MATH 261 under revision updated

You might also like

Download as pdf or txt
Download as pdf or txt
You are on page 1of 43

Course Author

Kwasi Baah Gyamfi received his B.Sc. (Mathematics) degree from K.N.U.S.T. He furthered
his studies at the Norwegean University of Science and Technology, Trondheim-Norway,
where he obtained his M.Sc degree in Mathematics. He joined the Faculty of Physical
Sciences of the K.N.U.S.T in 2001. K. Baah Gyamfi obtained his Phd at the KNUST in 2014.
Currently, he is a Senoir Lecturerr teaching Mathematics at the KNUST in the Department of
Mathematics.

i
UNIT 1

LINEAR TRANSFORMATION AND BILINEAR


FORMS
SESSION 1.2 Linear Transformation

1-2.1 Linear Mapping

Definition
Let X and Y be any vector spaces. To each vector x  X we assign a unique vector y  Y.
Then we say that a mapping (or transformation or operator) of X into Y is given. Such a
mapping is denoted by a capital letter, say F. Example: F: X  Y. The vector y Y assigned
to a vector x X is called the image of x and is denoted by F(x) = y.

F is called a linear mapping or linear transformation if for all vectors v and x in X and scalars
c K where K is a field if it satisfies the following conditions;
(a) F (v  x)  F (v)  F ( x)
(b) F (cx )  cF ( x) .
Thus F : X  Y is linear if it preserves vector addition and scalar multiplication.

Let X be a vector space over the field K. A mapping F : X  K is termed a linear functional
(or linear forms) if for every u , v X and every scalars q, r  K ; F (qu  rv)  qF (u )  rF (v)
That is a linear functional on X is a linear mapping from X to K.
Example
Let L be the operator defined by L( x )  3 x , for each x  R 2 then we have
L(x)  (3x)
  (3 x)
= L (x ) , scalar multiplication
L( x  y )  3( x  y )
 3x  3 y
 L( x)  L( y )
Hence L is a linear mapping.

1
1-2.2 Linear functional
Let V be a vector space over a field F. The dual space of V denoted by V  is the set of all
V   (V , F )  {F : V  F | F is linear}
linear maps from V to F that is eg. F:V  K s.t for all u, v V and q, r  K .
F (uq  vr )  qF (u )  rF (v )

and this is known as the linear functional. Thus linear functional is a vector space to its scalar
field.
The space of all linear functional is known as the Dual space of a vector V denoted by V 

Example 1
The mapping L: R2 R1 defined by L(x) = X1 + Xz is a linear transformation since L(x +
y) = = =

Example 2
Let L be the mapping from c[a,b] to If f and g are any
vectors in c[a,b] then is a linear functional on X.

=   f  x  dx    g  x  dx
b b

a a

=
Therefore, L is a linear transformation.

Example
Let V  K n , the vector space of n-tupples which we write as column vectors. Then the dual
space V* can be identified with the space of row vectors. Any linear functional
  (a1 , .......... ., an ) in V* has the representation
 x1 
 
 ( x1 , x2 , .......... ......., xn )  (a1 , a2 , .......... ......., an )  :   a1 x1  a2 x2  .......... .......... ....an xn
x 
 n

Example
The mapping L : R 2  R1 defined by L( x)  x1  x2 , is a linear transformation since
L( x   y )  ( x1   y1 )  ( x2   y2 )
  ( x1  x2 )   ( y1  y2 )
  L( x)   L( y )

Exercise
Let L : P1  P2 be defined as indicated. Is L a linear transformation? Justify your answer.
L[p(t)]=tp(t)+p(0)
2
a. L[p(t)]=tp(t)+t2+1
b. L(at+b)=at2+(a-b)t

1-2.3 Dual Space


Dual basis is a set of vectors that forms a basis for the dual space of a vector space. For finite
dimensional vector space V, the dual space V  is isomorphic to V, and for any given set of
basis vectors e1 , e2 ........, en  of V, there is an associated dual basis e1 , e2 ........., en  of V  .

