144 Chapter 3 Vector Spaces

(b) x, x − 1, x 2 + 1, x 2 − 1 for each of these four subspaces. (See Exercises 20

and 22 of Section 2.)
(c) x , x − x − 1, x + 1
2 2
(d) 2x, x − 2
17. Is it possible to find a pair of two-dimensional sub-
15. Let S be the subspace of P3 consisting of all poly- spaces U and V of R3 such that U ∩ V = {0}?
nomials p(x) such that p(0) = 0, and let T be the Prove your answer. Give a geometrical interpre-
subspace of all polynomials q(x) such that q(1) = tation of your conclusion. [Hint: Let {u1 , u2 } and
0. Find bases for {v1 , v2 } be bases for U and V , respectively. Show
(a) S (b) T (c) S ∩ T that u1 , u2 , v1 , v2 are linearly dependent.]
16. In R4 , let U be the subspace of all vectors of the 18. Show that if U and V are subspaces of Rn and
form (u 1 , u 2 , 0, 0)T , and let V be the subspace of U ∩ V = {0}, then
all vectors of the form (0, v2 , v3 , 0)T . What are the
dimensions of U , V , U ∩ V , U + V ? Find a basis dim (U + V ) = dim U + dim V

3.5 Change of Basis

Many applied problems can be simplified by changing from one coordinate system
to another. Changing coordinate systems in a vector space is essentially the same as
changing from one basis to another. For example, in describing the motion of a particle
in the plane at a particular time, it is often convenient to use a basis for R2 consisting of
a unit tangent vector T and a unit normal vector N instead of the standard basis {e1 , e2 }.
(See Application 2 in Section 3 of Chapter 2.)
In this section, we discuss the problem of switching from one coordinate system to
another. We will show that this can be accomplished by multiplying a given coordinate
vector x by a nonsingular matrix S. The product y = Sx will be the coordinate vector
for the new coordinate system.

Changing Coordinates in R2
The standard basis for R2 is {e1 , e2 }. Any vector x in R2 can be expressed as a linear
combination
x = x 1 e1 + x 2 e2
The scalars x1 and x2 can be thought of as the coordinates of x with respect to the
standard basis. Actually, for any basis {y, z} for R2 , it follows from Theorem 3.3.2 that
a given vector x can be represented uniquely as a linear combination
x = αy + βz
The scalars α and β are the coordinates of x with respect to the basis {y, z}. Let us order
the basis elements so that y is considered the first basis vector and z is considered the
second, and denote the ordered basis by [y, z]. We can then refer to the vector (α, β)T
as the coordinate vector of x with respect to [y, z]. Note that, if we reverse the order
of the basis vectors and take [z, y], then we must also reorder the coordinate vector.
The coordinate vector of x with respect to [z, y] will be (β, α)T . When we refer to a
basis using subscripts, such as {u1 , u2 }, the subscripts assign an ordering to the basis
vectors.

EXAMPLE 1 Let y = (2, 1)T and z = (1, 4)T . The vectors y and z are linearly independent and
hence they form a basis for R2 . The vector x = (7, 7)T can be written as a linear
 3.5 Change of Basis 145

combination
x = 3y + z
Thus, the coordinate vector of x with respect to [y, z] is (3, 1)T . Geometrically, the
coordinate vector specifies how to get from the origin to the point (7, 7) by moving
first in the direction of y and then in the direction of z. If, instead, we treat z as our first
basis vector and y as the second basis vector, then
x = z + 3y
The coordinate vector of x with respect to the ordered basis [z, y] is (1, 3)T . Geomet-
rically, this vector tells us how to get from the origin to (7, 7) by moving first in the
direction of z and then in the direction of y (see Figure 3.5.1).

7
3y

z
4 x

z 3y

y
2 4 6 7

Figure 3.5.1.

As an example of a problem for which it is helpful to change coordinates, consider

the following application:

APPLICATION 1 Population Migration

Suppose that the total population of a large metropolitan area remains relatively fixed;
however, each year 6% of the people living in the city move to the suburbs and 2% of
the people living in the suburbs move to the city. If, initially, 30% of the population
lives in the city and 70% lives in the suburbs, what will these percentages be in 10
years? 30 years? 50 years? What are the long-term implications?
The changes in population can be determined by matrix multiplications. If we set
⎧ ⎫ ⎧ ⎫
A=⎪ ⎩ 0.94 0.02 ⎪ ⎭ and x0 = ⎪ ⎩ 0.30 ⎪ ⎭
0.06 0.98 0.70
then the percentages of people living in the city and suburbs after 1 year can be calcu-
lated by setting x1 = Ax0 . The percentages after 2 years can be calculated by setting
x2 = Ax1 = A2 x0 . In general, the percentages after n years will be given by xn = An x.
If we calculate these percentages for n = 10, 30, and 50 years and round to the nearest
percent, we get
⎧ ⎫ ⎧ ⎫ ⎧ ⎫
⎪
x10 = ⎩
0.27 ⎪
⎭ x30 = ⎩ ⎪ 0.25 ⎪
⎭ x50 = ⎩ ⎪ 0.25 ⎪
⎭
0.73 0.75 0.75
146 Chapter 3 Vector Spaces

