MATH 110: LINEAR ALGEBRA

HOMEWORK #11
DAVID ZYWINA
5.2
Problem 3. In each of these questions we choose a standard basis , and the use [T]

to
compute the eigenvalues of T. We then nd a basis of eigenvectors of T (if possible); this
is the required basis.
(a) Let = 1, x, x
2
, x
3
be the standard basis. Then
[T]

=
_
_
_
_
0 1 2 0
0 0 2 6
0 0 0 3
0 0 0 0
_
_
_
_
.
The eigenvalues of [T]

are 0 (with multiplicity 4), but dimE

) = 1. Therefore [T]

(and hence T) is not diagonalizable.

[T]

=
_
_
0 0 1
0 1 0
1 0 0
_
_
.
The eigenvalues of T are the roots of the polynomial
det([T]

I) = det
_
_
0 1
0 1 0
1 0
_
_
= (1 ) det
_
1
1
_
= ( 1)
2
( + 1).
A vector ax
2
+bx+c is in E
1
(T) if cx
2
+bx+a = ax
2
+bx+a. Thus E
1
(T) has dimension
2 and is spanned by x, x
2
+ 1.
A vector ax
2
+ bx + c is in E
1
(T) if cx
2
+ bx + a = ax
2
bx a. Thus E
1
(T) has
dimension 1 and is spanned by x
2
1.
Therefore the basis = x, x
2
+ 1, x
2
1 will diagonalize T.
(c) Let = (1, 0, 0), (0, 1, 0), (0, 0, 1).
[T]

=
_
_
0 1 0
1 0 0
0 0 2
_
_
Date: December 1st.
1
2 DAVID ZYWINA
The eigenvalues of T are the roots of the polynomial
det([T]

I) = det
_
_
1 0
1 0
0 0 2
_
_
= ( 2)(
2
+ 1).
Therefore the only (real) eigenvalues are 2 with multiplicity 1. Thus we dont have enough
eigenvectors to construct a basis that diagonalizes T. Therefore T is not diagonalizable.
(d) Let = 1, x, x
2
.
[T]

=
_
_
1 0 0
1 1 1
1 1 1
_
_
A computation shows that det([T]

I) = t
3
+3t
2
2t = t(t
2
3t+2) = t(t1)(t2).
Thus the eigenvalues of T are 0, 1, 2.
In general, T(a +bx +cx
2
) = a + (a +b +c)x + (a +b +c)x
2
.
A vector a + bx + cx
2
is in E
0
(T) if a + (a + b + c)x + (a + b + c)x
2
= 0. Thus E
0
(T) is
spanned by x x
2
.
A vector a +bx +cx
2
is in E
1
(T) if a + (a +b +c)x + (a +b +c)x
2
= a +bx +cx
2
. Thus
E
1
(T) is spanned by 1 x x
2
.
A vector a + bx + cx
2
is in E
2
(T) if a + (a + b + c)x + (a + b + c)x
2
= 2a + 2bx + 2cx
2
.
Thus E
2
(T) is spanned by x +x
2
.
Therefore the basis = x x
2
, 1 x x
2
, x +x
2
diagonalizes T.
(e) Let = (1, 0), (0, 1), then
[T]

=
_
1 i
i 1
_
.
The eigenvalues of T are the roots of the polynomial
det([T] I) =
2
2 + 2,
which are 1 i. Now we to nd corresponding eigenvectors.
An eigenvector for 1 + i, would satisfy (z + iw, iz + w) = T(z, w) = (1 + i)(z, w) =
(z +iz, w +iw). Thus we must have w = z, which shows that (1, 1) is an eigenvector.
An eigenvector for 1 i, would satisfy (z + iw, iz + w) = T(z, w) = (1 i)(z, w) =
(z iz, w iw). Thus we must have w = z, which shows that (1, 1) is an eigenvector.
Therefore the basis = (1, 1), (1, 1) will diagonalize T.
(f) Let =
__
1 0
0 0
_
,
_
0 1
0 0
_
,
_
0 0
1 0
_
,
_
0 0
0 1
__
then
[T]

=
_
_
_
_
1 0 0 0
0 0 1 0
0 1 0 0
0 0 0 1
_
_
_
_
.
The eigenvalues of T are roots of the polynomial
det([T]

I) = ( 1)
3
( + 1),
MATH 110: HOMEWORK #1 3
that is 1, 1, 1, 1. The space E
1
(T) is easily seen to be spanned by the vectors
_

_
_
_
_
_
1
0
0
0
_
_
_
_
,
_
_
_
_
0
1
1
0
_
_
_
_
,
_
_
_
_
0
0
0
1
_
_
_
_
_

_
,
and the vector
_
_
_
_
0
1
1
0
_
_
_
_
spans E
1
(T). Therefore the basis
=
_

_
_
_
_
_
1
0
0
0
_
_
_
_
,
_
_
_
_
0
1
1
0
_
_
_
_
,
_
_
_
_
0
0
0
1
_
_
_
_
,
_
_
_
_
0
1
1
0
_
_
_
_
_

