Assignment 4/MATH 247/Winter 2010

Due: Tuesday, February 9

First, you find, in Part 1, a summary of some of the material discussed in class, and also
found in Chapter 6 of the text.

The problems to solve are in Part 2, starting on page 5.

Part 1: Matrix representation and change of basis: the special case for operators.

A linear operator is a linear mapping whose domain and codomain are the same space:

T :V → V .

(Linear operators are the most important, but of course, not the only type, of linear
mapping, which has the general form T : U → V , with possibly different vector spaces
U and V .)

1. Coordinate vectors

Let V be a vector space, U = (u1 ,..., un ) a basis of V ; dim(V ) = n . For an arbitrary

vector w in V , the coordinate vector of w relative to U , in notation [ w]U , is the
column vector

⎡ x1 ⎤
[ w]U = ⎢⎢ # ⎥⎥
⎢⎣ xn ⎥⎦

n
for which w = x1 ⋅ u1 + ... + xn ⋅ un = ∑ x ⋅u
i =1
i i .

Fact 1 (“Coordinates are linear functions”):

k k
If w = ∑ a j ⋅ w j = a1 ⋅ w1 + ... + ak ⋅ wk , then [w]U =
j =1
∑a
j =1
j ⋅ [ w j ]U .

1
 ⎡ x1 j ⎤
⎢ ⎥ n k
Proof: Write [ w j ]U = ⎢ # ⎥ . Then w j = ∑x ij ⋅ ui , and for w = ∑ a j ⋅ w j , we have
⎢ xnj ⎥ i =1 j =1
⎣ ⎦
⎡ x1 ⎤
= ⎢⎢ # ⎥⎥ , this means xi =
k n n k k
w= ∑a ⋅∑ x
j =1
j
i =1
ij ⋅ ui = ∑ (∑ a
i =1 j =1
j ⋅ xij ) ⋅ ui . For [ w]U ∑a
j =1
j ⋅ xij .
⎢⎣ xn ⎥⎦
⎡ x1 ⎤ ⎡ x1 j ⎤
⎢ ⎥
And this is the same as ⎢⎢ # ⎥⎥ =
k k

∑ a j ⋅ ⎢ # ⎥ , that is, [ w]U = ∑a j ⋅ [ w j ]U .

⎢⎣ xn ⎥⎦
j =1
⎢ xnj ⎥ j =1
⎣ ⎦

2. Matrix representation

Let T : V → V be a linear operator on V , U = (u1 ,..., un ) a basis of V ; dim(V ) = n .

The matrix of T relative to the basis U , in notation [T ]U , is the n × n matrix defined
as

[T ]U = [ [T (u1 )]U ,..., [T (un )]U ]

We have the relation

Fact 2: ` [T ( w)]U = A ⋅ [ w]U , (1)

where A = [T ]U , w any vector in V .

⎡ x1 ⎤
Proof: Let X = ⎢⎢ # ⎥⎥ = [ w]U . This means w =
n

∑ x ⋅ u . Therefore, since
i =1
i i T is linear,
⎢⎣ xn ⎥⎦
n
T ( w) = ∑ x ⋅ T (u ) .
i =1
i i Using Fact 1 (“coordinates are linear”), we infer

⎡ x1 ⎤
= ∑ [T (ui )]U ⋅ xi = [ [T (u1 )]U ,..., [T (un )]U ] ⋅ ⎢⎢ # ⎥⎥ =
n n
[T ( w)]U = ∑ x ⋅ [T (u )]
i =1
i i U
i =1
⎢⎣ xn ⎥⎦
A ⋅ X = A ⋅ [ w]U .

2
Fact 2’: The equality (1) determines the matrix A as A = [T ]U .

Proof: Apply (1) to the vector w = ui . We have [ui ]U = ei , the standard n × 1 column
vector with all but the i th component equal zero, and with the i th component equal 1.
For any n × n matrix A , A ⋅ ei = Ai , the i th column of A . Therefore, (1) now says that
[T (ui )]U = Ai . This is precisely to say that A = [T ]U .