Theorem
Suppose {v1, ……………, vn} is a basis of V over K. Let 1, …….n  V * be the linear
1 if i  j
functional defined by i (vi )  ij  
0 if i  j
Then (1 , .......... ....n ) is a basis of V*

Proof
We first show that (1 , .......... ....n ) spans V*. Let  be an arbitrary element of V* and
suppose  (v1 )  k1 , (v2 )  k2 .......... .........,  (vn )  kn
Set   k11  .......... .........  knn
Then  (v1 )  (k11  .......... ....  knn )(v1 )
 k11v1  .......... .........  knnv1
 k1.1 k2 . 0  .........  kn . 0  k1

Similarly for i  2, .......... ... n , we have


 k11vi    kii vi   knn vi  ki

Thus (vi) =  (vi) for i = 1, …………. , n Since  and  agree on the basis vectors,
    k11  k21   knn . Hence, spans V*.

We are supposed to show that {1, 2, ……….., n} is linearly independent.
Suppose a11 + a22 + ……………. + ann = 0
Multiply through by v1
0 = 0 (v1) = (a11 + a22 + …………… + ann)(v1)
= a11v1 + a22v1 + ………… annv1
= a1. 1 + a2.0 + …………… + an.0
= a1

0 = 0(v2) = (a11 + a22 + ……………. + ann)(v2)


= a11v2 + a22v2 + ………….. + ann(x2)
= a1.0 + a2.1 + ……………… an.0
3
= a2.
Similarly, for i = 3, ………….., n we have
0 = 0 (vi) = (a11 + ………….. + ann)(vi)
= a11vi + ………… aiIvi + ……….. annvi = ai
Hence a1 = 0, …………. an = 0. Hence {1, ……………. n} is linearly independent and so it
is a basis of V*

The above basis I is termed as the dual basis to {vi} or the dual basis by the kronecker delta
ij we have
1v1 = 1, 1v2 = 0, ………….. 1vn = 0
2v1 = 0, 2v2 = 1, …………., 2vn = 0
nv1 = 0, nv2 = 0, …………., nvn = 1

Example
Consider the following basis of R2 = {v1 = (2, 1), v2 = (3, 1)}. Find the dual basis {1, 2}.

Solution
The linear functional is given as 1(x, y) = ax + by and 2(x, y) = cx + dy. Such that
1v1 = 1 1v2 = 0 2v1 = 0, 2v2 = 1.

Thus
1v1  1 (2, 1)  2a  b  1
 or a = -1, b = 3
1v2  1 (3, 1)  3a  b  0
2v1  2 (2, 1)  2c  d  0
 or c = 1, d = -2
2v2  2 (2, 1)  3c  d  1 

Hence the dual basis is {1 (x, y) = -x + 3y, 2(x, y) = x –2y} (2)
Exercise
Consider the following basis of R2={v1=(1,1), v2=(1,2)}. Find the dual basis 1 , 2 

Theorem
Let {v1, ………….., vn} be a basis of V and let {1, ……….., n} be the dual basis of V*.
Then for any vector u V u = 1(u)v1 + 2(u)v2 + ……… + n(u)vn (1) and for any linear
functional  V*
 =  (v1) 1 + (2) 2 + ………….. + (vn) n. …………(2)

Proof
Suppose u = a1v1 + a2v2 + …………. + anvn …………… (3)
Then 1(u) = a11(v1) + a2 1(v2) + ……………. + an1(vn)

4
= a1. 1 + a2. 0 + …………. an. 0 = a1
2(u) = a12(v1) + a22(v2) + ……………. + an2(vn).
= a1. 0 + a2. 1 + ………….. + an. 0 = a2.
Similarly, for i = 3, ………….., n. we have
I(u) = a1I(v1) + a2 I(v2) + …………. + aiI(vi) + ………….. anI(vn).
= a1. 0 + a2. 0 …………+ ai. 1 + ……….. + an. 0 = ai
That is 1(u) = a1, 2(u) = a2, ………….., n(u) = an.

Substituting these resulst into (3) we obtain (1).


U = 1(u)v1 + 2(u) v2 + …………. + n(u)vn.

Next we prove (2). Applying the linear functional  to both sides of (1) we get.
(u) = 1(u) (v1) + 2(u) (v2) + ………… + n (vn).
= (v1) 1(u) + (v2) 2(u) + ………. + (vn) n(u).
= ((v1) 1 + ((v2) 2 + …………….+ ((vn) n.
= ((v1) 1 + (v2) 2 + …………… + (vn) n) (u)

Since the above holds for every u V,


 =  (v1) 1 + (v2) 2 + …………… + (Vn) n. as claimed.