_
diagonalizes T.
Problem 11. Since similar matrices have the same eigenvalues, and the eigenvalues of an
upper triangular matrix are exactly those on the diagonal; we know from our assumption
that there is an invertible Q such that
A = Q
_
_
_
_

1
I
m
1

0
2
I
m
2

0 0
.
.
.
0 0 0
k
I
m
k
_
_
_
_
Q
1
.
(a)
tr(A) = tr
_
_
_
_

1
I
m
1

0
2
I
m
2

0 0
.
.
.
0 0 0
k
I
m
k
_
_
_
_
=
k

i=1
tr(
i
I
m
i
) =
k

i=1
m
i

i
(b)
det(A) = det
_
_
_
_

1
I
m
1

0
2
I
m
2

0 0
.
.
.
0 0 0
k
I
m
k
_
_
_
_
=
k

i=1
det(
i
I
m
i
) =
k

i=1

m
i
i
Problem 14.
(a) The equations x

= x +y and y

= 3x y can be re-written as:

= A
_
x
y
_
, where A =
_
1 1
3 1
_
.
4 DAVID ZYWINA
Diagonalizing the matrix A, we nd that A = Q
_
2 0
0 2
_
Q
1
, where Q =
_
1/3 1
1 1
_
.
Let
_
w
z
_
= Q
1
_
x
y
_
, then
_
w
z
_

= Q
1
_
x
y
_

= Q
1
A
_
x
y
_
= Q
1
AQ
_
w
z
_
=
_
2 0
0 2
__
w
z
_
=
_
2w
2z
_
.
Thus we get the much simpler equations: w

= 2w and z

= 2z. Therefore we
have w = c
1
e
2t
and z = c
2
e
2t
, where c
1
, c
2
are constants.
_
x
y
_
= Q
_
c
1
e
2t
c
2
e
2t
_
=
_
1
3
c
1
e
2t
+c
2
e
2t
c
1
e
2t
+c
2
e
2t
_
Thus the solution is x =
1
3
c
1
e
2t
+c
2
e
2t
and y = c
1
e
2t
+c
2
e
2t
.
(b) The equations x

1
= 8x
1
+ 10x
2
and x

2
= 5x
1
7x
2
can be re-written as:
_
x
1
x
2
_

= A
_
x
1
x
2
_
, where A =
_
8 10
5 7
_
.
Diagonalizing the matrix A, we nd that A = Q
_
2 0
0 3
_
Q
1
, where Q =
_
1 2
1 1
_
.
Let
_
y
1
y
2
_
= Q
1
_
x
1
x
2
_
, then
_
y
1
y
2
_

= Q
1
_
x
1
x
2
_

= Q
1
A
_
x
1
x
2
_
= Q
1
AQ
_
y
1
y
2
_
=
_
2 0
0 3
__
y
1
y
2
_
=
_
2y
1
3y
2
_
.
Thus we get the much simpler equations: y

1
= 2y
1
and y

2
= 3y
2
. Therefore we
have y
1
= c
1
e
2t
and y
2
= c
2
e
3t
, where c
1
, c
2
are constants.where c
1
, c
2
are constants.
_
x
1
x
2
_
= Q
_
c
1
e
2t
c
2
e
3t
_
= Q
_
c
1
e
2t
c
2
e
3t
_
=
_
c
1
e
2t
2c
2
e
3t
c
1
e
2t
+c
2
e
3t
_
Thus the solution is x
1
= c
1
e
2t
2c
2
e
3t
and x
2
= c
1
e
2t
+c
2
e
3t
.
(c)
_
_
x
1
x
2
x
3
_
_

= A
_
_
x
1
x
2
x
3
_
_
, where A =
_
_
1 0 1
0 1 1
0 0 2
_
_
.
Diagonalizing the matrix A, we nd that
A = Q
_
_
1 0 0
0 1 0
0 0 2
_
_
Q
1
, where Q =
_
_
1 0 1
0 1 1
0 0 1
_
_
.
MATH 110: HOMEWORK #1 5
As in the earlier parts the functions
_
_
y
1
y
2
y
3
_
_
:= Q
1
_
_
x
1
x
2
x
3
_
_
satisfy the easier
dierential equations: y