3. Change of basis

Suppose U = (u1 ,..., un ) and V = (v1 ,..., vn ) are both bases of the space V ; dim(V ) = n .
The change-of –basis matrix from U to V is the matrix P , denoted sometimes by
[U →V ] , and defined as

P = [U →V ] = [ [v1 ]U ,..., [vn ]U ] .

Fact 3: For any vector w in V , we have

[ w]U = P ⋅ [ w]V

(and not the other way around!);

and, equivalently,

[ w]V = P −1 ⋅ [ w]U . (2)

⎡ x1 ⎤ ⎡ y1 ⎤
Proof: Let us write X = ⎢ # ⎥ for X = [ w]U , and Y = ⎢⎢ # ⎥⎥ = [ w]V . We want
⎢ ⎥
⎢⎣ xn ⎥⎦ ⎢⎣ yn ⎥⎦
X = P ⋅Y .

3
With P = ( pij )n×n , we have that the column vector [v j ]U is the same as
⎡ p1 j ⎤
⎢ ⎥
[v j ]U = ⎢ # ⎥. We have
⎢ pnj ⎥
⎣ ⎦

1 n 2 n n 3 n 4 n n
w = ∑
j=
yj ⋅vj
1
= ∑
j
y j ⋅ (∑ pij ⋅ ui )
=1 i=1
= ∑ y j ⋅pij ⋅ ui
i, j= 1
= ∑
i
(∑ y j ⋅ pij ) ⋅ ui
=1j =1

1 2 3
[explanations: = by the definition [ w]V ; = by the definitions of [v j ]U and P ; =
expresses the fact that the sum on the left can be written, in any order, as the sum of the
n 2 terms y j ⋅ pij ⋅ ui , one for each pair (i, j ) of indices i, j from 1 to n . On the right
4
of = , the sum is regrouped by collecting the terms that contain the same ui .]

and

n
w = ∑
i
xi ⋅ ui .
=1

n n
⎡ x1 ⎤ ⎡ y1 ⎤
Therefore: xi = ∑ y j ⋅ pij = ∑ pij ⋅ y j , which is the same as ⎢⎢ # ⎥⎥ = P ⋅ ⎢⎢ # ⎥⎥ , that
j =1 j =1
⎢⎣ xn ⎥⎦ ⎢⎣ yn ⎥⎦
is, X = P ⋅ Y .

Fact 3’: Equality (2) determines the matrix P as [U →V ] .

Proof: Apply (2) to the vector w = ui .

4. Change of basis and matrix representation

Suppose U = (u1 ,..., un ) and V = (v1 ,..., vn ) are both bases of the space V ; dim(V ) = n .
Suppose T : V → V be a linear operator on V . We have the three matrices

A = [T ]U , B = [T ]V , P = [U →V ] .

We have the following connection between them

4
Fact 4: A = P ⋅ B ⋅ P −1

or equivalently,

B = P −1 ⋅ A ⋅ P

Proof : By Fact 2’, it suffices to show that ( P ⋅ B ⋅ P −1 ) ⋅ [ w]U = A ⋅ [ w]U = [T ( w)]U . But

P −1 ⋅ [ w]U = [ w]V (see Fact 3)

B ⋅ [ w]V = [T ( w)]V (meaning of B = [T ]V )
and
P ⋅ [T ( w)]V = [T ( w)]U .

Therefore,

P ⋅ B ⋅ P −1 ⋅ [ w]U = P ⋅ B ⋅ [ w]V = P ⋅ [T ( w)]V = [T ( w)]U

as desired.

Part 2: The questions

[1] This problem is an exercise in comprehension of Part 1. Answer the questions

economically, for each question using previously obtained items when possible. Pretend
that the questions concerned vectors and matrices of large dimensions for which economy
of calculation would be of importance.