1-2.4 Hermitain forms


Let V be a vector space of finite dimension over the complex field . Let f: V x V   be
such that
(i) f(au1 + bu2, v) = af(u1, v) + bf(u2, v) (f is linear in the first variable)
(ii) f(u, v) = f(v, u)
Where a, b   and ui, v  V. then f is called the Hermitian on V.
(iii) f(u, av1 + bv2) = f (av1  bv2 , u
= af (v1, u)  bf (v2 , u)
= a f (v1 , u )  b f (v2 , u )
= a f(u1 v1) + b f(u1 v2) (f is conjugate linear)
t
A matrix A = (aij) is Hemitian if and only if A = A ie. aij = a ji (A = At ) (a transpose
conjugate)

Example
2 2  3i 4  5i 
 
 2  3i 5 6  2i  is a Hermitian matrix.
 4  5i 6  2i  7 

5
3 2i 4  i
 
But  2  i 6 i  is not a Hermitian matrix even though it is symmetric.
4  i i 3 

A real matrix is Hermitian if and only if it is symmetric.
Example
4 3 5
 
 3 2 1
5  6 
 1

Let A be a Hermitian matrix. Show that f is a Hermitian in Cn where f is defined by


F(x, y) = xt AY

Solution
For all a, b  C and all x1, x2, Y  Cn
f(ax1 + bx2, Y) = (ax1 + bx2)t AY
= (ax1t + bx2t)AY
= ax1tAY + bx2tAY
= af(x1, Y) + bf(x2, Y)
Hence f is linear in the first variable.

Also
t
f ( X Y )  X t AY  ( X t AY )
t
= Y At X
t
= Yt AX A  A
= YtAX
= f(Y, X)
Hence f is a Hamitian form on Cn. (we consider XtAY is a scalar hence equal to its
transpose).

SESSION 2-2 Bilinear Forms

Let V be a. vector space of finite dimension over a field K. A bilinear form on V is a mapping
F: U x V  K which satisfies the ff.
(i) f (au1 + bu2, v) = af (u1, v) + bf(u2, v)
(ii) f(u, av1 + bv2) = af(u, v1) + bf(u, v2)
for all a, b K and all ui U and vi V
Let the basis of V be (v1, v2, ………….. vn) and (u1, u2, ………….. un) the basis of U

6
Example: (i)
Let f be the dot product on Rn; that is f(u, v) = u.v= a1b1 + a2b2 + ………. + anbn
where u = (ai) and v = (bi). Then f is a bilinear form in Rn.

(iii) Let A = (aij) be any n x n matrix over K. Then A may be viewed as a bilinear from
f in Kn by defining f(X, Y) = Xt AY
 a11 a12 .......... .a1n   y1 
   
 a21 a22 .......... .a2 n   y2 
; = (x1, x2, …….. xn) 
: : :  : 
   
a   
 ni a n2 .......... .a nn   yn 
n
= 
i . j 1
aij xiyi = a11 x1y1 + a12x1y1 + …………. annxnyn.

The above formal expression in variables xi, yi, is termed bilinear polynomial corresponding
to the matrix A.

(iv) Define a function H: R2xR2  R by


 a   b1    a1   b1 
H   1 ,    = 2a1b1 + 3a1b2 + 4a2b1 – a2b2 for

 ,   R2
 2 
a  b2    a2   b2 
We show that H is a bilinear form on R2
2 3 a  b 
Let A =  , x   1 , y   1  then H(x, y) = xt Ay.
 4  1  a2   b2 
We denote the set of bilinear form in V by (V). A vector space structure is placed on (V)
by defining f + g and kf by (f + g) (u, v) = f(u, v) + g(u, v)
(kf) (u, v) = k f(u, v).
For any g, f  (V) and any k  K. Let v = Fn, where the elements are considered as column
vectors. For any AMnxn (F) define H: V x V  F by H(x, y) = xt Ay. for x, y V.