1
= y
1
, y

2
= y
2
and y

3
= 2y
3
. So there are constants c
1
, c
2
, c
3
such that y
1
= c
1
e
t
, y
2
= c
2
e
t
and y
3
= c
3
e
2t
.
_
_
x
1
x
2
x
3
_
_
= Q
_
_
y
1
y
2
y
3
_
_
= Q
_
_
c
1
e
t
c
2
e
t
c
3
e
2t
_
_
=
_
_
c
1
e
t
+c
3
e
2t
c
2
e
t
+c
3
e
2t
c
3
e
2t
_
_
.
Therefore the solutions are x
1
= c
1
e
t
+c
3
e
2t
, x
2
= c
2
e
t
+c
3
e
2t
and x
3
= c
3
e
2t
.
5.3
Problem 1.
(a) T; this is the corollary to Theorem 5.12
(b) T; this is a consequence of Theorem 5.13
(c) F; the coordinates also need to be non-negative
(d) F; it is true for columns, but not rows. See the examples in the book.
(e) T; this is a corollary to Theorem 5.15
(f) T; The Gerschgorin disks for the matrix are described as:
center 1 with radius [z[ +[ 1[ < 2
center 1 with radius [z[ +[1[ < 2
center z with radius [ 1[ +[1[ = 2.
It is easy to show that 3 is not in any of these disk, and hence Gerschgorins disk
Theorem implies that 3 is not an eigenvalue.
(g) T; this is Theorem 5.17
(h) F; a counter example is
_
0 1
1 0
_
(i) F; again a counter example is A =
_
0 1
1 0
_
. lim
m
A
m
does not converge since A
has an eigenvalue -1
(j) T; The convergence follows from Theorem 5.20(b), the rank one property follows
from Theorem 5.20(e)
Problem 2. e)
The eigenvalues of the matrix A =
_
2 1
4 3
_
are the roots of the polynomial det(A
I) =
2
2 = ( 2)( + 1).
If lim
m
A
m
exists, then Theorem 5.13 (the part of the theorem we need is proved in the
remarks afterwards in the text) implies that its eigenvalues of A are all in the set
C : [[ < 1 or = 1.
Since = 2 is not in this set, we nd that lim
m
A
m
does not exist.
6 DAVID ZYWINA
(f) The eigenvalues of the matrix A =
_
2 0.5
3 0.5
_
are the roots of the polynomial det(A
I) = (2 )(0.5 ) + 1.5 =
2
1.5 + 0.5 = ( 1)( 0.5).
A quick calculation shows that
_
1
2
_
is an eigenvector for = 1, and
_
1
3
_
is an
eigenvector for = 0.5.
Therefore we let Q =
_
1 1
2 3
_
and get
A = Q
_
1 0
0 0.5
_
Q
1
.
lim
m
A
m
= lim
m
Q
_
1
n
0
0 0.5
n
_
Q
1
= Q
_
lim
m
_
1
n
0
0 0.5
n
__
Q
1
= Q
_
1 0
0 0
_
Q
1
=
_
1 1
2 3
__
1 0
0 0
__
3 1
2 1
_
=
_
1 0
2 0
__
3 1
2 1
_
=
_
3 1
6 2
_
Problem 5. Let A =
_
0 1
0 0
_
and B =
_
0 0
1 0
_
. Since A
2
= B
2
= 0 we nd that
lim
m
A
m
= lim
m
B
m
= 0.
However, the matrix AB =
_
1 0
0 0
_
satises (AB)
m
= AB, for all m 1. Therefore
lim
m
(AB)
m
=
_
1 0
0 0
_
,=
_
0 0
0 0
_
= lim
m
A
m
lim
m
B
m
.
Problem 6. This will be a four-state Markov chain with the following states:
(1) recovered
(2) ambulatory
(3) bedridden
(4) dead.
The initial probability vector is
P =
_
_
_
_
0
0.3
0.7
0
_
_
_
_
.
MATH 110: HOMEWORK #1 7
The transition matrix is
A =
_
_
_
_
1 0.6 0.1 0
0 0.2 0.2 0
0 0.2 0.5 0
0 0 0.2 1
_
_
_
_
(we have made the reasonable assumptions that those people who recovered stay recovered,
and those people who died stay dead!).
Since
AP =
_
_
_
_
0.25
0.2
0.41
0.14
_
_
_
_
,
we see that after one month, 25% of the patients recover, 20% are ambulatory, 41% are
bedridden and 14% have died.
We now nd the eigenvalues of A.
det(A I) =
_
_
_
_
1 0.6 0.1 0
0 0.2 0.2 0
0 0.2 0.5 0
0 0 0.2 1
_
_
_
_
= (1 ) det
_
_
0.2 0.2 0
0.2 0.5 0
0 0.2 1
_
_
= (1 )
2
det
_
0.2 0.2
0.2 0.5
_
= ( 1)
2
(
2
0.7 + 0.06)
= ( 1)
2
( 0.1)( 0.6)
Therefore the eigenvalues of A are 1, 1, 0.1, 0.6.
The standard computations show that the eigenspace of = 1 has basis
_

