises
is
4
1
3
1
4
1
2
0
1
0
3
1
1
0 4
1
1
1
0
=
0
0
1
0
1
0
2
0
3
1
1
1
0
0
5
1
6
1
0
0
2
0
4
3
4 4 1
1 1 0
3 2 1
1 0 0
0
0
1
2
0
1
0
1
17. People like to consider the solutions of first order linear systems of equations which
are of the form
x (t) = Ax (t)
where here A is an n n matrix. From the theorem on the Jordan canonical form,
there exist S and S 1 such that A = SJS 1 where J is a Jordan form. Define
y (t) S 1 x (t) . Show y = Jy.
x = SJS 1 x
and so
S 1 x = J S 1 x , y = Jy
Now suppose (t) is an n n matrix whose columns are solutions of the above
differential equation. Thus
= A
Now let be defined by SS 1 = . Show
= J.
You have
= S 1 S
and so
S 1 S = S 1 AS = S 1 ASS 1 S
det () =
n
X
det i
i=1
115
Exercises
where i has the ith row differentiated and the other rows of left unchanged. Thus,
letting i denote the ith row of , you have that the j th entry of i is given by
X
ij (t) =
Jik kj = i ij + ai (i+1)j
k
Thus
where ai either is 0 or 1. Then since the determinant of a matrix having two equal
rows equals 0, it follows that
(det ()) =
n
X
i=1
and so
det () = Cetrace(J)t
19. Let A be an n n matrix and let J be its Jordan canonical form. Recall J is a block
diagonal matrix having blocks Jk () down the diagonal. Each of these blocks is of
the form
1
0
..
Jk () =
..
. 1
0
1
0
D =
.
..
k1
0
Show that D1 Jk () D has the same form as Jk () but instead of ones down the
super diagonal, there is down the super diagonal. That is Jk () is replaced with
..
..
.
0
Now show that for A an n n matrix, it is similar to one which is just like the Jordan
canonical form except instead of the blocks having 1 down the super diagonal, it has
.
That D1 Jk () D is of the
1
0
0
0 1
0
0
0
2
0
0
0
right form is
0
0
0
0 0
0
3
1 0 0
1 0 0
0 0
1 0
0 1 0 0 2
0 0
0 0 0
the 4 4 case.
0
0
0
3
116
Exercises
1
0
0 1
=
0
0
0
0
0
0
2
0
0
0
0
0 0
3
0 0
0 0
0 0
0 0
0 0 0
J1
0
..
.
Jm
0
2
2
0
0
0
3
3
D1
0
..
.
0
Dm
and on the left by the inverse of this matrix. Using block mutiplication you get the
desired modification. Note that this shows that every matrix in n n matrix having
entries in C is similar to one which is close to a diagonal matrix.
20. Let A be in L (V, V ) and suppose that Ap x 6= 0 for some x 6= 0. Show that Ap ek 6= 0
for some ek {e1 , , en } , a basis for V . If you have a matrix which is nilpotent,
(Am = 0 for some m) will it always be possible to find its Jordan form? Describe how
to do it if this is the case. Hint: First explain why all the eigenvalues are 0. Then
consider the way the Jordan form for nilpotent transformations was constructed in the
above.
Yes, it is possible to compute the Jordan form. In fact, the proof of existence tells how
to do it. A nilpotent matrix has all eigenvalues equal to 0. This is because if Ax =x,
then Ak x = k x and so for large k, you get k x = 0. Finding the eigenvalues is always
the difficulty.
21. Suppose A is an n n matrix and that it has n distinct eigenvalues. How do the minimal polynomial and characteristic polynomials compare? Determine other conditions
based on the Jordan Canonical form which will cause the minimal and characteristic
polynomials to be different.
If the eigenvalues are distinct, then the two polynomials are obviously the same. More
generally, A has a Jordan form
J (1 )
..
.
J (r )
117
Exercises
in the characteristic equation. If J (k ) is of the form
k 1
..
.
k
.
.
. 1
k
then J i I is of the form
J (1 i )
..
.
J (0)
..
.
J (r i )
where J (0) i = 0 and no smaller exponent will work. Therefore, the characteristic
polynomial and the minimal polynomial will coincide. However, if J (k ) is composed
of many blocks, the extreme case being when it is a diagonal matrix having k down
the diagonal, a smaller exponsne will work and so there will be a difference between
the characteristic and minimal polynomials.
For example, consider the minimal polynomial for the matrix
0 1 0 0
0 0 0 0
N =
0 0 0 1
0 0 0 0
a 1 0
a
0 a 0 or 0
0 0 b
0
0
0
b
In the first case, the two polynomials are the same. The characteristic polynomial is
( a)2 ( b)
In the first case
a 1
0 a
0 0
0
a
0 0
b
0
0
a
0
0
a
0 0
a
0
1
a
0
0
b
0 0
b
0
0 0
b 0
0 b
118
Exercises
0
= 0
0
ab
0
0
0
0 6= 0
0
a 0 0
a 0 0
a
0 a 0 0 a 0 0
0 0 b
0 0 a
0
0
= 0
0
0
a
0
0 0
0 0
0 0
0
b
0 0
b
0
0 0
b 0
0 b
24. In proving the uniqueness of the Jordan canonical form, it was asserted that if two
n n matrices A, B are similar, then they have the same minimal polynomial and also
that if this minimal polynomial is of the form
p () =
s
Y
ri
i ()
i=1
where the i () are irreducible, then ker (i (A)ri ) has the same dimension as ker (i (B)ri ) .
Why is this so? This was what was responsible for the blocks corresponding to an
eigenvalue being of the same size.
Since the two are similar, they come from the same linear transformation T . Therefore,
this is pretty obvious. Also, if B = S 1 AS, then
ri
r
r
ker (i (B) i ) = ker i S 1 AS
= ker S 1 (i (A) i ) S
r
25. Show that a given matrix is non defective (diagonalizable) if and only if the minimal
polynomial has no repeated roots.
If the matrix is non defective, then its Jordan form is
J (1 )
..
.
J (r )
119
Exercises
Conversely, suppose the minimal polynomial is of this form. Also suppose that some
J (i ) say J (i ) is not diagonal. Then consider
Y
(J i I)
i
J1 (1 i )
..
.
Ji (0)
..
.
Jr (r i )
where for some i, Ji (0) is not the zero matrix. It follows that the above product
cannot be zero either. In fact, you could let x Fri be such that J (0) x = y 6= 0.
Then
0
0
.. ..
. .
Y
(J i I)
x = z , z 6= 0
. .
i
.. ..
0
0
because Ji (i k ) is one to one on Fri whenever k 6= i. It follows that each of the
J (k ) must be a diagonal matrix.
the i being distinct. Thus the minimal polynomial would have no repeated roots and
so the matrix is non defective by the above problem.
120
Exercises
27. In Theorem 10.3.4 show that the cyclic sets can be arranged in such a way that the
length of xi+1 divides the length of xi .
You make this condition a part of the induction hypothesis. Then at the end when
m1
you extend to (A)
(V) ker ((A)) , you note that, thanks to Lemma 10.3.2,
m1
each of the yi for yi ker (A)
has length equal to a multiple of d, the degree
m1
m1
of () while the x for x (A)
(V ) \ ker (A)
each have length equal to
d thanks to the assumption that () is irreducible over the field of scalars and each
of these is in ker ( (A)).
28. Show that if A is a complex n n matrix, then
a Jordan block. Note that
0 0 1
1 0
0
0 1 0 0 1 0
1 0 0
0 0
1
0
0
Such a backwards identity will work for any size Jordan block. Therefore, J is similar
to J T because you can use a similarity matrix which is block diagonal, each block
being appropriate for producing the transpose of the corresponding Jordan block.
Thus there is a matrix S such that
J = S 1 J T S
Now also J = P 1 AP for a suitable P. Therefore,
P 1 AP = S 1 P 1 AP
and so
A =
=
T
1
P S 1 P T AT P T
SP 1
1
1
1
PT
SP 1
AT P T
SP 1
29. Let A be a linear transformation defined on a finite dimensional vector space V . Let
the minimal polynomial be
q
Y
m
i () i
i=1
o
n
be the cyclic sets such that ivi , , ivri
is a basis for
1
1
i P P
i
mi
i
ker (i (A) ). Let v = i j vj . Now let q () be any polynomial and suppose that
and let
ivi , , ivri
q (A) v = 0
n1
X
ai Ai x.
i=0
121
Exercises
Then
(A) v =
n1
X
ai Ai (A) x = 0
i=0
in the sum equals 0 and so from the first part of the argument q (A) restricted to
ker (i (A)mi ) equals 0. Since the whole space is a direct sum of these, it follows that
q (A) = 0.
F.28
Exercises
10.9
1. Letting A be a complex n n matrix, in obtaining the rational canonical form, one
obtains the Cn as a direct sum of the form
span x1 span xr
x2
z
}|
{ z
}|
{
x1 , , Am1 x1 , x2 , , Am2 x2 , , xl , Axl , , Aj xl
where there are k vectors in the above list. Since each span xi is A invariant, Aqk
is in the span of
x1
x2
z
}|
{ z
}|
{
x1 , , Am1 x1 , x2 , , Am2 x2 , , xl , Axl , , Aj xl , Aj+1 xl
122
Exercises
It follows that if p k + 2, then
qp Aqk = 0
m () l () + n () k ()
Thus you also have
=1
z }| {
r
m (A) l (A) x + n (A) k (A) k x = x
q
Y
nj
j ()
j=1
where the j () are irreducible. Explain using Corollary 8.3.11 why the irreducible
factors of the minimal polynomial are j () and why the minimal polynomial is of
the form
q
Y
r
j () j
j=1
where rj nj . You can use the Cayley Hamilton theorem if you like.
From the Cayley Hamilton theorem, the minimal polynomial divides the characteristic
polynomial and so from the mentioned corollary, it follows that the minimal polynomial
has the form described above with rj nj .
4. Find the minimal polynomial for
1 2 3
A= 2 1 4
3 2 1
by the above technique with the field of scalars being the rational numbers. Is what
you found also the characteristic polynomial?
1 2 3
2 1 4 , characteristic polynomial: X 3 3X 2 + 14
3 2 1
It is irreducible over the rationals by the rational root theorem. Therefore, the above
must be the minimal polynomial as well.
123
Exercises
5. Show, using the rational root theorem, the minimal polynomial for A in the above
problem is irreducible with respect to Q. Letting the field of scalars be Q find the
rational canonical form and a similarity transformation which will produce it.
To find the rational canonical form, you need to look for cyclic sets. You have a single
irreducible polynomial, 3 3 + 14 = (). Also (A) sends every vector to 0.
Therefore, you can simply start with a vector. Take e1 .
1
1 2
0 , 2 1
3 2
0
3
1
1 2
4 0 , 2 1
1
0
3 2
3
1 2 3
1
2 1 4 0
3 2 1
0
2
3
1
4 0
1
0
26
1
1
4
0 , 2 , 8 , 24
6
0
3
2
The last is definitely a linear combination of the first three because the first three are
an independent set. Hence, you ought to have these as columns. Then you have
1 1 4
1 2 3
1
0 2 8 2 1 4 0
0 3 2
3 2 1
0
0
= 1
0
0 14
0
0
1
3
1 4
2 8
3 2
1 2 1 1
2 3 0 2
1 3 2 4
1 2 1 2
This was just pulled out of the air. To find the rational canonical form, look for cyclic
sets.
1 0 0 0
1
1 2 1 1
1
1 2 1 1
1
0 1 0 0 0 2 3 0 2 0 2 3 0 2 0
0 0 1 0 0 , 1 3 2 4 0 , 1 3 2 4 0
0 0 0 1
0
1 2 1 2
0
1 2 1 2
0
1
2
1
1
2
3
3
2
3
1 1
1
0
0 2
2 4 0
1 2
0
1
2
,
1
1
2
3
3
2
1
0
2
1
4
1
1
0
2
4 0
2
0
124
Exercises
These reduce to
1
0
0
0
1
1
0 2
,
0 1
0
1
1
2
1
1
5
10
13
8
30
56
93
54
5
30
10 56
,
13 , 93
8
54
1
181
0
336
, row echelon form:
0
600
0
343
181
336
,
600
343
0
1
0
0
0
0
1
0
0 3
0 1
0 11
1
8
4 83 + 112 + + 3
Then
1
0
0
0
1 1
1 1
0 2
0 2
2 4 0 1
1 2
0 1
this. Also,
5 30
10 56
=
13 93
8 54
0
1
0
0
0
0
1
0
0
0
0
1
3
1
11
8
7. Let A : Q3 Q3 be linear. Suppose the minimal polynomial is ( 2) 2 + 2 + 7 .
Find the rational canonical form. Can you give generalizations of this rather simple
problem to other situations?
That second factor is irreducible over the rationals and so you would need the following
for the rational canonical form.
2 0 0
0 0 7
0 1 2
8. Find the rational canonical form with respect to the field of scalars equal to Q for the
matrix
0 0 1
A = 1 0 1
0 1 1
0 0
1 0
0 1
2
1
1 0
1 + 0 1
1
0 0
0
1
0 = 0
1
1
s t
s
t
1 1
0 0
1 1
125
Exercises
Starting with one of these vectors, lets find cyclic sets.