Example
Let f be the bilinear form on R2 defined by f((x1, x2), (y1, y2)) = 2x1y1 – 3x1 y2 + x2y2.
(i) Find the matrix A of f in the basis {u1 = (1, 0), u2 = (1, 1)}.
(ii) Find the matrix B of f in the basis {v1 = (2, 1), v2 = (1, -1)}
(iii) Find the transition matrix P from the basis {u1} to the basis {vi}, and verify that B
= PtAP

Solution
Set A = (aij) where aij = f(ui, uj).
2  3
The given matrix is   =D
0 1 

7
2  3  1 
a11 = f(u1, u1) = U1T Du1 = (1, 0)    = 2
0 1   0 

a12 = f(u1, u2)


2  3   1   1
= (1, 0)      (1, 0)   = -1
0 1   1  1
2  3  1   2
a21 = f(u2, u1) = (1, 1)      (1, 1)   = 2
0 1  0 0
2  3   1   1
a22 = f(u2, u2) = (1, 1)      (1, 1)   = 0
0 1   1  1
2  1
Thus A =   is the matrix of f in the basis (u1, u2)
2 0 
3 9
B =   is the matrix of f in the basis {v1, v2}.
 0 6 
v1  (2, 1)  (1, 0)  (1, 1)  u1  u2
v2  (1,  1)  2(1, 0)  (1, 1)  2u1  u2
1 2 t 1 1
P =   P   
 1  1  2  1
Check to see that B  Pt AP

Exercise
Let f be the bilinear form on R2 defined by
f   x1 , x2  ,  y1 , y2    3x1 y1  2 x1 y2  4 x2 y1  x2 y2
a. Find the matrix A of f in the basis u1  1,1 , u2  1, 2 
b. Find the matrix B of f in the basis v1  1, 1 , v2   3,1
c. Find the transition matrix P from ui  to vi  and verify that B  Pt AP

2-2.1 Symmetric bilinear form

Definition
A bilinear form f on V is symmetric if f(x, y) = f(y, x) or f(u, v) = f(v, u) for every u, v V
If A is a matrix representation of f, we can write f(x, y) = xt AY = (xt AY)t = YtAtX.

8
We use the fact that XtAY is a scalar and therefore equal its transpose. Thus if f is symmetric
YtAtX = f(X, Y) = f(Y, X) = YtAX. And since this is true for all vectors X, Y it follows that
A = At or A is symmetric conversely if A is symmetric, then f is symmetric.

2-2.2 Diagonalization of symmetric matrices


1 2  3
 
Let A  2 5  4  , a symmetric matrix
 3  4 8 
 
We want to diagonalize the above matrix. First we form a block matrix (A, I) and use
elementary row operations.
1 2 3 : 1 0 0
 
e.g. (A, I) =  2 5 4 : 0 1 0
 3  4 1 
 8:0 0
We apply R2  -2R1 + R2 and R3  3R1 + R3 to (A, I)
1 2 3 : 1 0 0
 
0 1 2 : 1 1 0  C2  -2C1 + C2 and C3  3C1 + C3 to A to obtain,
0 1 : 3 1 
 2 0

1 0 0 :1 0 0
 
0 1 2 : 2 1 0
0 1 : 3 0 1 
 2
We next apply the operation R3  -2R2 + R3
1 0 0 :1 0 0 1 0 0 :1 0 0
   
0 1 2 : 2 1 0  C3   2C2  C3 and then  0 1 0 : 2 1 0
0 5 : 7 2 1  0 0 5 : 7 2 1 
 0 

Now A has been diagonalized. We set


1 2 7 1 0 0
   
P = 0 1  2  and then P AP   0
T
1 0
0 1 0  5 
 0  0

Assignment

1 1 3
 
Let A=   1 2 1 , a symmetric matrix. Use elementary row operation to diagonalize A.
3 1 
 1

9
Definition
A bilinear form H on a finite-dimensional vector space V is called diagonalizable if there is
an ordered basis  for V such that  (H) is a diagonal matrix. If A is the matrix of bilinear
form and A has distinct eigenvalues then A is diagonalizable i.e. P-1AP = B.

Where P is the matrix of eigenvectors and B is the diagonal matrix with eigenvalues on the
main diagonal.