_
_
_
_
_
1
0
0
0
_
_
_
_
_
_
_
_
0
0
0
1
_
_
_
_
_

_
;
that
_
_
_
_
11
18
9
2
_
_
_
_
is an eigenvector for = 0.1; and that
_
_
_
_
2
1
2
1
_
_
_
_
is an eigenvector of = 0.6.
Thus
A = Q
_
_
_
_
1 0 0 0
0 1 0 0
0 0 0.1 0
0 0 0 0.6
_
_
_
_
Q
1
,
8 DAVID ZYWINA
where Q =
_
_
_
_
1 0 11 2
0 0 18 1
0 0 9 2
0 1 2 1
_
_
_
_
.
We need to compute v := lim
m
A
m
P.
v = lim
m
A
m
P
= lim
m
Q
_
_
_
_
1
m
0 0 0
0 1
m
0 0
0 0 0.1
m
0
0 0 0 0.6
m
_
_
_
_
Q
1
P
= Q
_
_
_
_
1 0 0 0
0 1 0 0
0 0 0 0
0 0 0 0
_
_
_
_
Q
1
P
=
_
_
_
_
1 0 11 2
0 0 18 1
0 0 9 2
0 1 2 1
_
_
_
_
_
_
_
_
1 0 0 0
0 1 0 0
0 0 0 0
0 0 0 0
_
_
_
_
Q
1
P
=
_
_
_
_
1 0 0 0
0 0 0 0
0 0 0 0
0 1 0 0
_
_
_
_
Q
1
P
To compute z := Q
1
P, is the same as solving the equation
_
_
_
_
1 0 11 2
0 0 18 1
0 0 9 2
0 1 2 1
_
_
_
_
z =
_
_
_
_
0
0.3
0.7
0
_
_
_
_
.
This can see to have solution z =
_
_
_
_
59/90
31/90
1/450
17/50
_
_
_
_
. (Note that we didnt need to compute Q
1
in order to compute Q
1
P.)
Thus
v =
_
_
_
_
1 0 0 0
0 0 0 0
0 0 0 0
0 1 0 0
_
_
_
_
Q
1
P =
_
_
_
_
1 0 0 0
0 0 0 0
0 0 0 0
0 1 0 0
_
_
_
_
_
_
_
_
59/90
31/90
1/450
17/50
_
_
_
_
=
_
_
_
_
59/90
0
0
31/90
_
_
_
_
.
Therefore eventually all of the patients either recover or die. In particular, 59% of the
patients will recover and 31% will die.
MATH 110: HOMEWORK #1 9
Problem 10(a). The initial probability vector is P =
_
_
0.3
0.3
0.4
_
_
, and the transition matrix
is A =
_
_
0.6 0.1 0.1
0.1 0.9 0.2
0.3 0 0.7
_
_
.
AP =
_
_
0.25
0.38
0.37
_
_
A
2
P =
_
_
0.225
0.441
0.334
_
_
The eigenvectors w of A with eigenvalue 1 are solutions of the equation
_
_
0.4 0.1 0.1
0.1 0.1 0.2
0.3 0 0.3
_
_
w = (A I)w = 0.
Solving this we see that E
1
(A) = span
_
_
_
_
_
1
3
1
_
_
_
_
_
. The unique probability vector in E
1
(A)
is v =
_
_
1/5
3/5
1/5
_
_
. By Theorem 5.20 we have lim
m
(A
m
P) = v =
_
_
1/5
3/5
1/5
_
_
.
Problem 23. Let A =
_
0 1
0 0
_
and B =
_
0 0
1 0
_
. Since A
2
= B
2
= 0 we nd that
e
A
= I +A/1! + 0 + 0 + = 1 +A =
_
1 1
0 1
_
and
e
B
= I +B/1! + 0 + 0 + = 1 +A =
_
1 0
1 1
_
.
Thus we have
e
A
e
B
=
_
2 1
1 1
_
.
Now consider the matrix A + B =
_
0 1
1 0
_
. It has eigenvalues 1 and 1, so there is an
invertible matrix Q such that
A+B = Q
_
1 0
0 1
_
Q
1
.
10 DAVID ZYWINA
We now compute e
A+B
.
e
A+B
= I + (A+B) +
(A+B)
2
2!
+
(A+B)
3
3!
+
= I +Q
_
1 0
0 1
_
Q
1
+
_
Q
_
1 0
0 1
_
Q
1
_
2
2!
+
_
Q
_
1 0
0 1
_
Q
1
_
3
3!
+
= I +Q
_
1 0
0 1
_
Q
1
+
Q
_
1 0
0 1
_
2
Q
1
2!
+
Q
_
1 0
0 1
_
3
Q
1
3!
+
= Q
_
_
_
_
_
I +
_
1 0
0 1
_
+
_
1 0
0 1
_
2
2!
+
_
1 0
0 1
_
3
3!
+
_
_
_
_
_
Q
1
= Q
_
I +
_
1 0
0 1
_
+
_
1
2
2!
0
0
(1)
2
2!
_
+
_
1
3
3!
0
0
(1)
3
3!
_
+
_
Q
1
= Q
_

