Linear Algebra Chapter 5.1

Applied Linear Algebra
OTTO BRETSCHER
http://www.prenhall.com/bretscher
Chapter 5
Orthogonality and Least Squares
Chia-Hui Chang
Email: chia@csie.ncu.edu.tw
National Central University, Taiwan
5.1 ORTHONORMAL BASES AND ORTHOGONAL PROJECTIONS

Not all bases are created equal.
Definition. 5.1.1
Qrthogonality, length, unit vectors
a. Tow vectors ~v and w
~ in Rn are called perpendicular or orthogonal if ~v w
~ = 0.
b. The length (or magnitude
or norm) of a
vector ~v in Rn is k~v k = ~v ~v .
c. A vector ~
u in Rn is called a unit vector if its
length is 1, (i.e., k~
uk = 1, or ~
u~
u = 1).
Explanation:
If ~v is a nonzero vector in Rn, then
~
u = k~1vk~v
is a unit vector.
1
Definition. 5.1.2 Orthonormal vectors

The vector v~1, v~2, . . . , v~m in Rn are called orthonormal if they are all unit vectors and orthogonal to one another:
(
v~i v~j =
1 if i=j,
0 if i6=j.
Example. 1.
The vectors e~1,e~2,. . .,e~n in Rn are orthonormal.
Example. 2.
"
For any scalar , the vectors
cos
sin
# "
sin
cos
are orthonormal.
Example. 3. The vectors
v~1 =
1/2
1/2
1/2
1/2
, v~2 =
1/2
1/2
1/2
1/2
, v~3 =
1/2
1/2
1/2
1/2
in R4 are orthonormal. Can you find a vector

v~4 in R4 such that all the vectors v~1,v~2,v~3,v~4
are orthonormal.
The following properties of orthonormal vectors are often useful:
Fact 5.1.3
a. Orthonormal vectors are linearly independent.
b. Orthonormal vectors v~1, . . ., v~n in Rn form
a basis of Rn.
Proof
a. Consider a relation
c1v~1+c2v~2+ +civ~i+ +cmv~m=~
0
Let us form the dot product of each side of
this equation with v~i:
(c1v~1 + c2v~2 + + civ~i + + cmv~m) v~i =
~
0 v~i = 0.
Because the dot product is distributive.
ci(~
vi v~i) = 0
Therefore, ci = 0 for all i = 1, . . ., m.
b. Any n linearly independent vectors in Rn
form a basis of Rn.
4
Definition. 5.1.4 Orthogonal complement

Consider a subspace V of Rn. The orthogonal
complement V of V is the set of those vectors
~
x in Rn that are orthogonal to all vectors in V :
V ={~
x in Rn: ~v ~
x = 0, for all ~v in V }.
Fact 5.1.5 If V is a subspace of Rn, then its
orthogonal complement V is a subspace of
Rn as well.
Proof
We will verify that V is closed under scalar
multiplication and leave the verification of the
two other properties as Exercise 23. Consider
a vector w
~ in V and a scalar k. We have
to show that kw
~ is orthogonal to all vectors
~v in V. Pick an arbitrary vector ~v in V. Then,
(kw)~
~ v =k(w
~ ~v )=0, as claimed.
5
Orthogonal projections
See Figure 5.
The orthogonal projection of a vector ~
x onto
one-dimentaional subspace V with basis ~v1 (unit
vector) is computed by:
projV ~
x=w
~ = (v~1 ~
x)v~1
Now consider a subspace V with arbitrary dimension m. Suppose we have an orthonormal
basis ~v1, ~v2, . . . , ~vm of V . Find w
~ in V such that
~
xw
~ is in V . Let
w
~ = c1~v1 + c2~v2 + + cm~vm
It is required that
~ w
x
~ =~
x c1~v1 c2~v2 cm~vm
is perpendicular to V ; i.e.:
6
~ w)
~vi (x
~ = ~vi (~
x c1~v1 c2~v2 cm~vm)
= ~vi ~
x c1(~vi ~v1) ci(~vi ~vi) cm(~vi ~vm)
= ~vi ~
x ci = 0
The equation holds if ci = ~vi ~
x.
Therefore, there is a unique w
~ in V such that
~
xw
~ is in V , namely,
w
~ = (~v1 ~
x)~v1 + (~v2 ~
x)~v2 + + (~vm ~
x)~vm
Fact 5.1.6 Orthogonal projection
Consider a subspace V of Rn with orthonormal
basis v~1,v~2,. . .,v~m. For any vector ~
x in Rn, there
is a unique vector w
~ in V such that ~
x-w
~ is in
V . This vector w
~ is called the orthogonal
projection of ~
x onto V , denoted by projV ~
x. We
have the formula
projV ~
x=(v~1 ~
x)v~1+ +(v~m ~
x)v~m.
The transformation T (~
x) = projV ~
x from Rn to
Rn is linear.
7
Example. 4
Consider the subspace V =im(A) of R4. where
A=
1
1
1 1
.
1 1
1
1
Find projV ~
x, for
~
x=
1
3
1
7
Solution
The two columns of A form a basis of V . Since
they happen to be orthogonal, we can construct an orthonormal basis of V merely by dividing these two vectors by their length (2 for
both vectors):
8
v~1 =
1/2
1/2
1/2
1/2
, v~2 =
1/2
1/2
1/2
1/2
Then,
projV ~
x=(v~1 ~
x)v~1+(v~2 ~
x)v~2=6v~1+ 2v~2=
1
3
1
3
=
2
To check this answer, verify that ~

x-projV ~
x is
perpendicular to both v~1 and v~2.
What happens when we apply Fact 5.1.6 to

the subspace V =Rn of Rn with orthonormal
basis v~1, v~2, ,v~n? Clearly, projV ~
x=~
x, for all ~
x
in Rn. Therefore,
~
x = (v~1 ~
x)v~1 + + (v~n ~
x)v~n,
for all ~
x in Rn. See Figure 7.
Fact 5.1.7
Consider an orthonormal basis v~1, ,v~n of Rn.
Then,
~
x = (v~1 ~
x)v~1 + + (v~n ~
x)v~n,
for all ~
x in Rn.
This is useful for compute the B-coordinate,
since ci = ~vi ~
x.
9
Example. 5
By using
paper and pencil, express the vector