Example
1 2
Consider the matrix A =  .
3 2 
1  2
Then we have =0
3 2  
(1 - ) (2 - ) – 6 = 0
2 - 3 + 2 – 6. = 0
2 - 3 - 4 = 0.  = 4,  = -1
Ax = x

When   4 , we have
1 2   x1   x1 
     4 
 3 3   x2   x2 
x1 + 2x2 = 4x1
3x1 + 2x2 = 4x2
-3x1 + 2x2 = 0
3x1 – 2x2 = 0
 2
3x1 = 2x2 =  
3
When   1 , we have
 1 2   x1   x1 
     1  
 3 2   x2   x2 

 x1  2 x2   x1
3 x1  2 x2   x2

 2 x1  2 x2  0
3 x1  3 x2  0

 1 
 x1   x2   
1
10
 1 1
 2 1 1
 5 5
p=   p   
3 1   3 2 
 
 5 5
Then A is similar to the diagonal matrix
 15 1
5  1 2   2 1 4 0
B = p-1 AP =  3
2  3      
 5 5   2 3 1  0  1

The diagonal elements 4 and –1 of the diagonal matrix B are the eigenvalues corresponding
to the given eigenvectors.
1 1 4 1  1 4
 
Let A =  3 2  1  Then det( A   I )  3 2 1 = 0 The eigenvalues
2 1  1   (1   )
 2 1
are 1 = 1 2 = -2 and 3 = 3. and the corresponding eigenvectors are V1 =
  1 1  1 
     
 4  V2    1 V3   2 
1  1  1 
     
1 1 1 1 2 3
  -1 1  
 P = (V1, V2, V3) =  4 1 2  p =  2  2 6
6 
1
 1 1   3 0  3 
1 2 3 1 1 4  1 1 1
1      
 p-1 AP = B = 2 2 6  3 2 1  4 1 2
6 
 3 0  3  2
 1  1 1
 1 1 
1 0 0
 
= 0 2 0
0 3 
 0
Definition
The difference between the number of positive eigenvalues and the number of negative
eigenvalues is called the signature.

Example
If a matrix A has as the eigenvalues, then the signature is
given as 8-1-1= 6

11
Exercise
1 1 i 2i 
 
Let H = 1  i 4 2  3i  , a Hermition matrix. Find a non-singular matrix
  2i 2  3i 7 
 
P s.t. PT H P is diagonal. Find also the signature(s

UNIT 2
INNER PRODUCT SPACES AND THEIR
PROPERTIES

SESSION 1.3 Inner Product Spaces

Definition
. Then is a scalar. The number is called the inner product of and and
is written as . This is referred to as dot product.

If , then the inner product of and is

 v1 
 
u  v  u t v   u1 , , un   
 
 
 vn 

(i) Compute and when and

Solution

Commutativity of the inner product holds in general.

12
Theorem
Let and be vectors in , and let c be a scalar, then
(a)
(b)
(c)
(d) and

1-3.1 The Norm of a vector


If is in , with entries , then the square root of u  v is defined because u  v is
nonnegative.

Definition
The length (or norm) of is the nonnegative scalar defined by:
v  vv  v  v  2
1
2
2  v and
2
n

Find the norm of

Let us suppose that , we can identity as a geometric point in the plane.


Then coincides with the notion of the length of the line segment from to .

x2

 v1 , v2 

v12  v22
| v2 |

| v1 |
0 x1

By the Pythagorean theorem,


Compute

13
v
A vector whose length is 1 is called a unit vector. The unit vector of is given as u  .
v
The process of creating from is sometimes called normalising . We say that is in the
direction as .

Example:
Let . Find a unit vector in the same direction as .

Solution
v  v  v  12  (2)2  22  9  3

Check that
 1   2   2 
2 2 2

v  vv          1
2

 3  3   3

Definition
For and in , the distance between and is
If u  u1 , , un , v  v1 , , vn

 u1  v1    u2  v2    un  vn 
2 2 2
Then u  v  

Eg. Compute the distance between the vectors and

Exercise
Let u = (2,3,2,-1) and v = (4,2,1,3). Find,
1. The norm of each of the vectors
2. The distance between the two vectors
3. The angle between the two vectors

14
SESSION 2-3 Orthogonal Vectors

v 0 v

 u  u  v   v  u  v 
 u u  u v  v u  v v
 u u  v v  u v  v u

Two square distances are equal 

Definition
Two vectors and in are orthogonal to each other if

Theorem
Two vectors and are orthogonal 

2-3.1 Angles in R2 and R3


If and are nonzero vectors then the angle  between the two line segments from the
origin to the points identified with and is given as