n=0
1
n
n!
0
0

n=0
(1)
n
n!
_
Q
1
= Q
_
e 0
0 e
1
_
Q
1
So to show e
A
e
B
,= e
A+B
, it suces to check that
_
2 1
1 1
_
,= Q
_
e 0
0 e
1
_
Q
1
.
However this is a consequence of the fact that these two matrices have dierent traces, and
hence are dierent.
tr(e
A
e
B
) = 2 + 1 = 3 and tr(e
A+B
) = e +e
1
3.086...
(Of course we could have also shown the two are dierent by explicitly computing Q and
then e
A+B
, but this would be signicantly more work.)
non-book questions
Problem (3).
(1) Take any w E
1
(P), so we have Pw = w. We nd that w E
1
(P
k
) since
P
k
w = P
k1
w = = P
2
w = Pw = w.
Thus E
1
(P) is a subspace of E
1
(P
k
). Since P is a probability matrix, we know that
E
1
(P) ,= 0. So E
1
(P) is a nonzero subspace of the 1-dimensional space E
1
(P
k
).
Therefore E
1
(P) = E
1
(P
k
), which is equivalent to what we needed to show.
(2) If B is a square matrix then rank B = rank B
t
(This is Corollary 2 on page 158, and
should be familiar from Math 54. It could be derived from the LU decomposition,
since transposing B essentially just transposes its LU decomposition). The dimension
MATH 110: HOMEWORK #1 11
theorem then implies that nullity B = nullity B
t
.
Now apply these general remarks to the matrix B := A I:
dimE

(A
t
) = dimN(A
t
I)
= nullity(A
t
I)
= nullity(AI)
t
= nullity(AI)
= dimN(A I)
= dimE

(A) = 1
(3) Following the hint, if A = (P
k
)
t
then A has all positive entries and unit row sums.
So Av = v has as one solution, v = u, where u is the vector of all ones. We want to
show there are no other independent solution. Take any 0 ,= v dimE
1
(A). Divide
v by v
k
(where v
k
is a component of v with maximal absolute value) to get w = v/v
k
.
Then w
k
= 1 and all other entries of w are have absolute value less than or equal to
1. Now Aw = w implies 1 = w
k
=

n
i=1
A
k,i
w
i
.
1 = [
n

i=1
A
k,i
w
i
[

i=1
[A
k,i
[[w
i
[ (triangle inequality)

i=1
A
k,i
(since A
k,j
> 0 and [w
i
[ 1
= 1
Thus all these inequalities are in fact equalities, since everything is sandwiched be-
tween 1 and 1. In particular each A
k,i
w
i
equals either A
k,i
or A
k,i
. Since w
k
= 1 we
must have A
k,i
w
i
= A
k,i
for each w
i
(otherwise we couldnt have all equalities above).
Since each A
k,i
is nonzero, each w
i
= 1 as desired.
(The hint about Gershgorin seems to be a red herring).
(4) Part 3 tells us that dimE
1
((P
k
)
t
) = 1. Part 2 then tells us that dimE
1
(P
k
) = 1.
Finally part 1 applies to give us that dimE
1
(P) = 1 as desired.
(5) Everything is laid out in the question until we get to (*):
(*) u
t
B
11
w
1
+u
t
B
12
w
2
= u
t
w
1
,
where
u is the vector of all 1s,
B
11
and B
12
have positive entries and column sums are < 1
w
1
has positive entries
w
2
has nonpositive entries
Thus u
t
B
11
w
1
= (u
t
B
11
)w
1
= x
t
w
1
, where x := u
t
B
11
and each 0 < x
i
< 1, so
u
t
B
11
w
1
<

i
(w
1
)
i
.
12 DAVID ZYWINA
Also u
t
B
12
w
2
0 by a similar argument. So
u
t
B
11
w
1
+u
t
B
12
w
2
<

i
(w
1
)
i
= u
t
w
1
contradicting (*). Thus there cannot be any nonpostive entries w
2
in w, and by
scaling we can make all the entries of w not just positive but sum to 1.
(6) The identity matrix I
n
is a probability matrix with dimE
1
(I
n
) = n.
Problem 4.
(1) If you draw a Gershgorin Circle for A = P
k
, it has a center at 0 < A
i,i
< 1 and radius
1 A
i,i
. So the Circle is tangent to the unit circle at 1 and otherwise lies entirely
inside the unit circle. So an eigenvalue must either lie inside the unit circle or equal
1.
(2) The question is how to guarantee that P does not have any eigenvalues on the unit
circle other than at 1.
If P had an eigenvalue on the unit circle somewhere else than 1, say at e
it
,= 1,
then P
m
would have an eigenvalue at e
imt
, for all m 1. We can pick arbitrarily
large m such that mt is not a multiple of 2. Thus we have innitely many m 1
such that P
m
has positive entries and has an eigenvalue of absolute value 1 which is
not 1. This however contradicts part 1.
Therefore, P has no eigenvalues on the unit circle besides 1.
5.4
Problem 18. (a) By denition f(t) = det(A tI), thus a
0
= f(0) = det(A 0) = det(A).
The matrix A is then invertible, if and only if, a
0
= det(A) is non-zero.
(b) The Cayley-Hamilton theorem says that f(A) = 0. More explicitly this can be written
as
(1)
n
A
n
+a
n1
A
n1
+ +a
1
A+a
0
= 0.
Multiplying both sides by A
1
we get:
(1)
n
A
n1
+a
n1
A
n2
+ +a
1
I +a
0
A
1
= 0.
Solving for A
1
we nd (recall that a
0
,= 0 by part (a))
A
1
=
1
a
0
_
(1)
n
A
n1
+a
n1
A
n2
+ +a
1
I
_
.
(c) Now consider the specic matrix A =
_
_
1 2 1
0 2 3
0 0 1
_
_
.
f(t) = det(AtI) = det
_
_
1 t 2 1
0 2 t 3
0 0 1 t
_
_
= (t1)(t2)(t+1) = t
3
+2t
2
+t2.
MATH 110: HOMEWORK #1 13
Part (b) then shows that
A
1
=
1
2
_
A
2
+ 2A+I
_
=
1
2
_
_