1
0
1 , 1
0
0
1
0 1
1
0 0
0 1 1 , 1 0
1 1
0
0 1
1
0
1
1 , 1 , 1
0
1
0
1
1
1 1
1
0
These of course will be dependent since otherwise we dont really have the minimal
polynomial. Thus the cycle can consist of only the first two. Thus we will have these
two as the first two columns and then we will have the eigenvector corresponding to
ker (A I)
1 0
1
ker 1 1 1
0
1
0
1
This eigenvector is just 0 . Then the rational canonical form of this matrix is
1
1
1 0 1
0 0
1 1 0 1 0
0
1 1
0 1
21
21
=
1
2
1
2
21
1
2
1
2
1
2
1
2
1
1 0 1
1 1 1 0
1
0
1 1
1
1 0 1
1 1 1 0
1
0
1 1
0
0
1
0 0
1 0
0 1
0 1
1 0
0 0
9. Let q () be a polynomial and C its companion matrix. Show the characteristic and
minimal polynomial of C are the same and both equal q ().
Let q () = a0 + a1 + + an1 n1 + n . Then its companion matrix is
0
0
a0
1 0
a1
C =
..
.
..
.
0
an1
Then you can show by induction that the characteristic polynomial of C is q (). See
Problem 43 on Page 66 It was also shown that q (C) = 0. (Cayley Hamilton theorem
for companion matrices.) Let p () be the minimal polynomial. Thus it has degree no
more than the degree of q () and thus must divide q (). Could it have smaller degree
126
Exercises
than q ()? No, it couldnt. To see this, consider, for example a companion matrix
for a monic polynomial of degree 4 denoted as A. The following gives the sequence
A0 , A, A2 , A3 . Note that these are linearly independent. Watch the ones to see this.
0 0 a
ad
1 0 0 0
0 0 0 a
0 1 0 0 1 0 0 b 0 0 b a + bd
0 0 1 0 0 1 0 c 1 0 c b + cd
0 1 d d2 + c
0 0 0 1
0 0 1 d
0 a
ad
a d2 +
c
2
0 b a + bd
b
d
+
c
+
ad
0 c b + cd a + c d2 + c + bd
1 d d2 + c b + d d2 + c + cd
This pattern will occur for any sized companion matrix. You start with ones down the
main diagonal and then the string of ones moves down one diagonal and then another
and so forth. After n 1 multiplications, you get a 1 in the lower left corner. Each
multiplication moves the ones into slots which had formerly been occupied with zeros.
That is why these are independent. Therefore, the degree of the minimal polynomial
is at least n and so the minimal and characteristic polynomials must coincide for any
companion matrix.
10. Use the existence of the rational canonical form to give a proof of the Cayley Hamilton
theorem valid for any field, even fields like the integers mod p for p a prime. The earlier
proof was fine for fields like Q or R where you could let but it is not clear the
same result holds in general.
The linear transformation has a matrix of the following form,
M1
..
M =
.
Mq
mk
p () =
q
Y
k ()
mk
k=1
C1k
..
.
Clkk
k=1
q Y
lk
Y
kj () =
k=1 j=1
q Y
lk
Y
k=1 j=1
k () j , rj mk
127
Exercises
It follows that
q (A) =
q Y
li
Y
k (A)rj
i=1 j=1
k rj
We know that k Cj
B1
..
B=
.
Bq
and () is a polynomial,
(B1 )
..
(B) =
.
(Bq )
Therefore, the above product for q (A) is the product of block diagonal matrices such
that every position on the diagonal has a zero matrix in at least one of the block
matrices in the product. This equals the zero matrix. Like this:
0 0 0
D 0 0
F 0 0
0 B 0 0 0 0 0 H 0
0 0 C
0 0 E
0 0 0
When you multiply the block matrices, it involves simply multiplying together the
blocks in the corresponding positions. Since at least one is zero, it follows the product
is the zero matrix.
11. Suppose you have two n n matrices A, B whose entries are in a field F and suppose G
is an extension of F. For example, you could have F = Q and G = C. Suppose A and
B are similar with respect to the field G. Can it be concluded that they are similar
with respect to the field F? Hint: Use the uniqueness of the rational canonical form.
Do A, B have the same minimal polynomial over F? Denote these by pA () and pB ()
You have
pA (A) = 0
and so pA (B) = 0 also because A and B are similar with respect to G. It follows that
pA () must divide pB () . Similar reasoning implies that pB () must divide pA ().
Therefore, these two minimal polynomials having coefficients in F coincide. Otherwise,
you could write
pB () = pA () + r ()
where the degree of r () is smaller than the common degree of pA () and pB ().
Then plugging in = B, you get a contradiction to the definition of pB () as having
the smallest degree possible. Say the common minimal polynomial over F is
q
Y
i ()mi
i=1
where i () is irreducible over F. Thus, each of these i is a polynomial with coefficients in F and hence in G, and so
i (A) , i (B)
128
Exercises
are similar over G. Letting vi and wj denote the A cyclic sets associated with
ker (i (A))
mi
, ker (i (B))
mi
respectively, it follows from Lemma 10.8.3 that the lengths of these vi and wi comprise exactly the same set of positive numbers. These lengths are identified with the
size of the Jordan blocks of the two similar matrices, i (A) , i (B). However, the
proof of this lemma exhibits a basis which yields the Jordan form, which happens to
involve only the field F. The companion matrices corresponding to two of these which
have the same length are therefore exactly the same, both being companion matrices
k
of i () for a suitable k. It follows that the rational canonical form for A, B over F
is exactly the same. Hence, by uniqueness, the two matrices are similar over F.
F.29
Exercises
11.4
1. Suppose the migration matrix for three
.5
.3
.2
locations is
0 .3
.8 0 .
.2 .7
Find a comparison for the populations in the three locations after a long time.
.5 0 .3
1 0 0
0.5
0
0.3
.3 .8 0 0 1 0 = 0.3 0.2
0
.2 .2 .7
0 0 1
0.2
0.2 0.3
1
3
0
0
1 0 53 0
2
10
9
3
15
0
0 , row echelon form: 0 1 10
0
10
1
1
3
0
0
0
0
10 0
5
5
It follows that the populations are
.6
t .9
1
where t is chosen such that the sum of the entries corresponds to the total amount at
the beginning.
P
2. Show that if i aij = 1, then if A = (aij ) , then the sum of the entries of Av equals
the sum of the entries of v. Thus it does not matter whether aij 0 for this to be so.
P P
P P
P
i
j aij vj =
j
i aij vj =
j vj
3. If A satisfies the conditions of the above problem, can it be concluded that limn An
exists?
Not necessarily. Consider the following matrix.
4
9
3 8
4
4
9
311 936
=
etc. The eigenvalues of this matrix are
3 8
312
937
1, 5. Thus there is no way that the limit can exist. Just do the powers of the matrix
to the eigenvector which corresponds to 5.
129
Exercises
4. Give
1.
1
0
0
0
1 , eigenvalues: 1, 1
0
5. Show that when a Markov matrix is non defective, all of the above theory can be proved
very easily. In particular, prove the theorem about the existence of limn An if the
eigenvalues are either 1 or have absolute value less than 1.
It suffices to show that limn An v exists for each v Rn . This is because you could
then assert that limn eTi An ej exists. But this is just the ij th entry of An . Let the
eigen pairs be
(1, v1 ) , (i , vi ) , i = 2, , n
and by assumption, these vi form a basis. Therefore, if v is any vector, there exist
unique scalars ci such that
X
v=
ci vi
i
It follows that
Am v = c1 v1 +
n
X
j=2
ci m vi c1 v1
5
2
5
A=
7
2
7
2
12
0
12
12
0
0
1
2
1
4
25
2
Does limn An exist? Note that all the rows sum to 1. Hint: This matrix is similar
to a diagonal matrix. The eigenvalues are 1, 1, 12 , 12 .
5
21 0 1
2
5
0
0 4
7
2 12 52
2
7
21 0 2
2
1 1 1 0
1 0 0 0
1
0 0 1
1
3 2 1 0 0
0 0
1 0 2
2
=
5
2
0 0 1 0 1
1
1
1
0 1
4
2 1 1 0
0 0 0 12
41 41 1 21
Now it follows that
1 1 1 0
3 2 1 0
An =
2 5 1 1
4
2 1 1 0
1
n
1
2n (1) + 1
2n
2n 3 (1)n + 1 2n
2
2
=
1n 2 (1)n + 1 1n
2
2
n
1
1
2n 2 (1) + 1
2n
(1)
0
0
0
1
1
1
1
0
1
0
1
0
1
1
2n
0
0 1
1
1
1
1
0
2n
4
4
n
2
0
(1) 2n + 1
0 3 (1)n 24n + 1
1
2 (1)n 23n + 1
2n
n
0 2 (1) 22n + 1
0
0
0
1
0
0 1
0 2
0 1
1 12
130
Exercises
2
4
A=
5
2
3
21
0
21
21
1
4
2
2
1
2
1
1
1
2
Note that the rows sum to 1 in this matrix also. Hint: This matrix is not similar
to a diagonal matrix but you can find the Jordan form and consider this in order to
obtain a formula for this product. The eigenvalues are 1, 1, 21 , 12 .
2 21 12 1
4
0
1 4
5
2 1 2
2
3 21 12 2
97 1 61 79
1 0 0 0
1 0
0
1
23 1 1 14 0 1 0 0 1 1
0
1
9
3
9
=
13 1 1 4 0 0 1 1 0 6 14 20
9
6
9
2
0 0 0 12
1 0
3
2
16
1 61 79
9
Now consider the Jordan block on the bottom.
It when raising this to the nth power, you get
=
=
1
2
X
n
2nk
k
0
k=0
1
0
2n
+n
1
0
n
1 21n
2
n
2n
1
0
n
2
0
1
2nk
1
2n1
0
1
2n1
0
0
1
0
k
0 1
0 0
n
97 1 61 97
(1)
0 0
0
1 0
0
1
23 1 1 14
1 1
0
1 0
0
0
1
9
3
9
13
1
4
1
1n
0
0
2
n
0
6
14
20
2n
9
6
9
1
7
1
1
0
3
2
16
1
0
0
0
2n
9
6
9
n
n
7
1 1n
7
1
1 1n
7
16
1 1n
2
n
(1)
+
1
1
n
n+1
92n
6
9
2n
22
9 (1) 92n 3 2
n
n
23
32
2 1n
1n
14 n 1 21n n 23 (1) + 1 2n 1
2
n
n+1
3
9
2
9 (1) 92n 3 2
92
4 n 1 21n n 13 (1)n + 1 1n 1 1 21n n + 1n 13 (1)n 22 n 1 21n n + 1
92
6
9
2
2
2
9
92
3
n
n
7
1 1n
1
1 1n
16
16
1 1n
n 16
n
n+1
92n 6 2
9 (1) + 1
2n 1
22
9 (1) 92n 3 2
8. Find limn An if it exists for the matrix
1
12
2
1
1
2
2
A=
1
1
2
3
2
2
3
2
21
21
3
2
3
2
0
0
0
1
131
Exercises
1
1 1 1 0
1 1
0
0
0
2 1 1 0 0
4 3 1 1
0
0
1
2
0
0
0
0
1
0
0
1
1
0
0 1
0
1
0 1
1 1
1 0
2 1
1 1 1 0
1 0 0 0
1
1 1
0 1n 0 0 1
0
0
2
2 1 1 0 0 0 1 0 1
4 3 1 1
0
0 0 0 1
you get
1 1 1 0
1 0 0 0
1
1 1
0 0 0 0 1
0
0
2 1 1 0 0 0 1 0 1
4 3 1 1
0 0 0 1
0
0 1 1 0
1 0 1 0
=
1
1
2 0
3
3
3 1
0
1
1
2
0
1
1
2
1
1
0
1
1
1
0
1
0
0
0
1
0
0
and so, in the limit,
0
1
0
0
0
1
9. Given an example of a matrix A which has eigenvalues which are either equal to 1,1,
or have absolute value strictly less than 1 but which has the property that limn An
does not exist.
1 0 0
An easy example is 0 0 1 , eigenvalues: 1, 1
0 1 0
1 0 0
1 0 0
0 0 1 alternates between the identity and 0 0 1 .
0 1 0
0 1 0
10. If A is an n n matrix such that all the eigenvalues have absolute value less than 1,
show limn An = 0.
The Jordan form is
J (1 )
..
.
J (r )
m
0.
11. Find an example of a 3 3 matrix A such that limn An does not exist but
limr A5r does exist.
An easy example is the following.
i(2/5)
e
0
0
A=
0
ei(2/5)
0
i(2/5)
0
0
e
i(2/5)
5
e
0
0
= I.
A5 =
0
ei(2/5)
0
i(2/5)
0
0
e
Therefore, limr A5r exists. However, limr Ar cannot exist because it will just
yield diagonal matrices which have various fifth roots of 1 on the diagonal.
132
Exercises
12. If A is a Markov matrix and B is similar to A, does it follow that B is also a Markov
matrix?
No. Start with a Markov matrix
1/2 1/3
1/2 2/3
1
5
3
13. P
In Theorem 11.1.3
P suppose everything is unchanged except that you assume either
a
1
or
ij
j
i aij 1. Would the same conclusion be valid? What if you dont
insist that each aij 0? Would the conclusion hold in this case?