15
Exercise

Let

1. Compute and
2. Find a unit vector in the direction of
3. Show that is orthogonal to
4. Show that is an orthogonal set where

2-3.2 The Cauchy-schwarz inequality


For all in …….. (1)

Proof
If then both sides of (1) are zero and hence (1) is true in this case.
If , let W be the subspace spanned by .
Recall that for any scalar .
v, u v, u v, u u, v
Thus projwv  u  u  2
u 
u, u u, u u u

Since , we have hence the result.

2-3.3 The Triangle Inequality


For all in

Proof
u  v  u  v, u  v  u, u  2 u, v  v, v
2

 u  2 u, v  v
2 2

Cauchy-Schwarz

Example
Find the length and the distance between the vectors u = (2, 3, 1) and v = (4, 1, -3). Find also
the angle  between the two vectors.
16
||v|| = (v, v)  (2, 3, 1). (2, 3, 1)  4  9  1  14
||u|| = (u, u)  (4, 1,  3). (4, 1,  3)  16  1  9  26
|(v, u)| = ||v||. ||u|| = 14. 26  364  2 91
(u. v) (4, 1,  3). (2, 3, 1)
 cos  =   (8  3  3
|| u || . || v || 364
 8 
  = cos-1   = 65.20
 364 

2.3.4 Matrices
(Ext. of IPS. and norm)
If the vectror space is V = M(mxn) matrix then the norm function is
(a) ||A|| = Max |Aij| is called the max norm.
i, j

n 
(b) ||A|| = Max  | Aij | is called the max row sum norm
 j 1 
2  3 2
E.g. A =   for i = 1  | aij | = |2| + |-3| = 5 for i = 2
1 2  j 1
2


j 1
|a2j| = |1| + |2| = 3

By def. ||A|| = max (5, 3) = 5


1

 n 2
(c) Frebonius norm ||A||F =   ( Aij ) 2 
 i, j 
 a11 a12   2  3
For A =     
 a21 a21  1 2 

||A||F = a112  a122  a21 


1
2
 a21
2 2

= 4  9  1  4  18
1

 n p
(d) A P norm is defined by ||A||p =   | Aij | p  , where P = 1, 2, 3 p = 2  Frebonius
 ij 
n
For a vector space V = R , the norm functions are
(a) ||v|| = max |Vj| called max. absolute norm
j

2  3
E.g V =  
4  6 
||v|| = max |Vj| = max [|2|, |-3|, |4|, |-6]
j

= max [2, 3, 4, 6] = 6

17
1

n 2
(b) ||v||2 =  | V j |2  called Euclidean norm
 j 1 
 
1
 ( 6 ) 2
 ||v||2 = 2 2  (3) 2  4 2 2

= 4  9  16  362 
1
65
1

 n p
(c) ||v||p =   | v j | 
 p

 j 1 
||v||3 = | 2 |3  | 3 |3  | 4 |3  | 6 |3 3
1

= 8  27  16  2163  3 315
1

Let X be the vector (4, -5, 3)T in R3. Compute


||x||1 ||x||2 and ||x||
||x||1 = |4| + |-5| + |3| = 12
||x||2 = 16  25  9  5 5
||x|| = max (|4|, |-5|, |3|) = 5
1    4
Let x1 =   and x2 =   .
 2 2 
||x1||2 + ||x2||2 = 4 + 16 = 20
||x1 + x2||2 16 max of the x1, x2 square
||x1||22 + ||x2||22 = 5 + 20 = 25 = ||x1 + x2||22

Exercise
1. (i) Let x1, x2, ………….. xn be distinct real numbers. For each pair of polynomials in
n

  p( x )
2
pn define (p, p) = i where xi = (I – 3)/2 for i = 1, ………… 5
i 1

(ii) Show next the functions x and x2 are orthogonal.


1

 5 2
(iii) Define the norm in p5 by ||p|| = ( p, p)    [ p( xi )]2 
 i 1 
(iv) Compute (a) ||x||, (b) ||x || (c) the distance between x and x2
2

2. (i) Find the length and the distance between the vectors .
(ii) Find also the angles between the two vectors.