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

_
_
1 6 6
0 4 3
0 0 1
_
_
+
_
_
3 4 2
0 5 6
0 0 1
_
_
_
_
=
_
_
1 1 2
0 1/2 3/2
0 0 1
_
_
6.1
Problem 1.
(a) T; an inner product is a function from V V F, for a vector space over the eld
F
(b) T; In the comments at the beginning of this section, we restrict to the case F = R
or C
(c) F; If F = C, then an inner product is not linear in the second component since scalars
come out as their conjugate, i.e x, cy) = cx, y)
(d) F; there are lots of inner products on R
n
, for example if , ) is an inner product then
, ) is an inner product for all real > 0.
(e) F; the proof given makes no use of dimension.
(f) F; directly from denition on p.331
(g) F; Consider x = (1, 0), y = (0, 1), z = (0, 2) R
2
and , ) is the usual dot product.
Then x, y) = 0 = x, z), but y ,= z.
(h) T; If x, y) = 0 for all x, then in particular if x = y we have y, y) = 0. From the
fourth axiom of an inner product space we then nd that y = 0.
Problem 6. The remaining proofs from Theorem 6.1. We will use the four inner product
axioms without comment.
(b) x, cy) = cy, x) = cy, x) = cy, x) = cx, y)
(c) 0, x) = 0 0, x) = 00, x, =)0. x, 0) = 0, x) = 0 = 0.
(d) If x = 0, then x, x) = 0 follows from part (c).
Conversely suppose that x, x) = 0. If x ,= 0, then this would contradict the last inner
product axiom. Therefore x = 0.
(e) Suppose that x, y) = x, z) for all x V . Then for all x V (using part (a) and (b)),
x, y z) = x, y) x, z) = 0.
In particular, if x = y z then we have y z, y z) = 0. Part (d) then implies that
y z = 0, i.e. y = z.
14 DAVID ZYWINA
Problem 7. The remaining proofs from Theorem 6.2.
(a)
[[cx[[ =
_
cx, cx) =
_
cx, cx) =
_
ccx, x) =
_
[c[
2
x, x) =
_
[c[
2
_
x, x) = [c[ [[x[[
(b) That [[x[[ 0 for all x V is clear from the denition since x, x) 0.
We see that x, x) = 0, if and only if, [[x[[ :=
_
x, x) = 0. The result then follows directly
from Theorem 6.1(d).
Problem 11. Take any x, y V .
[[x +y[[
2
+[[x y[[
2
= x +y, x +y) +x y, x y)
= (x, x) +x, y) +y, x) +y, y)) + (x, x) x, y) y, x) +y, y))
= 2x, x) + 2y, y)
= 2 [[x[[
2
+[[y[[
2
In R
2
it says that for any parallelogram, the sum of the squares of the length of the four
sides is equal to the sum of the squares of the lengths of the two diagonals.
Problem 12. That the set v
1
, , v
k
is orthogonal means, v
i
, v
j
) = 0 if i ,= j.