You cant conclude that the limit exists if the sign of aij is not restricted. For example,
4
9
3 8
has an eigenvalue of 5Pand so the limit of powers of this matrix cannot exist. What
about the case where i aij 1 instead of equal to 1? In this case, you do still
get the conclusion of this theorem. The condition was only used to get the entries
of An bounded independent of n and this can be accomplished just as well with the
inequality.
14. Let V be an n dimensional vector space and let x V and x 6= 0. Consider x
x,Ax, ,Am1 x where
Am x span x,Ax, ,Am1 x
and m
is the smallest such
that the above inclusion in the span takes place. Show
that x,Ax, ,Am1 x must be linearly independent. Next suppose {v1 , , vn }
is a basis for V . Consider vi as just discussed, having length mi . Thus Ami vi is a
linearly combination of vi ,Avi , ,Am1 vi for m as small as possible. Let pvi () be
the monic polynomial which expresses this linear combination. Thus pvi (A) vi = 0
and the degree of pvi () is as small as possible for this to take place. Show that the
minimal polynomial for A must be the monic polynomial which is the least common
multiple of these polynomials pvi ().
Let p () be the minimal polynomial. Then there exists ki () , a monic polynomial
such that
p () = ki () pvi () + ri ()
where ri () must be zero since otherwise, you would have ri (A) vi = 0 and this would
contradict the minimality of the degree of pvi (). Therefore, each of these pvi ()
divides p () . If q () is their least common multiple, then it follows that q (A) = 0
because every vector is a linear combination of these vectors vi . Therefore, p ()
divides q (). Thus
q () = p () k ()
Now each pvi () divides p () so it is a common multiple. It follows that k () = 1
since otherwise, q () wouldnt really be the least common multiple.
133
Exercises
15. If A is a complex Hermitian n n matrix which has all eigenvalues nonnegative, show
that there exists a complex Hermitian matrix B such that BB = A.
You have A = U DU for D a diagonal matrix having all nonnegative entries. Then
just note that B = U D1/2 U U D1/2 U, U is unitary. Clearly you should take B =
U D1/2 U.
16. Suppose A, B are n n real Hermitian matrices and they both have all nonnegative
eigenvalues. Show that det (A + B) det (A) + det (B).
Let P 2 = A, Q2 = B where P, Q are Hermitian and nonnegative. Then
P
A+B = P Q
Q
Now apply the Cauchy Binet formula to both sides.
X
det (A + B) =
det (Ci )2 det (P )2 + det (Q)2
i
= det P 2 + det Q2 = det (A) + det (B)
where the Ci are n n submatrices of P Q . Note that this does not depend on
the matrices being real. You could write
P
A + B = P Q
Q
Then
det (A + B)
det (Ci Ci )
det P
+ det Q
c
is an (n + 1) (n + 1) Hermitian nonnegative matrix where
b A
is a scalar and A is n n. Show that must be real, c = b, and A = A , A is
nonnegative, and that if = 0, then b = 0.
b
B =
= B and so you must have is real and b = c while A = A .
c A
Suppose = 0 and b 6= 0. I need to show that B is not nonnegative.
17. Suppose B =
}| {
0 b
x
b A
x
b x
x
= b x+x b + x Ax
b + Ax
z
If for some x, x Ax < 0, then let = 0 and you see that B is not nonnegative.
Therefore, A must be nonnegative. But now you could let x = b and then pick be
large and negative and again see that B is not nonnegative. Hence if = 0, then
b = 0. Now consider b 6= 0 and determine the sign of . Consider
B
z
}| {
b
=
b A
0
=
134
Exercises
18. If A is an n n complex Hermitian and nonnegative matrix, show that there exists
an upper triangular matrix B such that B B = A. Hint: Prove this by induction. It
is obviously true if n = 1. Now if you have an (n + 1) (n
+ 21) Hermitian
nonnegative
b
matrix, then from the above problem, it is of the form
, real.
b A
From the above problem, a generic (n + 1) (n + 1) Hermitian nonnegative matrix is
of the form just claimed where A is nonnegative and Hermitian and is real. In case
= 0, there is nothing to prove. You would just use
0 0
0 0
0 B
0 B
where B B = A. Otherwise, you consider
0
b B
0
b
B
B12
B22
0
B22
det (A)
= det (B11
) det (B22
) det (B11 ) det (B22 )
= det (B11
B11 ) det (B22
B22 )
This determinant is clearly equal to the product of the determinants of B11 and B22 .
= det (B11
B11 ) det (B22
B22 )
135
Exercises
This follows from induction. It is clearly true if n = 1. Now let A be nn, nonnegative
and Hermitian and assume the theorem is true for all nonnegative matrices which are
of smaller size. Then
B a
A=
a
It follows that B and are both nonnegative. Then from the above,
det (A) det (B) det ()
By induction, this implies
det (A)
F.30
Aii
Exercises
12.7
1. Find the best solution to the system
x + 2y = 6
2x y = 5
3x + 2y = 0
1 2
1 2
14 6
2 1 2 1 =
6 9
3 2
3 2
1 2
6
14 6
x
Solve
= 2 1 5
6 9
y
3 2
0
17
14 6
x
16
15
=
, Solution is:
1
6 9
y
7
45
2
0
0
1 0 1
1 1 0
2 0 0
1
1
1
5 6 25 5
6
6
6 6 30
6
6 6
1
1
1
= 61 6
5
6
0 0
6
6 5 6 30 5 6
1
1
1
2
6
5
6
5
5
0
0
3
15
5
5
136
Exercises
Then one example of an orthonormal basis extending the given vector is
2
1
1
5 6
30
6 6
5 5
1 6 , 1 5 6 ,
0
6
6
1
1
1
5
6
5 5
3
15
It satisfies all the axioms of an inner product obviously. The only ones which are not
obvious are whether (Ax, x) 0 and so forth. But A = U DU where D is diagonal
and U is orthogonal. Therefore,
2
where is the smallest eigenvalue which is assumed positive. Therefore, also (Ax, x) =
0 if and only if |x| = 0. The Cauchy Schwarz inequality says
|(Ax, y)| (Ax, x)1/2 (Ay, y)1/2
4. Let
||x|| max {|xj | : j = 1, 2, , n} .
T
Show this is a norm on Cn . Here x = x1 xn
. Show
1/2
The axioms of a norm are all obvious except for the triangle inequality. However,
kx + yk = max {|xi + yi | , i n} kxk + kyk
so even this one is all right. Also for some i
kxk
1/2
2
= |xi | = |xi |
5. Let
||x||1
Show this is a norm on Cn . Here x =
n
X
j=1
|xi |
"1/2
= |x| .
|xj | .
x1
xn
1/2
T
. Show
Of course. Just square both sides and the left has mixed terms which are not present
on the right.
137
Exercises
6. Show that if |||| is any norm on any vector space, then
|||x|| ||y||| ||x y|| .
kxk = kx y + yk kx yk + kyk and so
kxk kyk kx yk
Now repeat the argument with the x, y switched. Finally, |kxk kyk| equals either
kxk kyk or kyk kxk either way, you get the inequality.
7. Relax the assumptions in the axioms for the inner product. Change the axiom about
(x, x) 0 and equals 0 if and only if x = 0 to simply read (x, x) 0. Show the Cauchy
Schwarz inequality still holds in the following form.
|(x, y)| (x, x)
1/2
1/2
(y, y)
You have x =
(x, uj ) uj and y =
(x, y) =
(y, uj ) uj . Therefore,
(x, uj ) uj ,
X
j
(y, uk ) uk
j,k
n
X
(x, uk ) (y, uk )
k=1
9. Let the vector space V consist of real polynomials of degree no larger than 3. Thus a
typical vector is a polynomial of the form
a + bx + cx2 + dx3 .
For p, q V define the inner product,
(p, q)
x(1/2)
1/2 =
3 (2x 1)
1 2
x 2 )
0 (
=
1
2
2
2
2
kx (x , 3(2x1)) 3(2x1)(x ,1)1k
kx 6 3 3(2x1) 13 k
x(x,p1 )p1
kx(x,p1 )p1 k
R
1
p3 (x) =
= 6 5 x2 x + 16
138
Exercises
=
1
1
1
1
3
3
2
2800
kx 20 56 5(x x+ 6 ) 20 3 3(2x1) 4 k
2800
p4 (x) =
= 20 7 x3 32 x2 + 35 x
1
20
1
3
2
1, 3 (2x 1) , 6 5 x x +
, 20 7 x3 x2 +
6
2
is
3
1
x
5
20
10. Let Pn denote the polynomials of degree no larger than n 1 which are defined on an
interval [a, b] . Let {x1 , , xn } be n distinct points in [a, b] . Now define for p, q Pn ,
(p, q)
n
X
p (xj ) q (xj )
j=1
Show this yields an inner product on Pn . Hint: Most of the axioms are obvious. The
one which says (p, p) = 0 if and only if p = 0 is the only interesting one. To verify this
one, note that a nonzero polynomial of degree no more than n 1 has at most n 1
zeros.
If (p, p) = 0 then p equals zero at n distinct points and so it must be the zero polynomial
because it has degree at most n 1 and so it can have no more than n 1 roots.
11. Let C ([0, 1]) denote the vector space of continuous real valued functions defined on
[0, 1]. Let the inner product be given as
Z 1
(f, g)
f (x) g (x) dx
0
Show this is an inner product. Also let V be the subspace described in Problem 9.
Using the result of this problem, find the vector in V which is closest to x4 .
It equals
x4 , 1 1 + x4 , 3 (2x 1)
3 (2x 1)
1
1
+ x4 , 6 5 x2 x +
6 5 x2 x +
6
6
3
3
3
1
3
1
+ x4 , 20 7 x3 x2 + x
20 7 x3 x2 + x
2
5
20
2
5
20
Now it is just a matter of working these out.
R1
2
x4 , 3 (2x 1) = 0 x4 3 (2x 1) = 15
3
R 1 4
1
2
4
2
2
x , 6 5 x x + 6 = 0 x 6 5 x x + 16 = 35
5
R
1
1
1
1
x4 , 20 7 x3 32 x2 + 53 x 20
= 0 x4 20 7 x3 32 x2 + 35 x 20
= 70
7
Therefore, the closest point is
1
2
2
1
+
3 3 (2x 1) +
56 5 x2 x +
+
5 15
35
6
1
3
3
1
+
720 7 x3 x2 + x
70
2
5
20
2
1
9
= 2x3 x2 + x
7
7
70
139
Exercises
12. A regular Sturm Liouville problem involves the differential equation, for an unknown function of x which is denoted here by y,
(6.37)
= 0,
= 0.
and then integrate. Use the boundary conditions to show that y (a) z (a)z (a) y (a) =
0 and y (b) z (b) z (b) y (b) = 0. The formula, 6.37 is called an orthogonality relation.
It turns out there are typically infinitely many eigenvalues and it is interesting to write
given functions as an infinite series of these eigenfunctions.
Let y go with and z go with .
=
=
0
0
z (p (x) y ) y (p (x) z ) =
d
(p (x) y z p (x) z y)
dx
140
Exercises
Look at the stuff on the top line. From the assumptions on the boundary conditions,
C1 y (a) + C2 y (a)
C1 z (a) + C2 z (a)
= 0
= 0
and so
y (a) z (a) y (a) z (a) = 0
Similarly,
y (b) z (b) y (b) z (b) = 0
Hence, that stuff on the top line equals zero and so the orthogonality condition holds.
13. Consider the continuous functions defined on [0, ] , C ([0, ]) . Show
Z
(f, g)
f gdx
0
nq
o
2
is an inner product on this vector space. Show the functions
sin
(nx)
are
n=1
an orthonormal set. What does
of the vector space
q the dimension
qthis mean about
2
2
C ([0, ])? Now let VN = span
sin (x) , ,
sin (N x) . For f C ([0, ]) find
a formula for the vector in VN which is closest to f with respect to the norm determined
from the above inner product. This is called the N th partial sum of the Fourier series
of f . An important problem is to determine whether and in what way this Fourier
series converges to the function f . The norm which comes from this inner product is
sometimes called the mean square norm.
0
0
Now also
R
0
sin (nx)
are orthonormal.
1 4 1 1
2 1 2
3
A=
4 9 0
1
5 6 3
4
Find an orthonormal basis for V. Hint: You might first find a basis and then use the
Gram Schmidt procedure.
141
Exercises
1
2
4
5
4
1
9
6
1 1
x
y
2
3
0
1 z
3
4
w
9
7 t3
0
0
4 t3
= , Solution is: 7
0
t3
0
0
This is pretty easy. The space has dimension 1. Therefore, an orthonormal basis is
just
9
146
146
9
4 2 146
1
= 73
7
146
16 + 49 + 81 7 146
0
0
15. The Gram Schmidt process starts with a basis for a subspace {v1 , , vn } and produces an orthonormal basis for the same subspace {u1 , , un } such that
span (v1 , , vk ) = span (u1 , , uk )
for each k. Show that in the case of Rm the QR factorization does the same thing.
More specifically, if
A = v1 vn
and if
A = QR
q1
qn
then the vectors {q1 , , qn } is an orthonormal set of vectors and for each k,
span (q1 , , qk ) = span (v1 , , vk )
This follows from the way we multiply matrices and the definition of an orthogonal
matrix.