Let : V = \ 2 , E = (e1 , e2 ) the standard basis of \ 2 ;

⎛ 3⎞ ⎛ −2 ⎞
U = (u1 , u2 ) a basis of \ 2 , where u1 = ⎜ ⎟ , u2 = ⎜ ⎟ ;
⎝ −4 ⎠ ⎝ 3⎠

5
 ⎛1⎞ ⎛8⎞
V = (v1 , v2 ) another basis of \ 2 , where v1 = ⎜ ⎟ , v2 = ⎜ ⎟ ,
⎝ −2 ⎠ ⎝ −5 ⎠
⎛7⎞ ⎛ 6 ⎞
w1 = ⎜ ⎟ , w2 = ⎜ ⎟ , w = 5w1 + 6 w2 .
⎝11⎠ ⎝ −10 ⎠

1) Determine the following entities:

P = [E →U ], Q = [U → E ], X 1 = [ w1 ]E , X 2 = [ w2 ]E , Y1 = [ w1 ]U , Y2 = [ w2 ]U ,

X = [ w]E , Y = [ w]U .

Write down the equalities that are true for these entities on general grounds.

2) Let R = [U →V ], S = [V →U ] . Write down and prove equalities that

express R and S in terms of P = [E →U ] and M = [E →V ] ; the equalities should
hold on general grounds (hints: use Fact 3’, and argue similarly to the proof of Fact 4)).

Determine M , R and S .

3) Let the linear operator T : V → V be defined by:

T (u1 ) = −2u1 + 4u2 , T (u2 ) = −3u1 − 9u2 .

Use ,when reasonable, previously obtained matrices, and determine the matrices
[T ]U , [T ]E , [T ]V .

4) For w , w1 and w2 given above, determine the vectors T ( w1 ), T ( w2 ), T ( w) .

5) Suppose v = a ⋅ v1 − (a + 1) ⋅ v2 ; v1 , v2 are as given before. Determine T (v) .

[2] We define the following linear operators:

T1 : \ 2 → \ 2 is the reflection in the line l whose equation is 2 x − 5 y = 0 .

T2 : \ 2 → \ 2 is the orthogonal projection onto the line l ( l as before);

that is, T2 ( X ) = projl ( X ) .

6
T3 : \ 4 → \ 4 is the reflection of \ 4 (!) in the plane U = span (u1 , u2 ) , where
⎛1⎞ ⎛ −2 ⎞
⎜ ⎟ ⎜ ⎟
2 3
u1 = ⎜ ⎟ , u2 = ⎜ ⎟ . This means that T ( X ) = projU ( X ) − projU ⊥ ( X ) ; remember that
⎜ −1⎟ ⎜1⎟
⎜ ⎟ ⎜ ⎟
⎝0⎠ ⎝ −1 ⎠
always X = projU ( X ) + projU ⊥ ( X ) .

T4 : \ 4 → \ 4 is the orthogonal projection onto the plane U ( U is as before).

For each of the four cases T : V → V ,

a) determine a basis U of V for which [T ]U is easily obtained;

b) determine [T ]E , with E the standard basis (of V = \ n for n = 2 or 4);
⎛ −5 ⎞
c) calculate the values T ( w) where w = ⎜ ⎟ for the first two operators, and
⎝ −3 ⎠
⎛1⎞
⎜ ⎟
−1
w = ⎜ ⎟ for the last two.
⎜ 0⎟
⎜ ⎟
⎝ −2 ⎠

⎛1⎞ ⎛5⎞
[3] Let u1 = ⎜ 2 ⎟ , u2 = ⎜⎜ 1 ⎟⎟ , and let U = span (u1 , u2 ) , a subspace (plane) of \ 3 .
⎜ ⎟
⎜ −2 ⎟ ⎜ −1 ⎟
⎝ ⎠ ⎝ ⎠
There are two rotations of the plane U in itself, through 30D about the origin. (It is no
use now to say which one is clockwise, which one is counterclockwise, since such a
characterization depends on from which side of the plane we are viewing the plane itself.)
For each of these rotations, say T , and for a vector X in U , T ( X ) is a vector in U
such that T ( X ) = X and )( X , T ( X )) = 30D .