i=1
a
i
v
i

2
=
k

i=1
a
i
v
i
,
k

j=1
a
j
v
j
)
=
k

i=1
a
i
v
i
,
k

j=1
a
j
v
j
)
=
k

i=1
a
i
k

j=1
a
j
v
i
, v
j
)
=
k

i=1
a
i
a
i
v
i
, v
i
)
=
k

i=1
[a
i
[
2
[[v
i
[[
2
Problem 13. We check the axioms of an inner product. We will use the corresponding
axioms for , )
1
and , )
2
without comment. Take any x, y, z V and c F.
(a)
x +z, y) = x +z, y)
1
+x +z, y)
2
= x, y)
1
+z, y)
1
+x, y)
2
+z, y)
2
= (x, y)
1
+x, y)
2
) + (z, y)
1
+z, y)
2
)
= x, y) +z, y)
MATH 110: HOMEWORK #1 15
(b)
cx, y) = cx, y)
1
+cx, y)
2
= cx, y)
1
+cx, y)
2
= c(x, y)
1
+x, y)
2
)
= cx, y)
(c)
x, y) = x, y)
1
+x, y)
2
= x, y)
1
+x, y)
2
= y, x)
1
+y, x)
2
= y, x)
(d) Suppose x ,= 0.
x, x) = x, x)
1
+x, x)
2
> x, x)
2
> 0.
Problem 15. (a) If x is a multiple of y, say x = cy, then
[x, y)[ = [cy, y)[ = [cy, y)[ = [c[ [[y[[
2
= [[cy[[ [[y[[ = [[x[[ [[y[[ .
The same idea works if y is a multiple of x.
Conversely suppose that [x, y)[ = [[x[[ [[y[[. We will show that one of the vectors x or y is
a multiple of the other. We may assume that y ,= 0 because the result is trivial in the other
case.
As the hint suggests dene
a =
x, y)
[[y[[
2
and z = x ay.
(This is not a completely random denition, it is the Gram-Schmidt process from the next
section)
First we note, using our hypothesis, that
[a[ =
[x, y)[
[[y[[
2
=
[[x[[ [[y[[
[[y[[
2
=
[[x[[
[[y[[
.
Now note that z and y are orthogonal:
z, y) = x ay, y) = x, y) ay, y) = x, y)
x, y)
[[y[[
2
y, y) = 0.
Then since z and ay are orthogonal, Exercise 10 show that [[ay +z[[
2
= [[ay[[
2
+[[z[[
2
. Thus
we have
[[x[[
2
= [[ay +z[[
2
= [[ay[[
2
+[[z[[
2
= [a[
2
[[y[[
2
+[[z[[
2
=
_
[[x[[
[[y[[
_
2
[[y[[
2
+[[z[[
2
= [[x[[
2
+[[z[[
2
,
16 DAVID ZYWINA
and cancelling [[x[[
2
from both sides we get [[z[[
2
= 0. This the implies that z = 0, and that
x = ay as desired.
(b) We are going to show that [[x +y[[ = [[x[[ + [[y[[, if and only if, the vectors x and y
point in the same direction or more precisely there exists a real number c 0 such that
x = cy or y = cx.
First suppose that x = cy for c 0 (the y = cx case is similar).
[[x +y[[ = [[cy +y[[ = [[(1 +c)y[[ = [1+c[ [[y[[ = [[y[[+c [[y[[ = [[y[[+[c[ [[y[[ = [[y[[+[[cy[[ = [[y[[+[[x[[
Conversely suppose that [[x +y[[ = [[x[[ + [[y[[. We may assume that y ,= 0. The proof
of the triangle inequality has the following string of inequalities, with the last line coming
from our assumption.
[[x +y[[
2
= [[x[[
2
+ 2'x, y) +[[y[[
2
[[x[[
2
+ 2[x, y)[ +[[y[[
2
[[x[[
2
+ 2 [[x[[ [[y[[ +[[y[[
2
= ([[x[[ +[[y[[)
2
= [[x +y[[
2
Since the rst and last terms are the same, all of the inequalities are in fact equalities. In
particular we nd that
[x, y)[ = [[x[[ [[y[[ and [x, y)[ = 'x, y).
From part (a) the equality [x, y)[ = [[x[[ [[y[[ implies that there is a scalar c such that
x = cy. It remains to show that c is real and c 0.
We have x, y) = cy, y) = c [[y[[
2
, thus c = x, y)/ [[y[[
2
. So it suces to show that x, y)
is real and x, y) 0.
Let x, y) = +i for , R. Then

2
+
2
= [x, y)[
2
= ('x, y))
2
=
2
.
Canceling
2
from both sides we see that
2
= 0, hence x, y) = R.
x, y) = 'x, y) = [x, y)[ 0
Problem 20). (a) x, y V a real inner product space.
1
4
[[x +y[[
2

1
4
[[x y[[
2
=
1
4
x +y, x +y)
1
4
x y, x y)
=
1
4
(x, x) +x, y) +y, x) +y, y))
1
4
(x, x) x, y) y, x) +y, y))
=
1
4
(x, x) + 2x, y) +y, y))
1
4
(x, x) 2x, y) +y, y))
= x, y)
MATH 110: HOMEWORK #1 17
(b) x, y V a complex inner product space.
4