16. Verify the parallelogram identify for any inner product space,
|x + y|2 + |x y|2 = 2 |x|2 + 2 |y|2 .
Why is it called the parallelogram identity?
2
|x + y| + |x y| = (x + y, x + y) + (x y, x y)
= |x|2 + |y|2 + 2 (x, y) + |x|2 + |y|2 2 (x, y) .
17. Let H be an inner product space and let K H be a nonempty convex subset. This
means that if k1 , k2 K, then the line segment consisting of points of the form
tk1 + (1 t) k2 for t [0, 1]
is also contained in K. Suppose for each x H, there exists P x defined to be a point
of K closest to x. Show that P x is unique so that P actually is a map. Hint: Suppose
z1 and z2 both work as closest points. Consider the midpoint, (z1 + z2 ) /2 and use the
parallelogram identity of Problem 16 in an auspicious manner.
Suppose there are two closest points to x say k1 , k2 both work. Then
K and
2
2
x k1 + k2 = x k1 + x k2
2
2
2
2
2
k1 +k2
2
is also in
142
Exercises
2
2
k2 k1 2
+ 2 x k1 + 2 x k2
2
2
2
2
2
2
k2 k1
+ d2
2
18. In the situation of Problem 17 suppose K is a closed convex subset and that H
is complete. This means every Cauchy sequence converges. Recall from calculus a
sequence {kn } is a Cauchy sequence if for every > 0 there exists N such that
whenever m, n > N , it follows |km kn | < . Let {kn } be a sequence of points of K
such that
lim |x kn | = inf {|x k| : k K}
n
This is called a minimizing sequence. Show there exists a unique k K such that
limn |kn k| and that k = P x. That is, there exists a well defined projection map
onto the convex subset of H. Hint: Use the parallelogram identity in an auspicious
manner to show {kn } is a Cauchy sequence which must therefore converge. Since K
is closed it follows this will converge to something in K which is the desired vector.
Let x be given and let {kn } be a minimizing squence.
|x kn | inf {|x y| : y K}
Then
2
kn km 2 x kn
x kn 2
x km 2
x km
+
+
= 2
+2
2
2
2
2
2
1
2
|x kn | +
2
1
2
|x kn | +
2
2
1
kn + km
2
|km x| x
2
2
1
2
|km x| 2
2
|y (x + t (w x))|
143
Exercises
where w K and x K. It equals
2
f (t) = |y x| + t2 |w x| 2t Re hy x, w xi
Suppose x is the point of K which is closest to y. Then f (0) 0. However, f (0) =
2 Re hy x, w xi . Therefore, if x is closest to y,
Re hy x, w xi 0.
Next suppose this condition holds. Then you have
2
|y (x + t (w x))| |y x| + t2 |w x| |y x|
hP x P y,y P yi
hP y P x,x P xi
0
0
Thus
hP x P y,x P xi 0
Hence
hP x P y,x P xi hP x P y,y P yi 0
and so
hP x P y,xy (P x P y)i 0
|xy| |P x P y| hP x P y,P x P yi = |P x P y|
21. Give an example of two vectors in R4 x, y and a subspace V such that x y = 0 but
P xP y 6= 0 where P denotes the projection map which sends x to its closest point on
V.
Try this. V is the span of e1 and e2 and x = e3 + e1 , y = e4 + e1 .
P x = (e3 + e1 , e1 ) e1 + (e3 + e1 , e2 ) e2 = e1
P y = (e4 + e1 , e1 ) e1 + (e4 + e1 , e2 ) e2 = e1
P xP y = 1
22. Suppose you are given the data, (1, 2) , (2, 4) , (3, 8) , (0, 0) . Find the linear regression
line using the formulas derived above. Then graph the given data along with your
regression line.
You draw the graphs. You want to solve
a+b=2
2a + b = 4
3a + b = 8
b=0
144
Exercises
Of course there is no solution. The least squares solution is to solve
T
1 1
1 1
1 1
2
2 1 2 1 a
2 1 4
=
3 1 3 1 b
3 1 8
0 1
0 1
0 1
0
13
14 6
a
34
5
=
, Solution is:
Then the best line is
6 4
b
14
52
y=
13
2
x
5
5
23. Generalize the least squares procedure to the situation in which data is given and you
desire to fit it with an expression of the form y = af (x) + bg (x) + c where the problem
would be to find a, b and c in order to minimize the error. Could this be generalized
to higher dimensions? How about more functions?
Say you had ordered pairs given which you wanted the curve to go through. Say these
are (xi , yi ) for i n. Then you would be finding least squares solutions for a, b, c to
yi = af (xi ) + bg (xi ) + c, i = 1, , n
You would simply obtain a least squares solutions to this. The principles are the same
for higher dimensions.
24. Let A L (X, Y ) where X and Y are finite dimensional vector spaces with the dimension of X equal to n. Define rank (A) dim (A (X)) and nullity(A) dim (ker (A)) .
r
Show that nullity(A) + rank (A) = dim (X) . Hint: Let {xi }i=1 be a basis for ker (A)
r
nr
nr
and let {xi }i=1 {yi }i=1 be a basis for X. Then show that {Ayi }i=1 is linearly
independent and spans AX.
m
Let {Ayi }i=1 be a basis for AX and let {xi }i=1 be a basis for ker (A).
Then {y1 , , ym , x1 , , xr } is linearly independent. In fact it spans X because if
z X, Then
m
X
Az =
ci Ayi
i=1
Pm
and so z i=1 ci yi ker (X) and so it is in the span of the {xi }i=1 . Therefore,
z is in the span of the given vectors and so this list of vectors is a basis. Hence the
dimension of X is equal to m + r = rank (A) + dim (ker (A)).
25. Let A be an mn matrix. Show the column rank of A equals the column rank of A A.
Next verify column rank of A A is no larger than column rank of A . Next justify the
following inequality to conclude the column rank of A equals the column rank of A .
rank (A) = rank (A A) rank (A )
= rank (AA ) rank (A) .
r
Hint: Start with an orthonormal basis, {Axj }j=1 of A (Fn ) and verify {A Axj }j=1
is a basis for A A (Fn ) .
Pr
Say you have i=1 ci A Axi = 0. Then for any appropriate y,
!
!
r
r
X
X
0=
ci A Axi , y =
ci Axi , Ay
i=1
i=1
145
Exercises
In particular, you could have y =
i ci xi .
r
X
ci Axi = 0
i=1
F.31
Exercises
12.9
T
1. Here are three vectors in R4 : (1, 2, 0, 3) , (2, 1, 3, 2) , (0, 0, 1, 2) . Find the three
dimensional volume of the parallelepiped determined by these three vectors.
(1, 2, 0, 3) (1, 2, 0, 3) = 14
(1, 2, 0, 3) (2, 1, 3, 2) = 10
(1, 2, 0, 3) (0, 0, 1, 2) = 6
(2, 1, 3, 2) (2, 1, 3, 2) = 18
(2, 1, 3, 2) (0, 0, 1, 2) = 1
(0, 0, 1, 2) (0, 0, 1, 2) = 5
volume is
218
14 10 6
det 10 18 1 = 218
6 1 5
T
2. Here are two vectors in R4 : (1, 2, 0, 3) , (2, 1, 3, 2) . Find the volume of the parallelepiped determined by these two vectors.
(1, 2, 0, 3) (1, 2, 0, 3) = 14
(1, 2, 0, 3) (2, 1, 3, 2) = 10
146
Exercises
(2, 1, 3, 2) (2, 1, 3, 2) = 18
det
volume is
14 10
10 18
= 152
152
T
3. Here are three vectors in R2 : (1, 2) , (2, 1) , (0, 1) . Find the three dimensional
volume of the parallelepiped determined by these three vectors. Recall that from the
above theorem, this should equal 0.
It does equal 0.
4. Find the equation of the plane through the three points (1, 2, 3) , (2, 3, 1) , (1, 1, 7) .
5. Let T map a vector space V to itself. Explain why T is one to one if and only if T is
onto. It is in the text, but do it again in your own words.
This is because if T is one to one, it takes a basis to a basis. If T is onto, then it takes
a basis to a spanning set which must be independent, since otherwise, there would be
a linearly independent set of vectors which would also span and yet have fewer vectors
than a basis.
6. Let all matrices be complex with complex field of scalars and let A be an nn matrix
and B a m m matrix while X will be an n m matrix. The problem is to consider
solutions to Sylvesters equation. Solve the following equation for X
AX XB = C
where C is an arbitrary n m matrix. If q () is a polynomial, show first that if
AX XB = 0, then q (A) X Xq (B) = 0. Next define the linear map T which maps
the n m matrices to the n m matrices as follows.
T X AX XB
Show that the only solution to T X = 0 is X = 0 if and only if (A) (B) = .
Conclude that there exists a unique solution for each C to Sylvesters equation if and
only if (A) (B) = .
q (A) X Xq (B) = q (A) X = 0
Now explain why q (A)1 exists if and only if (A) (B) = .
Consider the first part. Multiply by An1 . Then An1 AX An1 XB = 0. Since
AX = XB, you can make multiple switches and write this as
An X XB n = 0
Now it follows that the desired result holds for a polynomial. Consider T which maps
the vector space of n m matrices to itself. Let q () be the characteristic polynomial
of B as suggested and suppose T X = 0. Then
q (A) X Xq (B) = 0 = q (A) X.
Qm
Let the characteristic polynomial for B be i=1 ( i ) . In case there are no shared
eigenvalues, it follows that
m
Y
q (A) =
(A i I)
i=1
147
Exercises
det (UI )
"1/2
where I is a choice of m rows which are selected from U so that UI is the square matrix
which results from these rows. The geometric meaning of the above expression is this.
det (UI ) gives the m dimensional volume of the parallelepiped which is obtained from
a projection onto the m dimensional subspace of Rn determined by setting the other
n m variables equal to 0. Thus the volume is the square root of the sum of the
squares of the areas of these projections.
F.32
Exercises
13.12
Since y is arbitrary, this shows (AB) x = B A x and since x is arbitrary, this shows
the formula.
4. Using the singular value decomposition, show that for any square matrix A, it follows
that A A is unitarily similar to AA .
You have A = U V where is the singular value matrix and U, V are unitary of the
right size. Therefore, A A = V U U V = V 2 V . Similarly, AA = U 2 U. Then
2 = U AA U and so
A A = V U AA U V
Since these matrices are all square, this does it. U V is unitary.
5. Let A, B be a m n matrices. Define an inner product on the set of m n matrices
by
(A, B)F trace (AB ) .
148
Exercises
Show this is an inner product
P satisfying all the inner product axioms. Recall for M an
n n matrix, trace (M ) ni=1 Mii . The resulting norm, ||||F is called the Frobenius
norm and it can be used to measure the distance between two matrices.
This was done earlier.
2j
n
X
xk p (k ) vk =
k=1
k=1
n
X
xk 0vk = 0
k=1
Hence p (A) = 0. Now drop the assumption that A is nondefective. From the above,
there exists a sequence Ak which is non defective which converges to A and also
pk () p () uniformly on compact sets because these characteristic polynomials
are defined in terms of determinants of the corresponding matrix. See the above
construction of the Ak . It is probably easiest to use the Frobinius norm for the last
part.
kpk (Ak ) p (A)kF kpk (Ak ) p (Ak )kF + kp (Ak ) p (A)kF
149
Exercises
The first term converges to 0 because the convergence of Ak to A implies all entries
of Ak lie in a compact set. The second term converges to 0 also because the entries of
Ak converge to the corresponding entries of A.
9. Prove that Theorem 13.4.6 and Corollary 13.4.7 can be strengthened so that the
condition
on
the Ak is necessary as well as sufficient. Hint: Consider vectors of the
x
form
where x Fk .
0
Say the matrix is positive definite. Then you could just look at
x
T
x
0 M (A)
= xT M (A)k x
0
and this needs to be positive. Hence those principle minors have all positive determinants.
10. Show directly that if A is an n n matrix and A = A (A is Hermitian) then all the
eigenvalues and eigenvectors are real and that eigenvectors associated with distinct
eigenvalues are orthogonal, (their inner product is zero).
(x, x) so = .
Now if x is an eigenvector,
(x, x) = (Ax, x) = (x,Ax) =
A
x =Ax =
x and so x + x
is also an eigenvector. Hence, you can assume that
the eigenvectors are real. If x
is pure imaginary, you could simply multiply by i and
get an eigenvector which is real.
11. Let v1 , , vn be an orthonormal basis for Fn . Let Q be a matrix whose ith column
is vi . Show
Q Q = QQ = I.
This follows from how we multiply matrices.
12. Show that an n n matrix Q is unitary if and only if it preserves distances. This
means |Qv| = |v| . This was done in the text but you should try to do it for yourself.
13. Suppose {v1 , , vn } and {w1 , , wn } are two orthonormal bases for Fn and suppose Q is an n n matrix satisfying Qvi = wi . Then show Q is unitary. If |v| = 1,
show there is a unitary transformation which maps v to e1 .
This is easy because you show it preserves distances.