In fact, as n increases, the sequence of vectors xn = An x0 converges to a limit

x = (0.25, 0.75)T . The limit vector x is called a steady-state vector for the process.
To understand why the process approaches a steady state, it is helpful to switch
to a different coordinate system. For the new coordinate system, we will pick vectors
u1 and u2 , for which it is easy to see the effect of multiplication by the matrix A. In
particular, if we pick u1 to be any multiple of the steady-state vector x, then Au1 will
equal u1 . Let us choose u1 = (1 3)T and u2 = (−1 1)T . The second vector was
chosen because the effect of multiplying by A is just to scale the vector by a factor of
0.92. Thus, our new basis vectors satisfy
⎧ ⎫⎧ ⎫ ⎧ ⎫
Au1 = ⎪
⎩ 0.94 0.02 ⎪
⎭⎪⎩1⎪ ⎭=⎪⎩1⎪⎭ = u1
0.06 0.98 3 3
⎧ ⎫⎧ ⎫ ⎧ ⎫
Au2 = ⎪
⎩ 0.94 0.02 ⎪
⎭⎪⎩ −1 ⎪ ⎩ −0.92 ⎪
⎭=⎪ ⎭ = 0.92u2
0.06 0.98 1 0.92

The initial vector x0 can be written as a linear combination of the new basis vectors:
⎧ ⎫ ⎧ ⎫ ⎧ ⎫
x0 = ⎪
⎩ 0.30 ⎪
⎭ = 0.25 ⎪
⎩1⎪ ⎩ −1 ⎪
⎭ − 0.05 ⎪ ⎭ = 0.25u1 − 0.05u2
0.70 3 1

It follows that
xn = An x0 = 0.25u1 − 0.05(0.92)n u2

The entries of the second component approach 0 as n gets large. In fact, for n > 27,
the entries will be small enough so that the rounded values of xn are all equal to
⎧ ⎫
⎩ 0.25 ⎪
0.25u1 = ⎪ ⎭
0.75

This application is an example of a type of mathematical model called a Markov

process. The sequence of vectors x1 , x2 , . . . is called a Markov chain. The matrix A
has a special structure in that its entries are nonnegative and its columns all add up
to 1. Such matrices are called stochastic matrices. More precise definitions will be
given later when we study these types of applications in Chapter 6. What we want to
stress here is that the key to understanding such processes is to switch to a basis for
which the effect of the matrix is quite simple. In particular, if A is n × n, then we will
want to choose basis vectors so that the effect of the matrix A on each basis vector u j
is simply to scale it by some factor λ j ; that is,

Au j = λ j u j j = 1, 2, . . . , n (1)

In many applied problems involving an n × n matrix A, the key to solving the problem
often is to find basis vectors u1 , . . . , un and scalars λ1 , . . . , λn such that (1) is satisfied.
The new basis vectors can be thought of as a natural coordinate system to use with the
matrix A, and the scalars can be thought of as natural frequencies for the basis vectors.
We will study these types of applications in more detail in Chapter 6.
 3.5 Change of Basis 147

Changing Coordinates
Once we have decided to work with a new basis, we have the problem of finding the
coordinates with respect to that basis. Suppose, for example, that instead of using the
standard basis [e1 , e2 ] for R2 , we wish to use a different basis, say,
⎧ ⎫ ⎧ ⎫
⎩3⎪
u1 = ⎪ ⎭ , u2 = ⎪ ⎩1⎪ ⎭
2 1
Indeed, we may want to switch back and forth between the two coordinate systems.
Let us consider the following two problems:
I. Given a vector x = (x1 , x2 )T , find its coordinates with respect to u1 and u2 .
II. Given a vector c1 u1 + c2 u2 , find its coordinates with respect to e1 and e2 .
We will solve II first, since it turns out to be the easier problem. To switch bases from
{u1 , u2 } to {e1 , e2 }, we must express the old basis elements u1 and u2 in terms of the
new basis elements e1 and e2 . Thus, we have

u1 = 3e1 + 2e2
u2 = e1 + e2
It then follows that
c1 u1 + c2 u2 = (3c1 e1 + 2c1 e2 ) + (c2 e1 + c2 e2 )
= (3c1 + c2 )e1 + (2c1 + c2 )e2
Thus, the coordinate vector of c1 u1 + c2 u2 with respect to {e1 , e2 } is
⎧ ⎫ ⎧ ⎫⎧ ⎫
⎪ 3c + c ⎪ ⎪ 3 1 ⎪
⎭⎪⎩ c1 ⎪
x=⎩ 1 2 ⎭=⎩ ⎭
2c1 + c2 2 1 c2
If we set ⎧ ⎫
U = (u1 , u2 ) = ⎪
⎩3 1⎪ ⎭
2 1
then, given any coordinate vector c with respect to {u1 , u2 }, to find the corresponding
coordinate vector x with respect to {e1 , e2 }, we simply multiply U times c:
x = Uc (2)
The matrix U is called the transition matrix from the ordered basis {u1 , u2 } to the basis
{e1 , e2 }.
To solve problem I, we must find the transition matrix from {e1 , e2 } to {u1 , u2 }.
The matrix U in (2) is nonsingular, since its column vectors, u1 and u2 , are linearly
independent. It follows from (2) that
c = U −1 x
Thus, given a vector
x = (x1 , x2 )T = x1 e1 + x2 e2
we need only multiply by U −1 to find its coordinate vector with respect to {u1 , u2 }.
U −1 is the transition matrix from {e1 , e2 } to {u1 , u2 }.
148 Chapter 3 Vector Spaces