k=1
i
k

x +i
k
y

2
=
4

k=1
i
k
x +i
k
y, x +i
k
y)
=
4

k=1
i
k
(x, x +i
k
y) +i
k
y, x +i
k
y))
=
4

k=1
i
k
(x, x) +i
k
x, y) +i
k
y, x) +i
k
i
k
y, y))
=
4

k=1
(i
k
x, x) +i
k
i
k
x, y) +i
2k
y, x) +i
2k
i
k
y, y))
=
4

k=1
(i
k
x, x) +x, y) + (1)
k
y, x) +i
k
y, y))
= (
4

k=1
i
k
)x, x) + (
4

k=1
1)x, y) + (
4

k=1
(1)
k
)y, x) + (
4

k=1
i
k
)y, y)
= 0 x, x) + 4x, y) + 0 y, x) + 0 y, y)
= 4x, y)
Problem 24. (a) V = M
mn
(F), [[A[[ = max
i,j
[A
i,j
[ for all A V .
Take A, B V and a F.
(1) [[A[[ 0 is clear.
[[A[[ = 0 max
i,j
[A
i,j
[ = 0 [A
i,j
[ = 0 for all i, j A
i,j
= 0 for all i, j A = 0
(2) [[aA[[ = max
i,j
[aA
i,j
[ = max
i,j
[a[[A
i,j
[ = [a[ max
i,j
[A
i,j
[ = [a[ [[A[[
(3)
[[A+B[[ = max
i,j
[A
i,j
+B
i,j
[
max
i,j
([A
i,j
[ +[B
i,j
[) max
i,j
[A
i,j
[ + max
i,j
[B
i,j
[ = [[A[[ +[[B[[
(d) V = R
2
, [[(a, b)[[ = max([a[, [b[).
Take (a, b), (x, y) V and c R.
(1) [[(a, b)[[ 0 is clear.
[[(a, b)[[ = 0 max([a[, [b[) = 0 [a[ = [b[ = 0 a = b = 0 (a, b) = 0
(2) [[c(a, b)[[ = max([ca[, [cb[) = max([c[[a[, [c[[b[) = [c[ max([a[, [b[) = [c[ [[(a, b)[[
(3)
[[(a, b) + (x, y)[[ = [[(a +x, b +y)[[ = max([a +x[, [b +y[)
max([a[ +[x[, [b[ +[y[) max([a[, [b[) + max([x[, [y[) = [[(a, b)[[ +[[(x, y)[[
18 DAVID ZYWINA
Problem 25. Suppose that the norm from Exercise 24(d) did come from an inner product
, ). Then using Exercise 20 we can recover the inner product. That is for x = (x
1
, x
2
), y =
(y
1
, y
2
) R
2
we would have
x, y) :=
1
4
[[x +y[[
2

1
4
[[x y[[
2
.
However, this inner product does not satisfy all of the required axioms. For example:
2(1, 0), (1, 1)) =
1
4
[[2(1, 0) + (1, 1)[[
2

1
4
[[2(1, 0) (1, 1)[[
2
=
1
4
[[(3, 1)[[
2

1
4
[[(1, 1)[[
2
=
1
4
3
2

1
4
1
2
=
8
4
= 2
2(1, 0), (1, 1)) =
1
2
[[(1, 0) + (1, 1)[[
2

1
2
[[(1, 0) (1, 1)[[
2
=
1
2
[[(2, 1)[[
2

1
2
[[(0, 1)[[
2
=
1
2
2
2

1
2
1
2
=
3
2
Thus 2(1, 0), (1, 1)) ,= 2(1, 0), (1, 1)), and in particular , ) is not an inner product.
non-book question
Problem (7). Prove that x, y) =

n
i=1
2x
i
y
i
+

n1
i=1
x
i
y
i+1
+

n1
i=1
y
i
x
i+1
is an inner
product on R
n
.
The rst three of the inner product axioms are easy. Take any x, y, z R
n
and c R.
x +z, y) =
n

i=1
2(x
i
+z
i
)y
i
+
n1

i=1
(x
i
+z
i
)y
i+1
+
n1

i=1
y
i
(x
i+1
+z
i+1
)
= (
n

i=1
2x
i
y
i
+
n1

i=1
x
i
y
i+1
+
n1

i=1
y
i
x
i+1
)
+(
n

i=1
2z
i
y
i
+
n1

i=1
z
i
y
i+1
+
n1

i=1
y
i
z
i+1
)
= x, y) +z, y)
MATH 110: HOMEWORK #1 19
cx, y) =
n

i=1
2(cx
i
)y
i
+
n1

i=1
(cx
i
)y
i+1
+
n1

i=1
y
i
(cx
i+1
)
= c(
n

i=1
2x
i
y
i
+
n1

i=1
x
i
y
i+1
+
n1

i=1
y
i
x
i+1
)
= cx, y)
y, x) =
n

i=1
2y
i
x
i
+
n1

i=1
y
i
x
i+1
+
n1

i=1
x
i
y
i+1
=
n

i=1
2x
i
y
i
+
n1

i=1
x
i
y
i+1
+
n1

i=1
y
i
x
i+1
= x, y)
It remains to verify the nal axiom. As the hint suggests, we express x, x) as a sum of
squares.
x, x) =
n

i=1
2x
2
i
+
n1

i=1
2x
i
x
i+1
=
n

i=1
2x
2
i
+
n1

i=1
_
(x
i
+x
i+1
)
2
x
2
i
x
2
i+1
_
= 2
n

i=1
x
2
i
+
n1

i=1
x
2
i

n1

i=1
x
2
i+1
+
n1

i=1
(x
i
+ x
i+1
)
2
= x
2
n
+x
2
1
+
n1

i=1
(x
i
+x
i+1
)
2
From this expression it is clear that x, x) 0 for all x R
n
.
Finally suppose that x, x) = 0.
Our explicit expression gives us that x
1
= x
n
= 0, and x
i
= x
i+1
for i = 1, . . . , n 1. It is
then clear that x
1
= x
2
= = x
n1
= x
n
= 0. That is, x = 0, and this concludes the proof
of the last axiom.