14. Finish the proof of Theorem 13.6.5.
15. Let A be a Hermitian matrix so A = A and suppose all eigenvalues of A are larger
than 2 . Show
2
(Av, v) 2 |v|
Where here, the inner product is
(v, u)
n
X
vj uj .
j=1
150
Exercises
16. Suppose A + A has all negative eigenvalues. Then show that the eigenvalues of A
have all negative real parts.
You have
0 >
=
((A + A ) x, x) = (Ax, x) + (A x, x)
(Ax, x) + (Ax, x)
|x| = Re () |x|
0 > |x| +
2
2
2
ei n 00
ei n 10
ei n 20
2
2
2
ei n 11
ei n 21
ei n 01
2
2
1 ei 2
n 02
ei n 12
ei n 22
U=
n
..
..
..
.
.
.
2
2
2
ei n 0(n1) ei n 1(n1) ei n 2(n1)
where
2
ei n (n1)0
2
ei n (n1)1
2
ei n (n1)2
..
.
ei n (n1)(n1)
Now argue U is unitary and use this to establish the result. To show this verify
each row has length 1 and the inner product of two different rows gives 0. Now
2
2
Ukj = ei n jk and so (U )kj = ei n jk .
Take the inner product of two rows.
n1
X
ei n jk ei n jl
j=0
n1
X
ei n jk ei n jl =
j=0
n1
X
ei n j(lk)
j=0
1e
i 2
n n(lk)
2
1 ei n (lk)
= 0 if l 6= k
Pn1
2
because ei n n(lk) = 1. What if l = k? then the sum reduces to j=0 1n 1n = 1
and so this matrix is unitary. Therefore, its inverse is just the transpose conjugate
which yields the other formula.
18. Let f be a periodic function having period 2. The Fourier series of f is an expression
of the form
n
X
X
ikx
ck e lim
ck eikx
n
k=
k=n
and the idea is to find ck such that the above sequence converges in some way to f . If
f (x) =
ck eikx
k=
151
Exercises
and you formally multiply both sides by eimx and then integrate from 0 to 2,
interchanging the integral with the sum without any concern for whether this makes
sense, show it is reasonable from this to expect
Z 2
1
cm =
f (x) eimx dx.
2 0
Now suppose you only know f (x) at equally spaced points 2j/n for j = 0, 1, , n.
Consider the Riemann sum for this integral obtained
n from using the left endpoint of
the subintervals determined from the partition 2
j
. How does this compare with
n
j=0
the discrete Fourier transform? What happens as n to this approximation?
Multiply by eimx and formally integrate both sides. This leads to the formula for
cm . Now use a left sum. This would be
n1
1 X
2
2
2
im 2
j
n
f
j e
(j + 1)
j
2 j=0
n
n
n
=
n1
2
2
1X
f
j eim n j
n j=0
n
1 f by the
n
2
n j / n.
unitary matrix U
20. Let A be a complex m n matrix. Using the description of the Moore Penrose inverse
in terms of the singular value decomposition, show that
1
lim (A A + I)
0+
A = A+
where the convergence happens in the Frobenius norm. Also verify, using the singular
value decomposition, that the inverse exists in the above formula.
Recall that the Moore Penrose inverse is
1
0
V
U
0
0
0
where A = U
V . The matrices U, V are unitary and of the right size. First
0 0
of all, observe that there is no problem about the existence of the inverse mentioned
in the formula. This is because
((A A + I) x, x) |x|
and so it is a one to one matrix. Now consider the limit assertion. The left side equals
152
Exercises
1
0
0
V
UU
V + I
V
0 0
0 0
0
0+
2
1
0
0
= lim V
V + I
V
U
0 0
0 0
0
lim
Now
2
V
0
=
=
0
0
2
V
0
2
V
0
= V
= V
0
0
0
0
V + I
V + V V
+ I
2 + I 0
0
I
! 2
1
+ I
0
=V
1
1
1
0
0
0
0
0
0
0
0
1
0
U
0 0
"
0
0
U
0 0
1 I
!
1
2 + I
0
0
!
1
Of course and 2 and so forth are diagonal matrices. Thus 2 + I
is also a
! 2
1
diagonal matrix having on the diagonal an expression of the form i +
i and
this clearly converges as 0 to 1
.
Therefore,
the
above
converges
to
i
1
0
V
U = A+
0
0
+
21. Show that A+ = (A A) A . Hint: You might use the description of A+ in terms of
the singular value decomposition.
1
0
0
Recall that A+ = V
U where A = U
V . The was the
0
0
0 0
diagonal matrix which consisting of square roots of eigenvalues of A A, arranged in
decreasing order from top left toward lower right.
2
0
A A = V
V
0 0
2
0
0
V
V V
U =V
U = A+
0
0
0 0
0
0
153
Exercises
F.33
Exercises
14.7
1. Solve the system
using the
solving it
4 1
1 5
0 2
1 1
x
1
5 2 y = 2
2 6
z
3
4
1
0
1 1
x
1
7 2 y = 2
2 4
z
3
1 1
x
1
7 2 y = 2
2 4
z
3
Gauss Seidel method and the Jacobi method. Check your answer by also
using row operations.
5
1
x
1
94
7
2 y = 2 , Solution is: 94
67
4
z
3
94
using the
solving it
5 1
1 7
0 2
4
1
0
Gauss Seidel method and the Jacobi method. Check your answer by also
using row operations.
9
1
x
1
100
21
2 y = 2 , Solution is: 100
43
6
z
3
100
using the
solving it
4 1
1 7
0 2
5
1
0
Gauss Seidel method and the Jacobi method. Check your answer by also
using row operations.
5
1
x
1
118
9
2 y = 2 , Solution is: 118
42
4
z
3
59
4. If you are considering a system of the form Ax = b and A1 does not exist, will either
the Gauss Seidel or Jacobi methods work? Explain. What does this indicate about
finding eigenvectors for a given eigenvalue?
No. These methods only work when there is a unique solution.
5. For ||x|| max {|xj | : j = 1, 2, , n} , the parallelogram identity does not hold.
Explain.
Let x = (1, 0) , y = (0, 1) . Then
kx + yk2 + kx yk2
2
2 kxk + 2 kyk
= 1+1=2
= 4
154
Exercises
Show the norm || which comes from an inner product is strictly convex.
Let x, y be as described.
If x 6= y, then
x + y
2
x y
2
1
1
2
2
2
+
2
2
= 2 kxk + 2 kyk = kxk
x + y
2
2
2
< kxk
7. A norm |||| is said to be uniformly convex if whenever ||xn || , ||yn || are equal to 1 for
all n N and limn ||xn + yn || = 2, it follows limn ||xn yn || = 0. Show the
norm || coming from an inner product is always uniformly convex. Also show that
uniform convexity implies strict convexity which is defined in Problem 6.
The usual norm satisfies the parallelogram law. Thus
|xn yn |
=
=
155
Exercises
P
1/2
Pm
m
2
Say wn = k=1 cnk vk . Define a norm on V by kwk
where w =
k=1 |ck |
Pm
c
v
.
This
gives
a
norm
on
V
which
is
equivalent
to
the
given
norm
on V .
k=1 k k
Pm
Thus if wn w, it follows that cn c in Cm and so w =
c
v
and
so
k=1 k k
V is closed. Let w, w be two points and suppose without loss of generality that
dist (w, V ) dist (w , V ) 0. Let v V be such that kw vk < dist (w , V ) + .
Then
|dist (w, V ) dist (w , V )| =
kw vk kw vk +
kw w k + kw vk kw vk +
kw w k +
Then since is arbitrary, this shows that |dist (w, V ) dist (w , V )| kw w k and
so this is continuous as claimed. Now let w
/ V. Thus dist (w, V ) > 0 because if this
is not so, there wouild exist vn w and by the first part, w V . Let v V be such
wv
that 2 dist (w, V ) > kw vk. Then you can consider z = kwvk
. If v1 V,
kz v1 k =
wv
v1 kw vk
kw vk
kw vk
1
kw v kw vk v1 k
2 dist (w, V )
dist (w, V )
1
=
2 dist (w, V )
2
Since v1 is arbitrary, it follows that dist (z, V ) 12 . If you have an infinite dimensional
normed linear space, the above argument shows that you can obtain an infinite sequence of vectors {xn } , each kxn k = 1 and such that dist (xn+1 , span (x1 , , xn ))
1/2 for every n. Therefore, there is no convergent subsequence because no subsequence
is a Cauchy sequence. Therefore, the unit ball is not sequentially compact.
11. Suppose (A) < 1 for A L (V, V ) where V is a p dimensional vector space having
a norm ||||. You can use Rp or Cp if you like. Show there exists a new norm ||||||
such that with respect to this new norm, |||A||| < 1 where |||A||| denotes the operator
norm of A taken with respect to this new norm on V ,
|||A||| sup {|||Ax||| : |||x||| 1}
Hint: You know from Gelfands theorem that
||An ||
1/n
<r<1
provided n is large enough, this operator norm taken with respect to ||||. Show there
exists 0 < < 1 such that
A
< 1.
You can do this by arguing the eigenvalues of A/ are the scalars / where (A).
Now let Z+ denote the nonnegative integers.
n
A
|||x||| sup n x
nZ+
156
Exercises
First show this is actually a norm. Next explain why
n+1
A
|||Ax||| sup n+1 x |||x||| .
nZ+
Just pick larger than the absolute value of all the eigenvalues which are each less
than 1. Now use the norm suggested. First of all, why is this a norm? If |||x||| = 0,
then kxk = 0 because
kxk |||x|||
It is also clear that |||cx||| = |c| |||x|||. What about the triangle inequality?
n
n
n
A
A
A
|||x + y||| sup
n (x + y)
sup
n x
+
n y
nZ+
nZ+
n
n
A
A
n y
|||x||| + |||y|||
+
sup
sup
x
n
nZ
nZ
+
Now why is the operator norm with respect to this new norm less than 1?
n
n+1
A
A
|||Ax||| sup n Ax = sup n+1 x |||x|||
nZ+
nZ+
12. Establish a similar result to Problem 11 without using Gelfands theorem. Use an
argument which depends directly on the Jordan form or a modification of it.
(A) < 1 and A = S 1 J S where J is in modified Jordan form having or 0 on the
super diagonal the diagonal entries of every block being less than 1 in absolute value.
Let < 1 be larger than the absolute values of all eigenvalues. Now what if you define
|||x||| kSxk ? Then
|||Ax|||
kSAxk = kJ Sxk
max i (Sx)i + (Sx)i+1
157
Exercises
P
14. A matrix A is diagonally dominant if |aii | > j6=i |aij | . Show that the Gauss Seidel
method converges if A is diagonally dominant.
It is the eigenvalues of B 1 C which are important. Say
B 1 Cx = x
Then you must have
det (B C) = 0
Illustrating what happens in the Gauss Seidel method in three dimensions, this requires
In other words the above matrix must have a zero determinant. However, since the
original matrix was given to be diagonally dominant, this matrix cannot have a zero
determinant unless || < 1.
P
15. Suppose f () = k=0 an n converges if || < R. Show that if (A) < R where A is
an n n matrix, then
X
f (A)
an An
k0
converges in L (Fn , Fn ) . Hint: Use Gelfands theorem and the root test.
Since the given series converges if || < R, it follows that
lim sup |an n |
1/n
<1
and so
lim sup |an |
1/n
|| < 1 if || < R
1/n
n 1/n
1
R
1/n
1/n
kAn k
1/n
<
1
R=1
R
n
n
1/n
1/n
158
Exercises
Now letting z = u + iv where u, v are real valued, show
u + b2 u
= 0, u (0) = 1, u (0) = 0
v + b2 v
= 0, v (0) = 0, v (0) = b.
Next show u (t) = cos (bt) and v (t) = sin (bt) work in the above and that there is at
most one solution to
w + b2 w = 0 w (0) = , w (0) = .
Thus z (t) = cos (bt) + i sin (bt) and so y (t) = eat (cos (bt) + i sin (bt)). To show there
is at most one solution to the above problem, suppose you have two, w1 , w2 . Subtract
them. Let f = w1 w2 . Thus
f + b2 f = 0
and f is real valued. Multiply both sides by f and conclude
!
2
2
d (f )
2f
+b
=0
dt
2
2
Thus the expression in parenthesis is constant. Explain why this constant must equal
0.
The first claim is a routine computation. The next claim is also routine computations.
What about the uniqueness assertion? It suffices to show that if , = 0 then the
only solution is 0. Multiply both sides of the differential equation by w . Then you get
It follows that
1 d
b2 d " 2
2
(w ) +
w =0
2 dt
2 dt
2
(w ) + w2
is a constant. From the initial conditions, this constant can only be 0. The rest follows.
17. Let A L (Rn , Rn ) . Show the following power series converges in L (Rn , Rn ).
k k
X
t A
k=0
k!
You might want to use Lemma 14.4.2. This is how you can define exp (tA). Next show
using arguments like those of Corollary 14.4.3
d
exp (tA) = A exp (tA)
dt
so that this is a matrix valued solution to the differential equation and initial condition
(t) = A (t) , (0) = I.
This (t) is called a fundamental matrix for the differential equation y = Ay. Show
t (t) y0 gives a solution to the initial value problem
y = Ay, y (0) = y0 .