18
UNIT 3

ORTHOGONALIZATION PROCESS AND BEST


APPROXIMATION

Introduction
In this unit we treat Orthogonal and Orthonormal sets. We introduce you to orthogonal
projection and the Gram-Schmidt Orthogonalization process. We also treat the Best
approximation and the method of least square.

SESSION 1-4 Orthogonalization Process

1-4.1 Orthogonal Sets

Definition
A set of vectors in is said to be an orthogonal set if each pair of distinct
vectors from the set is orthogonal, that is if whenever .

Example

Show that is an orthogonal set where

Solution
Let us consider the three possible pairs of distinct vectors, namely, and

Each pair of distinct vectors is orthogonal, and so is an orthogonal set.

19
Theorem 1
If is an orthogonal set of nonzero vectors in , then S is linearly
independent and hence is a basis for the subspace spanned by S

Proof
If for some scalars then
0  0  u1  (c1u1  c2u2   c pu p )u1
 (c1u1 )u1  (c2u2 )u1   (c pu p )u1
 c1 (u1u1 )  c2 (u2u1 )   c p (u pu1 )

 c1 (u1  u1 )
Since is nonzero, is not zero and so . Similarly, must be zero. Thus
S is linearly independent.

Let S be an orthogonal set of nonzero vectors in and let W be the subspace spanned by S.
then S is called an orthogonal basis for W because it is both an orthogonal set and a basis for
W. If there are n vectors in S, then W and S is an orthogonal basis for .

Theorem 2
Let be an orthogonal basis for a subspace W of . Then each y in W, has a
unique representation as a linear combination of . In fact if
Then

Example

Let where and is an orthogonal basis for

. Express the vector as a linear combination of the vectors in S

Solution

Then by theorem 2 above


y  u1 y  u2 y  u3
y u1  u2  u3
u1  u1 u2  u2 u3  u3

20
11 12 33
 u1  u2  u
11 6 33 3
2
 u1  2u2  2u3

Exercise
Which of the following are orthogonal sets of vectors?
a. {(1,-1,2),(0,2,-1),(-1,1,1)}
b. {(1,2,-1,1),(0,-1,-2,0),(1,0,0-1)}
c. {(0,1,0,-1),(1,0,1,1),(-1,1-1,2)}

1-4.2 Orthonormal sets


A set is an orthonormal set if it is an orthogonal set of unit vectors. If W is the
subspace spanned by such a set, then by Theorem 1, is an orthonormal basis for
W, since the set is automatically linearly independent. The simplest example of an
orthonomal set is the standard basis . For

Example
Show that is an orthonormal basis of where

Solution
v  v  3
1 2  2  1 0
66 66 66
and
Thus is an orthogonal set. Also

v1  v1  9  1  1 1
11 11 11

v2  v2  1  4  1  1
6 6 6

v3  v3  1
 16  49  1
66 66 66
Which shows that and are unit vectors. Thus is an orthonomal set.

Theorem
A matrix has orthonomal columns if and only if ( I the identity matrix)

21
Proof
Let assume that U has only three columns in .
Let

Then (1)

We know that the columns of U are orthogonal 


(2)
The column of U all have unit length if and only if (3)

From (2) and (3) we have

Theorem
Let U be an matrix with orthonomal columns and let and be in . Then
(a)
(b)
(c) if and only of

Properties (a) and (c) say that linear mapping preserves length and orthogonality.

Example
Let

 1 2 
 2 3 
 1  
U  2  and x   2 
2 3  3 
   
 0 1 
 3 

Verify that

Solution

22
1-4.3 Orthogonal projection
Given nonzero vectors u and y in , consider the problem of decomposing y into the sum of
two vectors, one a multiple of u and the other orthogonal to u.

We wish to write where for some scalar α, and z is some vector


orthogonal to u. See figure below,

z  y  yˆ y

ŷ   u u
Find α to make orthogonal to u.
Given any scalar α, let , then is orthogonal to u if and only if

Then is orthogonal to u if and only and

Then we have the following


(a) is the orthogonal project of onto u
(b) is the component of orthogonal to u .

Example
Let and . Find the orthogonal projection of unto u. Then write as the
sum of two orthogonal vectors, one in space and one orthogonal to u.