Determine the matrix [T ]U for both rotations as T as follows:

1) Determine an orthogonal basis V = (v1 , v2 ) of U by the following

< u ,v >
formulas: v1 = u1 , v2 = u2 − 2 1 ⋅ v1 (this is called the Gram-Schmidt process,
< v1 , v1 >
applied to two vectors; soon we will learn more about it).

7
 1
2) Normalize the basis V : form Vˆ = (vˆ1 , vˆ2 ), where vˆi = ⋅ vi .
vi
Vˆ = (vˆ1 , vˆ2 ) is an orthonormal basis of U : v̂1 ⊥ v̂2 , and v1 = v2 = 1 . (Do not use
decimal approximations. Leave quantities such as 2 as they are, without calculating
them.)

3) Write down the matrix [T ] ˆ : this is just an ordinary rotation matrix,

V
[T ] ˆ =[T ']E , for either of the two rotations T ' : \ 2 → \ 2 of \ 2 through 30 D .
V

4) Apply change-of-basis to calculate [T ]U .

As an application,

5) determine the value of T (2u1 + 3u2 ) numerically as a 3 ×1 vector in

\ 3 , for both rotations as T . (Leave expressions “algebraically’; do not calculate square
roots.)

Rotations of \ 3 (preliminaries to problem [4])

Let l be a line passing through the origin, and let α be an angle. Let U be the plane
through the origin that is perpendicular to l : in our general terminology, U = l ⊥ , the
orthogonal complement of l .
A rotation in \ 3 about the axis l through the angle α is a linear operator
T : \3 → \3 for which
a) T (v) = v for the direction vector v of the line l ; and
b) whenever u ∈ U , we also have T (u ) ∈ U (for which we say that U is
invariant under T ) , and the operator T b U : U → U on U , defined by
(T b U )(u ) = T (u ) ( u ∈ U ) , is a rotation of the plane U through angle α . In short, T
acts in the plane U as a rotation of a plane

Since, unless α = 0 or α = 180D , there are two possibilities for the rotation
T b U : U → U as in [3], there are two rotations T about the axis l through angle α .

8
[4] This is a continuation of problem [3]. We let U be the plane defined in [3]. We let
α = 30D as in [3]. We let l be the line perpendicular to U through the origin: l = U ⊥
(and thus U = l ⊥ ).

The task is to calculate [T ]E with the standard basis E of \ 3 , for the two rotations
T : \3 → \3 about the axis l through the angle α = 30D .

To do this, use the work in [3]. Recipe: Calculate a unit vector v̂3 that lies in the line l .
(there are two possibilities; it is enough to take one, either one) . The three vectors
vˆ1 , vˆ2 , vˆ3 form an orthonormal basis V = (vˆ1 , vˆ2 , vˆ3 ) of \ 3 . Write down the matrix [T ] 
V
directly, using the work in [3]. Apply change-of basis to obtain [T ]E . As an application
⎛1⎞
(“tasting”), calculate the two rotated vectors T (⎜⎜1⎟⎟) , for both rotations T : \3 → \3 .
⎜1⎟
⎝ ⎠
(Leave expressions “algebraically’; do not calculate square roots.)

Remark: When Vˆ = (vˆ1 ,..., vˆn ) is an orthonormal system of n × 1 column vectors, then
the n × n matrix P = [vˆ1 ,..., vˆn ] is called an orthogonal matrix. An orthogonal matrix
P has the property that P tr ⋅ P = I n (as it is easily seen), and thus P −1 = P tr : the inverse
is easy to calculate!