159
Exercises
The series converges absolutely by the ratio test and the observation that
Ak
k
kAk . Now we need to show that you can differentiate it. Writing the difference
quotient gives
k
k
k1 k
k1 k
(t
+
h)
t
Ak
X
X
X
1
ks (k, h)
A
s (k, h)
A
=
=
h
k!
k!
(k 1)!
k=0
k=1
k=1
where s (k, h) is between t and t + h. Thus for each k, limh0 s (k, h) = t. I want to
P k1 Ak
show that the limit of the series is k=1 t(k1)!
.
k1 k
X
s (k, h)
A
k=1
(k 1)!
k1
s (k, h)
tk1 Ak
X
X
tk1 Ak
=
(k 1)!
(k 1)!
k=1
k=1
X
(k 1) p (k, h)k2 (s (k, h) t) Ak
(k 1)!
k=1
X
p (k, h)k2 (s (k, h) t) Ak
(k 2)!
k=2
where p (k, h) is also between t and t + h. Now the norm of this is dominated by
|h|
k2
k
X
|p (k, h)|
kAk
(k 2)!
k=2
k1 k
X
s (k, h)
A
k=1
(k 1)!
X
tk1 Ak
0
(k 1)!
k=1
k1 k1
X
X
tk1 Ak
t
A
=A
= A exp (At)
(k 1)!
(k 1)!
k=1
k=1
When you plug in t = 0, you get I. Therefore, this is the solution to the given
differential equation.
18. In Problem 17 (t) is defined by the given series. Denote by exp (t (A)) the numbers
exp (t) where (A) . Show exp (t (A)) = ( (t)) . This is like Lemma 14.4.7.
Letting J be the Jordan canonical form for A, explain why
(t)
k k
X
t A
k=0
k!
=S
k k
X
t J
k=0
k!
S 1
and you note that in J k , the diagonal entries are of the form k for an eigenvalue
of A. Also J = D + N where N is nilpotent and commutes with D. Argue then that
k k
X
t J
k=0
k!
is an upper triangular matrix which has on the diagonal the expressions et where
(A) . Thus conclude
( (t)) exp (t (A))
160
Exercises
Next take et exp (t (A)) and argue it must be in ( (t)) . You can do this as
follows:
(t) et I
k k
k k
X
X
t A
t
I
k!
k!
k=0
k
X
k=0
k k1
X
t X
k=0
k!
k k1
X
t X
k=0
k=0
t k
A k I
k!
k!
j=1
Akj j (A I)
(6.38)
Akj j
j=1
converges to something in L (R , R ). To do this, use the ratio test and Lemma 14.4.2
after first using the triangle inequality. Since (A) , (t) et I is not one to one
and so this establishes the other inclusion. You fill in the details. This theorem is a
special case of theorems which go by the name spectral mapping theorem.
P (tA)k x
If (A) , then Ax = x and so
= et x and so exp (t (A))
k=0
k!
(exp (tA)) . Now consider exp (tA) . You have A = S 1 JS where J is Jordan form.
Thus it is simple to write that
exp (tA) = S 1
k
X
(tJ)
k=0
k!
Now J is block diagonal and the blocks have a constant down the main diagonal. Say
..
.
. ..
When you raise such a block to the exponent k, you get an upper triangular matrix
which has k down the diagonal. Therefore, exp (tA) is of the form
S 1 BS
where B is an upper triangular matrix which has for its diagonal entries et for
(A). These diagonal entries are the eigenvalues of exp (tA). It follows that
(exp (tA)) exp (t (A)) .
19. Suppose (t) L (V, W ) where V, W are finite dimensional inner product spaces and
t (t) is continuous for t [a, b]: For every > 0 there there exists > 0 such that
if |s t| < then || (t) (s)|| < . Show t ( (t) v, w) is continuous. Here it is
the inner product in W. Also define what it means for t (t) v to be continuous
and show this is continuous. Do it all for differentiable in place of continuous. Next
show t || (t)|| is continuous.
161
Exercises
As to continuity of t k (t)k ,
|k (t)k k (s)k| k (t) (s)k
from the triangle inequality.
20. If z (t) W, a finite dimensional inner product space, what does it mean for t z (t)
to be continuous or differentiable? If z is continuous, define
Z
as follows.
w,
z (t) dt W
z (t) dt
a
Show that this definition is well defined and furthermore the triangle inequality,
Z
Z
b
b
z (t) dt
|z (t)| dt,
a
a
and fundamental theorem of calculus,
Z t
d
z (s) ds = z (t)
dt
a
hold along with any other interesting properties of integrals which are true.
Differentiability is the same as usual.
z (t + h) z (t) z (t) h = o (h)
If t z (t) is continuous, then from the above problem, so is t (z (t) , w) and so
Rb
(z (t) , w) dt makes sense. Does there exist an element I of the inner product space
a
Rb
such that (w, I) = a (w, z (t)) dt? This is so in a finite dimensional inner product
space because
Z b
w
(w, z (t)) dt
a
162
Exercises
is a linear map and so it can be represented by a unique I W so that
Z b
(w, I) =
(w, z (t)) dt
a
Since the space is finite dimensional, this is the same as saying that
R t+h
z (s) ds
lim t
= z (t) .
h0
h
You can just apply the above to the finitely many Fourier coefficients. Therefore, the
usual Fundamental theorem of calculus holds. What about the triangle inequality?
Z
!
Z b
b
z (t) dt = sup w,
z (t) dt
a
|w|=1
a
Z
Z b
b
(w, z (t)) dt sup
|(w, z (t))| dt
= sup
|w|=1 a
|w|=1 a
Z b
|z (t)| dt.
a
as follows.
w,
(t) dt (v)
Rb
Show this is well defined and does indeed give a (t) dt L (V, W ) . Also show the
triangle inequality
Z
Z
b
b
(t) dt
|| (t)|| dt
a
a
where |||| is the operator norm and verify the fundamental theorem of calculus holds.
Z t
(s) ds = (t) .
a
Also verify the usual properties of integrals continue to hold such as the fact the
integral is linear and
Z b
Z c
Z c
(t) dt +
(t) dt =
(t) dt
a
and similar things. Hint: On showing the triangle inequality, it will help if you use
the fact that
|w|W = sup |(w, v)| .
|v|1
163
Exercises
You should show this also.
That last assertion follows by the Cauchy Schwarz inequality which shows the right
side is no larger than |w| and then by taking v = w/ |w|. That the integral is well
Rb
defined follows because w a (w, (t) v) dt is a linear mapping and so, by the
Riesz representation theorem, there exists a unique I (v) W such that (w, I (v)) =
Rb
(w, (t) v) dt. Now also, I is a linear function of v because of the fact that each
a
Rb
(t) is linear. Therefore, you can denote by a (t) dt L (V, W ) this I. Thus the
definition of the integral is well defined. The standard properties of the integral are
now fairly obvious. As for the triangle inequality, it goes the same way as in the above
problem.
Z
Z
!
Z b
b
b
(t) dt
sup
(t) dt (v)
= sup sup w,
(t) dt (v)
a
|v|1
|v|1 |w|1
a
a
Z
Z
b
b
= sup sup
(w, (t) (v)) dt
k (t)k dt.
|v|1 |w|1 a
a
Now consider the fundamental theorem of calculus. For simplicity, let h 0 + .
Z
Z
1 t+h
1 t+h
(s) ds (t)
=
(s) (t) ds
h t
h t
Z t+h
1
k (s) (t)k ds
h t
Ku (s) ds.
" Kt
we
u0 eKt
and therefore,
w (t) e
Kt
u0
Ks
ds = u0
1
eKt
K
K
164
Exercises
Therefore, w (t) u0
eKt
K
1
K
u (t) u0 + K u0
eKt
1
K
K
= u0 eKt
Az (s) ds
and so
||z (t)||
Hint: A nice way to approach this problem is to show you can reduce it to the
consideration of the initial value problem
z = J z, z (0) = z0
where J is the modified Jordan canonical form where instead of ones down the main
diagonal, there are down the main diagonal. You have y = S 1 J Sy and so you
just let z = Sy. Then z 0 if and only if y 0.
Then
z = Dz + N z
165
Exercises
where D is the diagonal matrix obtained from the eigenvalues of A and N is a nilpotent
matrix commuting with D which is very small provided is chosen very small. Now
let (t) be the solution of
= D, (0) = I
described earlier as
X
(1)k tk Dk
.
k!
k=0
The reason these commute is that D is block diagonal, each block having constant
entries down the diagonal while the same is true of N . The blocks associated with N
commute with the blocks associated with D and so D and N commute. Thus (t)
commutes with D and N .
Next argue
Then integrating,
(t) z (t) z0 =
e 1
0
..
(t) =
.
n t
0
e
and now observe that if is chosen small enough, ||N || is so small that each component
of z (t) converges to 0.
It follows right away from the definition of (t) that (t) is given by the above
diagonal matrix where here the i are negatives of the eigenvalues of A. Thus their
real parts are all bounded below by some > 0. It follows that
|( (t) x)i | et |xi |
and so
|| (t) z (t)|| et kz (t)k
166
Exercises
Therefore, from the above inequality,
kz (t)k kz0 k et + et
Of course the right side converges to 0 if you choose small enough that kN k < .
25. Using Problem 24 show that if A is a matrix having the real parts of all eigenvalues
less than 0 then if
(t) = A (t) , (0) = I
it follows
lim (t) = 0.
Show (t) = (nt) . Hint: Subtract and show the difference satisfies = A, (0) =
0. Use uniqueness.
27. If the real parts of the eigenvalues of A are all negative, show that for every positive
t,
lim (nt) = 0.
n
Hint: Pick Re ( (A)) < < 0 and use Problem 18 about the spectrum of (t)
and
for the spectral radius along with Problem 26 to argue that
theorem
Gelfands
(nt) /ent < 1 for all n large enough.
n
P
28. Let H be a Hermitian matrix. (H = H ) . Show that eiH n=0 (iH)
is unitary.
n!
You have H = U DU where U is unitary and D is a real diagonal matrix. Then you
have
i
e 1
n
X
(iD)
..
eiH = U
U = U
U
.
n!
n=0
ein
and this is clearly unitary because each matrix in the product is.
29. Show the converse of the above exercise. If V is unitary, then V = eiH for some H
Hermitian.
To do this, note first that V is normal because V V = V V = I. Therefore, there
exists a unitary U such that V = U DU where D is the diagonal matrix consisting
of the eigenvalues of V down the main diagonal. Since V is unitary, it preserves all
lengths and so each diagonal entry of D is of magnitude 1. Thus you have
i
e 1
k
X
(iD)
.
.
V =U
U
U = U
.
k!
k=0
ein
167
Exercises
where D =
..
.
n
k
X
(iU DU )
k=0
= eiH
k!
exists, show
U = (I + iH) (I iH)
To see this, note first that (I U ) , (I + U ) commute. Taking the adjoint of the
above operator, it is easy to see that this equals
i (I + U )
(I U )
= i (U U + U )
1
= i (U + I)
= i (I + U )
(U U U )
U U (U I)
(I U ) = i (I U ) (I + U )
31. Suppose that A L (V, V ) where V is a normed linear space. Also suppose that
kAk < 1 where this refers to the operator norm on A. Verify that
(I A)1 =
Ai
i=0
This is called the Neumann series. Suppose now that you only know the algebraic
1
condition (A) < 1. Is it still the case that the Neumann series converges to (I A) ?
Consider partial sums of the series.
q
q
q
X i
X
i
X
i
A
A
kAk
i=p
i=p
i=p
P
i
1
which converges to 0 as p, q because i=0 kAk = 1kAk
< . Therefore, by
completeness of L (V, V ) , this series converges. Why is is the desired inverse? This is
obvious when you multiply both sides by (I A) . Thus
(I A)
X
i=0
Ai = lim (I A)
n
n
X
i=0
Ai = lim I An+1 = I
n
168
Exercises
n+1
because
An+1
kAk
and this converges to 0 as n because kAk < 1. Yes,
it the series still converges. Let (A) < r < R < 1. Then for all n large enough, you
have
1/n
kAn k
<R
and so kAn k < Rn . It follows that the partial sums of the series form a Cauchy
sequence and so the series converges.
F.34
Exercises
15.3
1. In Example 15.1.10 an eigenvalue was found correct to several decimal places along
with an eigenvector. Find the other eigenvalues along with their eigenvectors.
1 2 3
The matrix was 2 2 1
3 1 4
20
1 2 3
8. 485 7 1015 6. 190 4 1015 1. 188 8 1016
2 2 1 = 6. 190 4 1015 4. 515 9 1015 8. 672 7 1015
3 1 4
1. 188 8 1016 8. 672 7 1015 1. 665 6 1016
16
1. 586 4 10
1. 157 3 1016 2. 222 5 1016
0
2. 212 9 1011 8. 403 1 1011
0
0
4. 160 5 1011
6. 662 1
1. 218 8 104 7. 374 9 105
1. 139 5
1. 156 9
= 1. 218 8 104
7. 374 9 105
1. 156 9
0.801 49
0.534 92
The largest eigenvalue is close to 6. 662 1 and an eigenvector is 0.390 23 . The
0.749 39
others can be found by looking at the lower right block. You could do the same as the
above or you could just graph the characteristic function and zoom in on the roots.
They are approximately 1.7 and 1.3. Next use shifted inverse power method to find
them more exactly and to also find the eigenvectors.