1
~
x= 2 as a linear combination of
3
1 2 , v~ = 1
v~1 = 1
,
v
~
=
2
3
2
3
3
3
1 .
2
Solution
Since v~1,v~2,v~3 is an orthonormal basis of R3,
we have
~
x = (v~1 ~
x)v~1 + (v~2 ~
x)v~2 + (v~3 ~
x)v~3 = 3v~1 +
v~2 + 2v~3.
10
From Pythagoras to Cauchy

Example. 6
Consider a line L in R3 and a vector ~
x in R3.
What can you say about the relationship between the lengths of the vectors ~
x and projL~
x?
Solution
Applying the Pythagorean theorem to the shaded
right triangle in Figure 8, we find that
k projL~
x kk ~
xk.
The statement is an equality if (and only if) ~
x
is on L.
Does this inequality hold in higher dimensional
cases? We have to examine whether the
Pythagorean theorem holds in Rn.
11
Fact 5.1.8 Pythagorean theorem

Consider two vectors ~
x and ~
y in Rn. The equation
k~
x+~
y k2=k ~
x k2 + k ~
y k2
holds if (and only if)~
x and ~
y are orthogonal.
(See Figure 9.)
Proof The verification is straightforward:
k~
x+~
y k2= (~
x+~
y ) (~
x+~
y)
=~
x~
x + 2(~
x~
y) + ~
y~
y
=k ~
x k2 +2(~
x~
y )+ k ~
y k2
=k ~
x k2 + k ~
y k2
if (and only if) ~
x~
y = 0.
12
Fact 5.1.9 Consider a subspace V of Rn and

a vector ~
x in Rn. Then,
k projV ~
x kk ~
x k.
The statement is an equality if(and only if)~
x
is in V .
Proof we can write ~
x = projV ~
x+(~
xprojV ~
x)and
apply the Pythagorean theorem(see Figure 10):
k~
x k2=k projV ~
x k2 + k ~
x projV ~
x k2.
It follows that k projV ~
x kk ~
x k, as claimed.
13
Let V be a one-dimensional subspace of Rn

spanned by a(nonzero) vector ~
y . We introduce
the unit vector
~
u = k~1yk ~
y
in V . (See Figure 11.)
We know that
projV ~
x = (~
u~
x)~
u = k~y1k2 (~
y~
x)~
y.
for any ~
x in Rn. Fact 5.1.9 tells us that
k~
x kk projV ~
x k=k k~y1k2 (~
y~
x)~
y k=
1 |~
y~
x| k ~
y k.
k~
y k2
To justify the last step, note that k k~v k= |k| k
~v k, for all vectors ~v in Rn and all scalars k.
(See Exercise 25(a).) We conclude that
|~
x~
y|
k~
yk
k ~
x k.
14
Fact 5.1.10 Cauchy-Schwarz inequality

If ~
x and ~
y are vectors in Rn, then
|~
x~
y | k ~
x kk ~
y k.
The statement is an equality if (and only if) ~
x
and ~
y are parallel.
Definition. 5.1.11
Angle between two vectors Consider two
nonzero vectors ~
x and ~
y in Rn. The angle
between these vectors is defined as
~
y
cos = k~x~xkk~
.
yk
Note that is between 0 and , by definition

of the inverse cosine functiion.
15
Example. 7
Find the angle between the vectors
1
0
y=
~
x = and ~
1
0
0
Solution
~
y
1 = 1
=
cos = k~x~xkk~
12
2
yk
16
Correlation
Consider two characteristics of a population,
with deviation vectors ~
x and ~
y . There is a
positive correlation between the two characteristics if (and only if) ~
x~
y > 0.
Definition. 5.1.12
Correlation coefficient
The correlation coefficient r between two characteristics of a population is the cosine of the
angle between the deviation vectors ~
x and ~
y
for the two characteristics:
~
y
r = cos() = k~x~xkk~
yk
Exercise 5.1: 7, 9, 12, 19, 23, 24, 25, 28

17

Linear Algebra Chapter 5.1

Uploaded by

Copyright:

Available Formats

Linear Algebra Chapter 5.1

Uploaded by

Document Information

Original Description:

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Linear Algebra Chapter 5.1

Uploaded by

Copyright:

Available Formats

Applied Linear Algebra

5.1 ORTHONORMAL BASES AND ORTHOGONAL PROJECTIONS

Definition. 5.1.2 Orthonormal vectors

For any scalar , the vectors

Example. 3. The vectors

in R4 are orthonormal. Can you find a vector

Definition. 5.1.4 Orthogonal complement

To check this answer, verify that ~

What happens when we apply Fact 5.1.6 to

paper and pencil, express the vector

From Pythagoras to Cauchy

Fact 5.1.8 Pythagorean theorem

Fact 5.1.9 Consider a subspace V of Rn and

Let V be a one-dimensional subspace of Rn

Fact 5.1.10 Cauchy-Schwarz inequality

Note that is between 0 and , by definition

Exercise 5.1: 7, 9, 12, 19, 23, 24, 25, 28

You might also like