EXAMPLE 2 Let u1 = (3, 2)T , u2 = (1, 1)T , and x = (7, 4)T . Find the coordinates of x with respect
to u1 and u2 .

Solution
By the preceding discussion, the transition matrix from {e1 , e2 } to {u1 , u2 } is the inverse
of ⎧ ⎫
U = (u1 , u2 ) = ⎪⎩3 1⎪ ⎭
2 1
Thus, ⎧ ⎫⎧ ⎫ ⎧ ⎫
⎩ 1 −1 ⎪
c = U −1 x = ⎪ ⎭⎪⎩7⎪ ⎩ 3⎪
⎭=⎪ ⎭
−2 3 4 −2
is the desired coordinate vector and
x = 3u1 − 2u2

EXAMPLE 3 Let b1 = (1, −1)T and b2 = (−2, 3)T . Find the transition matrix from {e1 , e2 } to
{b1 , b2 } and the coordinates of x = (1, 2)T with respect to {b1 , b2 }.

Solution
The transition matrix from {b1 , b2 } to {e1 , e2 } is
⎧ ⎫
B = (b1 , b2 ) = ⎪ ⎩ 1 −2 ⎪ ⎭
−1 3
and hence the transition matrix from {e1 , e2 } to {b1 , b2 } is
⎧ ⎫
B −1 = ⎪ ⎩3 2⎪ ⎭
1 1
The coordinate vector of x with respect to {b1 , b2 } is
⎧ ⎫⎧ ⎫ ⎧ ⎫
c = B −1 x = ⎪⎩3 2⎪ ⎭⎪ ⎩1⎪ ⎭=⎪⎩7⎪⎭
1 1 2 3
and hence
x = 7b1 + 3b2
Now let us consider the general problem of changing from one basis [v1 , v2 ] of
R2 to another basis {u1 , u2 }. In this case, we assume that, for a given vector x, its
coordinates with respect to {v1 , v2 } are known:
x = c1 v1 + c2 v2
Now we wish to represent x as a sum d1 u1 + d2 u2 . Thus, we must find scalars d1 and
d2 so that
c1 v1 + c2 v2 = d1 u1 + d2 u2 (3)
If we set V = (v1 , v2 ) and U = (u1 , u2 ), then equation (3) can be written in matrix
form:
V c = Ud
 3.5 Change of Basis 149

It follows that
d = U −1 V c
Thus, given a vector x in R2 and its coordinate vector c with respect to the ordered
basis {v1 , v2 }, to find the coordinate vector of x with respect to the new basis {u1 , u2 },
we simply multiply c by the transition matrix S = U −1 V .

EXAMPLE 4 Find the transition matrix corresponding to the change of basis from {v1 , v2 } to
{u1 , u2 }, where
⎧ ⎫ ⎧ ⎫ ⎧ ⎫ ⎧ ⎫
⎩5⎪
v1 = ⎪ ⎭, ⎩7⎪
v2 = ⎪ ⎭ and ⎩3⎪
u1 = ⎪ ⎭, ⎩1⎪
u2 = ⎪ ⎭
2 3 2 1

Solution
The transition matrix from {v1 , v2 } to {u1 , u2 } is given by
⎧ ⎫⎧ ⎫ ⎧ ⎫
−1 ⎪ 1 −1 ⎪ ⎪ 5 7 ⎪ ⎪ 3 4 ⎪
S=U V =⎩ ⎭⎩ ⎭=⎩ ⎭
−2 3 2 3 −4 −5

The change of basis from {v1 , v2 } to {u1 , u2 } can also be viewed as a two-step
process. First we change from {v1 , v2 } to the standard basis, {e1 , e2 }, and then we
change from the standard basis to {u1 , u2 }. Given a vector x in R2 , if c is the coordinate
vector of x with respect to {v1 , v2 } and d is the coordinate vector of x with respect to
{u1 , u2 }, then
c1 v1 + c2 v2 = x1 e1 + x2 e2 = d1 u1 + d2 u2
Since V is the transition matrix from {v1 , v2 } to {e1 , e2 } and U −1 is the transition matrix
from {e1 , e2 } to {u1 , u2 }, it follows that

Vc = x and U −1 x = d

and hence
U −1 V c = U −1 x = d
As before, we see that the transition matrix from {v1 , v2 } to {u1 , u2 } is U −1 V (see
Figure 3.5.2).

V
[v1, v2] [e1, e2]

U –1
U –1V

[u1, u2]

Figure 3.5.2.