1 2 3
1 0 0
0.977 92 5. 047 3
3. 47
2 2 1 1.7 0 1 0 = 5. 047 3 33. 47
21. 136
3 1 4
0 0 1
3. 47
21. 136 13. 281
169
Exercises
Rather than use this method as described, I will just use the variation of it based on
the QR algorithm which involves raising the matrix to a power which was just used
to find the first eigenvalue.
20
0.977 92 5. 047 3
3. 47
5. 047 3 33. 47
21. 136 =
3. 47
21. 136 13. 281
6. 044 8 1031
3. 893 4 1032 2. 458 8 1032
3. 893 4 1032
2. 507 7 1033 1. 583 7 1033 :
32
2. 458 8 10
1. 583 7 1033 1. 000 2 1033
0.529 98
0.127 88
2
0.529 42
0.847 36
4. 112 3 10
32
33
4. 644 3 10
2. 991 4 10
1. 889 2 1033
0
9. 738 8 1027 3. 859 8 1028
0
0
3. 342 9 1027
T
0.130 15 3. 322 8 102
0.990 94
0.838 32
0.529 98
0.127 88
2
0.529 42
0.847 36
4. 112 3 10
0.977 92 5. 047 3
3. 47
5. 047 3 33. 47
21. 136
3. 47
21. 136 13. 281
47. 602
6. 755 1 105 2. 994 7 105
6. 755 1 105 6. 320 5 102
0.232 90
5
2. 994 7 10
0.232 90
0.190 38
1
1.7
0.130 15
suitable eigenvector is just the first column of the orthogonal matrix 0.838 32 .
0.529 42
How
well
does
it
work?
1 2 3
0.130 15
0.218 53
2 2 1 0.838 32 = 1. 407 5
3 1 4
0.529 42
0.888 91
0.130 15
0.218 52
0.838 32 (1. 679 0) = 1. 407 5
0.529 42
0.888 90
1
1 2 3
1 0 0
2 2 1 + 1.3 0 1 0
3 1 4
0 0 1
170
Exercises
20
16. 948 7. 810 9
8. 119 2
7. 810 9 3. 278 5 3. 802 7 =
8. 119 2 3. 802 7 3. 689 6
0.801 56
0.397 63 0.700 98
0.592 05
27
4. 584 7 10
2. 090 8 1027 2. 183 7 1027
0
1. 658 9 1022
2. 792 9 1022
0
0
4. 620 1 1021
T
0.834 83 0.544 13 8. 355 9 102
0.380 73 0.461 03
0.801 56
0.397 63 0.700 98
0.592 05
0.801 56
0.397 63 0.700 98
0.592 05
24. 377
9. 597 9 105
5. 267 105
0.127 02
1. 802 1 102
= 9. 597 9 105
5
2
5. 267 10
1. 802 1 10
0.334 17
1
+1.3
= 24. 377, Solution is: 1. 341. The approximate eigenvector is the first column.
How
well
does it work?
1 2 3
0.834 83
1. 119 5
2 2 1 0.380 73 = 0.510 57
3 1 4
0.397 63
0.533 24
0.834 83
1. 119 5
0.380 73 (1. 341) = 0.510 56
0.397 63
0.533 22
It works well. You could do more iterations if desired.
3 2 1
2. Find the eigenvalues and eigenvectors of the matrix A = 2 1 3 numerically.
1 3 2
In this case the exact eigenvalues are 3, 6. Compare with the exact answers.
13
3 2 1
4. 353 6 109 4. 353 6 109 4. 353 6 109
2 1 3 = 4. 353 6 109 4. 353 6 109 4. 353 6 109 =
1 3 2
4. 353 6 109 4. 353 6 109 4. 353 6 109
0.707 11
0.577 35 0.408 25
0.707 11
7. 540 7 109
7. 540 7 109
7. 540 7 109
0
1. 533 3 1019 1. 533 3 1019
0
0
2. 736 9 1048
Exercises
171
29
0.577 35 0.816 50 1. 437 8 10
0.577 35 0.408 25
0.707 11
0.577 35 0.408 25
0.707 11
6. 000 0
0
0
1. 5
0.866 03
= 1. 262 2 1029
29
2. 524 4 10
0.866 03
1. 5
0.577 35
An eigenvalue is 6 and an eigenvector is 0.577 35 . Now lets find the others.
0.577 35
1
3 2 1
1 0 0
2 1 3 1.5 0 1 0
1 3 2
0 0 1
2. 740 7
0.592 59
1. 925 9
= 0.592 59 7. 407 4 102 0.740 74
1. 925 9
0.740 74
1. 407 4
20
2. 740 7
0.592 59
1. 925 9
0.592 59 7. 407 4 102 0.740 74 =
1. 925 9
0.740 74
1. 407 4
12
3. 034 1 10
8. 129 9 1011 2. 221 1 1012
8. 129 9 1011 2. 178 4 1011
5. 951 5 1011 :
12
11
2. 221 1 10
5. 951 5 10
1. 626 0 1012
0.788 68
0.425 47 0.443 81
0.211 33 0.490 28 0.845 56
0.577 35 0.760 66 0.296 76
0
3. 850 7 106
3. 843 7 107
0
0
1. 927 6 107
0.788 68
0.425 47 0.443 81
2. 740 7
0.592 59
1. 925 9
0.211 33 0.490 28 0.845 56 0.592 59 7. 407 4 102 0.740 74
0.577 35 0.760 66 0.296 76
1. 925 9
0.740 74
1. 407 4
0.788 68
0.425 47 0.443 81
0.211 33 0.490 28 0.845 56
0.577 35 0.760 66 0.296 76
4. 309 4
9. 614 1 106 1. 276 5 105
= 9. 614 1 106
0.223 59
0.195 59
1. 276 5 105
0.195 59
0.136 42
0.788 68
1
0.211 33 . How well
1.5 = 4. 309 4, Solution is: 1. 732 1 and an eigenvector is
0.577 35
does
it
work?
3 2 1
0.788 68
1. 366
2 1 3 0.211 33 = 0.366 02
1 3 2
0.577 35
1.0
172
Exercises
0.788 68
1. 366 1
0.211 33 1. 732 1 = 0.366 04
0.577 35
1.0
Next you can find the other one.
1
3 2 1
1 0 0
2 1 3 + 1.5 0 1 0
1 3 2
0 0 1
2
4. 444 4 10
0.711 11 0.622 22
0.711 11
2. 622 2
2. 044 4
=
0.622 22
2. 044 4
1. 288 9
20
4. 444 4 102 0.711 11 0.622 22
0.711 11
2. 622 2
2. 044 4 =
0.622 22
2. 044 4
1. 288 9
11
2. 178 4 10
8. 129 9 1011 5. 951 5 1011
8. 129 9 1011 3. 034 1 1012 2. 221 1 1012 :
5. 951 5 1011 2. 221 1 1012 1. 626 0 1012
12
1. 030 8 10
3. 847 1 1012 2. 816 3 1012
0
1. 437 1 107
2. 791 7 107
0
0
1. 927 6 107
T
0.211 32 0.490 28 0.845 56
0.788 68 0.425 47 0.443 81
0.577 35
0.760 66 0.296 76
0.711 11
2. 622 2
2. 044 4 0.788 68 0.425 47 0.443 81 :
0.622 22
2. 044 4
1. 288 9
0.577 35
0.760 66 0.296 76
4. 309 4
2. 468 3 106 1. 167 8 105
2. 468 3 106
0.280 95
6. 479 4 102
5
2
1. 167 8 10
6. 479 4 10
0.161 74
0.211 32
1
0.788 68 . How
+1.5 = 4. 309 4, Solution is: 1. 732 1 and an eigenvector is
0.577 35
well
does
it
work?
3 2 1
1 0 0
0.211 32
2 1 3 + 1. 732 1 0 1 0 0.788 68
1 3 2
0 0 1
0.577 35
5
2. 262 8 10
= 6. 262 8 105 which is pretty close to 0 so this worked well.
7. 935 106
3 2 1
3. Find the eigenvalues and eigenvectors of the matrix A = 2 5 3 numerically.
1 3 2
The exact eigenvalues are 2, 4 + 15, 4 15. Compare your numerical results with
the exact values. Is it much fun to compute the exact eigenvectors?
Exercises
173
20
3 2 1
2 5 3 =
1 3 2
0.415 99
0.806 58
0.419 97
0.779 18 7. 803 9 102 0.621 92
0.468 86
0.585 95
0.660 94
17
17
3. 482 5 10
6. 522 6 10
3. 924 8 1017
0
1. 539 1013 1. 324 1 1013
0
0
1. 399 9 1012
3 2 1
0.415 99
0.806 58
0.419 97
0.779 18 7. 803 9 102 0.621 92 2 5 3
1 3 2
0.468 86
0.585 95
0.660 94
0.415 99
0.806 58
0.419 97
0.779 18 7. 803 9 102 0.621 92
0.468 86
0.585 95
0.660 94
7. 873 0
8. 768 9 105 3. 295 6 105
1. 746 2
0.641 05
= 8. 768 9 105
5
3. 295 6 10
0.641 05
0.380 82
0.415 99
An eigenvalue is 7. 873 0 and the approximate eigenvector is 0.779 18 . How well
0.468 86
does
it
work?
3 2 1
1 0 0
0.415 99
2 5 3 7. 873 0 0 1 0 0.779 18
1 3 2
0 0 1
0.468 86
4
1. 007 3 10
= 2. 414 105
8. 478 105
It works pretty well. Next we find the other eigenvalues and eigenvectors which go
with them.
1
3 2 1
1 0 0
2 5 3 1.75 0 1 0 =
1 3 2
0 0 1
3. 295 6 1. 006 3
1. 106 9
1. 006 3 0.276 73
0.704 4
2
1. 106 9 0.704 4 2. 515 7 10
20
3. 295 6 1. 006 3
1. 106 9
1. 006 3 0.276 73
0.704 4
1. 106 9 0.704 4 2. 515 7 102
174
Exercises
0
2. 995 0 106
1. 376 106
0
0
5. 368 8 105
0.704 4
2
0.301 51 0.384 47 0.872 51
1. 106 9 0.704 4 2. 515 7 10
4. 000 0
1. 828 8 106 8. 371 4 106
1. 828 8 106
0.155 70
7. 669 2 102
6
2
8. 371 4 10
7. 669 2 10
0.608 53
0.904 53
1
0.301 52 . How well does it
1.75 = 4, Solution is: 2.0. An eigenvector is then
0.301 51
work?
3 2 1
1 0 0
0.904 53
0.000 02
2 5 3 2 0 1 0 0.301 52 = 0.000 03
1 3 2
0 0 1
0.301 51
0.000 03
1
3 2 1
1 0 0
2 5 3 0.38 0 1 0 =
1 3 2
0 0 1
0.493 54
7. 815 4 102 0.449 38
7. 815 4 102
1. 056 5
1. 908 3
0.449 38
1. 908 3
2. 639 1
20
2
0.493 54
7. 815 4 10
0.449 38
7. 815 4 102
1. 056 5
1. 908 3 =
0.449 38
1. 908 3
2. 639 1
7. 593 8 109
4. 459 9 1010
6. 738 1010
4. 459 9 1010 2. 619 4 1011 3. 957 3 1011 :
6. 738 1010
3. 957 3 1011 5. 978 7 1011
0.549 53
0.834 58 3. 863 5 102
0.830 22
0.540 31
0.137 12
10
11
8. 115 9 10
4. 766 6 10
7. 201 3 1011
0
4. 418 3 106
1. 380 8 106
0
0
6. 663 6 105
T
9. 356 7 102 0.107 43
0.989 8
0.549 53
0.834 58 3. 863 5 102
0.830 22
0.540 31
0.137 12
2
0.493 54
7. 815 4 10
0.449 38
7. 815 4 102
1. 056 5
1. 908 3
0.449 38
1. 908 3
2. 639 1
175
Exercises
0.549 53
0.834 58 3. 863 5 102 =
0.830 22
0.540 31
0.137 12
3. 952 9
2. 044 4 105 1. 021 4 105
2. 044 4 105
0.182 25
0.145 62
1. 021 4 105
0.145 62
0.568 55
1
.38 = 3. 952 9,Solution is: 0.127 02 and an approximate eigenvector is then
9. 356 7 102
3 2 1
1 0 0
9. 356 7 102
2 5 3 0.127 02 0 1 0
0.549 53
1 3 2
0 0 1
0.830 22
2. 388 105
= 5. 469 9 105
3. 754 4 105
0 2 1
4. Find the eigenvalues and eigenvectors of the matrix A = 2 5 3 numerically.
1 3 2
I dont know the exact eigenvalues in this case. Check your answers by multiplying
your numerically computed eigenvectors by the matrix.
50
0 2 1
5. 045 4 1042 1. 454 4 1043 8. 826 9 1042
2 5 3 = 1. 454 4 1043 4. 192 3 1043 2. 544 4 1043
1 3 2
8. 826 9 1042 2. 544 4 1043 1. 544 2 1043
0.284 32 0.756 78
0.588 6
= 0.819 59 0.510 39 0.260 32
0.497 42 0.408 40 0.765 36
0
7. 091 8 1038 8. 072 9 1037
0
0
3. 007 5 1038
0.284 32 0.756 78
0.588 6
0.819 59 0.510 39 0.260 32
0.497 42 0.408 40 0.765 36
0.284 32 0.756 78
0.588 6
0.819 59 0.510 39 0.260 32
0.497 42 0.408 40 0.765 36
7. 514 5
9. 130 9 105
9. 130 9 105
0.541 46
1. 248 7 105
0.344 28
0
2
1
2 1
5 3
3 2
1. 248 7 105
0.344 28
2
2. 686 9 10
0.284 32
Thus an eigenvalue is 7. 514 5 and an eigenvector is approximately 0.819 59 . How
0.497 42
well does it work?
176
Exercises
0 2 1
0.284 32
2. 136 6
2 5 3 0.819 59 = 6. 158 9
1 3 2
0.497 42
3. 737 9
0.284 32
2. 136 5
0.819 59 7. 514 5 = 6. 158 8 .
0.497 42
3. 737 9
This has essentially found it. However, we dont know the others yet. begin with the
one which is closest to 0.541 46
0 2 1
1 0 0
2 5 3 + .541 46 0 1 0
1 3 2
0 0 1
5. 323 8
2. 181 4
0.480 23
0.393 89 0.393 38
= 2. 181 4
0.480 23 0.393 38
1. 046 8
20
5. 323 8
2. 181 4
0.480 23
2. 181 4
0.393 89 0.393 38 =
0.480 23 0.393 38
1. 046 8
15
5. 483 4 10
2. 055 8 1015 2. 530 4 1014
2. 055 8 1015 7. 707 7 1014 9. 487 0 1013 :
2. 530 4 1014 9. 487 0 1013 1. 167 7 1013
0.935 48
0.353 06
1. 491 6 102
0.350 73
0.932 81
8. 286 3 102
2
2
4. 316 9 10
7. 228 5 10
0.996 45
15
15
5. 861 6 10
2. 197 6 10
2. 704 9 1014
0
2. 167 1 1010 1. 796 2 109
0
0
8. 270 5 107
T
0.935 48
0.353 06
1. 491 6 102
0.350 73
0.932 81
8. 286 3 102
2
2
4. 316 9 10
7. 228 5 10
0.996 45
5. 323 8
2. 181 4
0.480 23
2. 181 4
0.393 89 0.393 38
0.480 23 0.393 38
1. 046 8
0.935 48
0.353 06
1. 491 6 102
0.350 73
0.932 81
8. 286 3 102 =
2
2
4. 316 9 10
7. 228 5 10
0.996 45
5
6. 163 8
1. 742 0 10
1. 159 3 106
1. 742 0 105
0.513 51
0.577 10
1. 159 3 106
0.577 10
0.979 41
1
+.541 46
0.935 48
0 2 1
1 0 0
0.935 48
2 5 3 + 0.703 70 0 1 0
:
0.350 73
2
1 3 2
0 0 1
4. 316 9 10
177
Exercises
6. 276 106
8. 299 106
6. 025 3 106
0 2 1
1.0 1.0 1.0
2 5 3 = 1.0
1.0 2.0
1 3 2
1.0 2.0 4.0
20
1.0 1.0 1.0
1.0
1.0 2.0
1.0 2.0 4.0
1. 287 1 1013
2. 778 8 1013 5. 314 3 1013
5. 999 6 1013 1. 147 4 1014 :
= 2. 778 8 1013
13
5. 314 3 10
1. 147 4 1014 2. 194 2 1014
0
1. 509 3 109
1. 706 9 109
0
0
6. 192 3 109
T
0.209 85 0.930 85 0.299 13
0.453 05 0.178 55 0.873 42
0.866 43 0.318 81 0.384 26
5. 288 0
2. 017 9 105 2. 856 3 105
2. 017 9 105
0.916 86
0.727 58
5
2. 856 3 10
0.727 58
0.371 12
0.209 85
1
0.453 05 .
= 5. 288 0, Solution is: 0.189 11. An approximate eigenvector is
0.866 43
How well does it work?
0 2 1
1 0 0
0.209 85
1. 473 4 105
2 5 3 0.189 11 0 1 0 0.453 05 = 1. 628 6 105
1 3 2
0 0 1
0.866 43
9. 422 7 106
.
0 2 1
5. Find the eigenvalues and eigenvectors of the matrix A = 2 0 3 numerically.
1 3 2
I dont know the exact eigenvalues in this case. Check your answers by multiplying
your numerically computed eigenvectors by the matrix.
I will do this one a little differently. You can easily find that the eigenvalues are
approximately 4.9, 2.7, .3. To find the eigen pair which goes with .3 to the following.
1
0 2 1
1 0 0
2 0 3 .3 0 1 0 =
1 3 2
0 0 1
178
Exercises
1. 138 5
4. 788 7 102 0.754 22
4. 788 7 102
0.180 77
0.347 18
0.754 22
0.347 18
0.468 10
20
1. 138 5
4. 788 7 102 0.754 22
4. 788 7 102
0.180 77
0.347 18 =
0.754 22
0.347 18
0.468 10
17805.
3431. 1 12214.
3431. 1
661. 21 2353. 7 =
12214. 2353. 7 8378. 3
21863.
4213.0
14997.
0
2. 261 7 102 6. 348 1 102
0
0
0.281 26
A similar matrix is
T
0.814 41 0.321 53 0.483 08
0.156 94
0.923 48 0.350 07
0.558 67 0.209 28 0.802 55
1. 138 5
4. 788 7 102 0.754 22
4. 788 7 102
0.180 77
0.347 18
0.754 22
0.347 18
0.468 10
1. 665 1
7. 437 4 106 1. 163 3 105
7. 437 4 106
0.296 62
0.142 05
5
1. 163 3 10
0.142 05
0.174 36
Thus an eigenvalue is close to 1.6651 and so the eigenvalue for the original matrix
1
is the solution of .3
= 1.6651, Solution is: 0.300 56. Then what about the
eigenvector? We have
1
(A .3I)1 v =
v
.3
0 2 1
0.814 41
0.244 79
2 0 3 0.156 94 = 0.047 19
1 3 2
0.558 67
0.167 89
0.814 41
0.244 78
0.156 94 (0.300 56) = 4. 717 0 102
0.558 67
0.167 91
This worked very well. Now lets find the pair associated with the eigenvalue being
close to 2.7.
1
1 0 0
0 2 1
2 0 3 + 2.7 0 1 0
1 3 2
0 0 1
Exercises
20
7. 969 8 13. 823 7. 127 4
13. 823 25. 248 13. 175 =
7. 127 4 13. 175 7. 105 8
0.439 25
0.891 55
0.110 45
0.795 85
0.443 21
0.412 55
0.416 76 9. 330 8 102 0.904 21
0
7. 478 5 1026 3. 772 7 1026
0
0
3. 494 4 1026
T
0.439 25
0.891 55
0.110 45
0.795 85
0.443 21
0.412 55
0.416 76 9. 330 8 102 0.904 21
39. 777
1. 791 2 104 6. 881 105
1. 791 2 104
0.336 06
0.128 95
5
6. 881 10
0.128 95
0.210 75
179
0.110 45
0.412 55 =
0.904 21
1
Eigenvalue, +2.7
= 39. 777, Solution is: 2. 674 9. The eigenvector would be the first
column of the above orthogonal matrix on the right.
0 2 1
0.439 25
1. 174 9
2 0 3 0.795 85 = 2. 128 8
1 3 2
0.416 76
1. 114 8
0.439 25
1. 174 9
0.795 85 (2. 674 9) = 2. 128 8 This worked very well.
0.416 76
1. 114 8
1
0 2 1
1 0 0
2 0 3 4.9 0 1 0
1 3 2
0 0 1
40
1. 753 6 2. 962 0 3. 668 8
2. 962 0 4. 446 3 5. 621 =
3. 668 8 5. 621 6. 735 1
180
Exercises
0.379 20
0.739 6
0.556 06
0.584 81 0.657 26 0.475 40
0.717 08 0.144 92
0.681 76
44
3. 018 2 10
4. 654 7 1044
0
6. 539 3 1039
0
0
0.379 20
0.739 6
0.556 06
0.584 81 0.657 26 0.475 40
0.717 08 0.144 92
0.681 76
0.379 20
0.739 6
0.556 06
0.584 81 0.657 26 0.475 40
0.717 08 0.144 92
0.681 76
13. 259
1. 247 5 104
1. 247 5 104
0.142 61
1. 129 2 105 2. 290 2 102
1
4.9
5. 707 5 1044
5. 579 8 1039
4. 797 6 1039
T
1. 129 2 105
2. 290 2 102
0.181 74
0 2 1
0.379 20
1. 886 7
2 0 3 0.584 81 = 2. 909 6
1 3 2
0.717 08
3. 567 8
0.379 20
1. 886 7
0.584 81 4. 975 4 = 2. 909 7 . It worked well.
0.717 08
3. 567 8
3 2 3
T
6. Consider the matrix A = 2 1 4 and the vector (1, 1, 1) . Find the shortest
3 4 0
distance between the Rayleigh quotient determined by this vector and some eigenvalue
of A.
T
1
3 2 3
1
q = 1 2 1 4 1 13 = 7. 333 3
1
3 4 0
1
3 2 3
1
1
2 1 4 1 7. 333 3 1
3 4 0
1
1
|7. 333 3 q |
3
= 0.471 41
1 2 1
T
7. Consider the matrix A = 2 1 4 and the vector (1, 1, 1) . Find the shortest
1 4 5
distance between the Rayleigh quotient determined by this vector and some eigenvalue
of A.
181
Exercises
T
1
1
q= 1 2
1
1
2 1
1
1 4 1 13 = 7
4 5
1
|7 q | = 2. 449 5
3 2 3
T
8. Consider the matrix A = 2 6 4 and the vector (1, 1, 1) . Find the shortest
3 4 3
distance between the Rayleigh quotient determined by this vector and some eigenvalue
of A.
T
1
3 2 3
1
q = 1 2 6 4 1 13 = 8.0
1
3 4 3
1
|q 8|
9.
Using
3
2
3
3
2
3
3. 266 0
2
6
4
3
1
1
4 1 8 1
3
1
1
Gerschgorins
theorem, find upper and lower bounds for the eigenvalues of A =
2 3
6 4 .
4 3
10 12
10. Tell how to find a matrix whose characteristic polynomial is a given monic polynomial.
This is called a companion matrix. Find the roots of the polynomial x3 + 7x2 + 3x + 7.
0 0 7
The companion matrix is 1 0 3
0 1 7
20
0 0 7
8. 062 3 1014 5. 408 5 1015 3. 628 2 1016
1 0 3 = 2. 253 5 1014 1. 511 7 1015 1. 014 1 1016
0 1 7
7. 726 4 1014 5. 183 2 1015 3. 477 1016
0.707 72 0.258 58
0.657 47
= 0.197 82 0.820 86 0.535 78
0.678 23 0.509 24 0.529 79
0
4. 570 4 1010 2. 079 1 1010
0
0
3. 151 4 1011
0.707 72 0.258 58
0.657 47
0 0 7
0.197 82 0.820 86 0.535 78 1 0 3
0.678 23 0.509 24 0.529 79
0 1 7
0.707 72 0.258 58
0.657 47
0.197 82 0.820 86 0.535 78
0.678 23 0.509 24 0.529 79
182
Exercises
6. 708 3
= 4. 147 2 105
1. 391 6 105
5. 850 6
5. 220 9
0.154 76
1. 187 6
0.936 81 0.446 46
Clearly a real root is close to 6. 708 3. Then the other roots can be obtained from
the lower right block.
0.154 76
1. 187 6
, eigenvalues: 0.145 85 + 1. 011i, 0.145 85 1. 011i
0.936 81 0.446 46
How well do these work? Try the last one. Evaluating the polynomial at this value of
x gives
6. 350 5 104 + 2. 065 8 104 i
36482.
27300.0
49899.
2. 448 9 105
14010.0
9182.0
77199.
1. 949 9 105
0.260 30
6. 965 1 102
0.941 85
0.200 79
9. 996 0 102
0.251 72
0.209 32
0.939 59
0.940 38
0.202 95
0.253 14
0.102 07
0.194 78
0.943 71
7. 090 6 102 0.257 75
0
70622. 2. 084 2 105 3. 625 1 105
0
0
2. 130 5
5. 038
0
0
0
2. 683 4
T
0.260 30
6. 965 1 102
0.941 85
0.200 79
9. 996 0 102
0.251 72
0.209 32
0.939 59
0.940 38
0.202 95
0.253 14
0.102 07
0.194 78
0.943 71
7. 090 6 102 0.257 75
0 0 0 1
1 0 0 1
0 1 0 4
0 0 1 3
0.260 30
6. 965 1 102
0.941 85
0.200 79
9. 996 0 102
0.251 72
0.209 32
0.939 59
0.940 38
0.202 95
0.253 14
0.102 07
0.194 78
0.943 71
7. 090 6 102 0.257 75
0.613 47
4. 251 0
0.485 69
2. 096 8
0.503 89
2. 337 6
0.115 42 0.330 94
11. Find
0
1
0
0
183
Exercises
0.157 34 0.304 46
0.974 48 0.108 46
, eigenvalues: 0.024 44 + 0.528 23i, 0.024 44 0.528 23i