Solution

23
The orthogonal projection of onto u is and the component
of orthogonal to u is

The sum of these two vectors is , that is

.
Which shows that is an orthogonal set.

Exercise

1. Show that , and is an orthogonal

basis for and express as a linear combination of the .


2. Let and compute the orthogonal project of onto u.

1-4.4 Gram-Schmidt orthogonalization process:


Lemma
An Orthonormal set u1 , u2 , , un  is linearly independent and for any
v V , the vector w  v, u1 u1  v, u2 u2   v, ui ui is orthogonal to each set ui

Theorem
Let {v1, ………….., vn} be an arbitrary basis of an IPS. V. Then there exists an orthonormal
basis {u1, …………. un} of V s.t. the transition matrix from {vi} to {ui} is triangular; that is
for i = 1, …………. n
ui = ai1 v1 + ai2 v2 + ……….. aiivi

Proof
v1
We set ui = ; then {u1} is orthonormal. Next we set w2 = v2 – (v2, u1) u1 and u2 =
|| v1 ||
w2/||w2||.

24
By the above Lemma above, w2 (and hence u2) is orthogonal to u1; then {ui, u2} is
orthonormal.
Next we set W3 = v3 – (v3, u1) u1 – (v3, u2) u2 and u3 = w3/||w3||.

By the Lemma above, w3 (and hence u3) is orthogonal to u1 and u2; then {u1, u2, u3} is
orthonormal.

In general, after obtaining {u1, ……….. un} we set


wi 1
wi 1  vi 1  vi 1 , u1 u1   vi 1 , ui ui and ui 1  , wi 1  0
wi 1
By induction we obtain an Orthonormal set u1 , u2 , , un  which is independent and hence a
basis of V. The specific construction guarantees that the transition matrix is indeed triangular.

Example
1  1 4
 
1 4  2
Given  find the orthonormal basis for the column space of A.
1 4 2
 
1  1 
 0 
Let v1 = (1 1 1 1)t v2 = (-1 4 4 –1) and v3 = (4 –2, 2, 0)

Step 1 ||v1|| = 12  12  12  12  4 =2
v1 1
u1 =  (1 1 1 1).
|| v1 || 2
Step 2
w2 = v2 – (v2, u1) u1 = (-1, 4, 4, -1) – (v2, u2) u1
1 
= (-1, 4, 4, -1) - 3  (1, 1, 1, 1) .
2 
= 5 2 [-1, 1, 1, -1]
w2 5 ( 1, 1, 1,  1) 5 ( 1, 1, 1,  1) 1
u2 =  2  2 = (1, 1, 1,  1)
|| w2 || 25 5 2
w3 = v3 – (v3, u1) u1 – (v3, u2)u2
1
= (4 – 2, 2,0) (v3, u1) (-1, 1, 1, 1) – (v3, u2)  1
2
(1, 1, 1,  1)
2
= (4, -2, 2, 0) - (1 1 1 1)  (1 1 1  1)
= (4, -2, 2, 0) – (2, 0, 0, 2).
= (2 – 2 2 –2)
2(1 –1 1 –1)
w3 2(1  1 1  1) 1
u3 =   (1 –1, 1 –1)
|| w3 || 16 2
Check that |uj| = 1 for j = 1, 2, 3 and u1 u2 = u1. u3 = u3 = 0
25
Exercise 1:
Given {v1 = (1, 1, 1) v2 = (0, 1, 1), v3 = (0, 0, 1) and using Gram-Schimdt othogonalization
process find the orthonormal vectors {ui} by normalising vi

Exercise 2:
Consider the basis S  u1 , u2 , u3  for R 3 where
u1  1,1,1 , u2   1,0, 1 , and u3   1, 2,3 .
Use the Gram-Schmidt process to transform S to an orthonormal basis in R 3 .

SESSION 2-4 Best Approximation

Theorem
The Best Approximation Theorem
Let W be a subspace of be a vector in and be the orthogonal projection of onto
W determined by an orthogonal basis of W. Then is the closest point in W to , in the
sense that for all v in W distinct from .

The vector is called the best approximation to by elements of W. The distance from to
given by , can be regarded as the ‘error’ of using in place of . By the above
theorem, the error is minimized when .( is the error).

Example

If and and W = span . Then the closest point in

W to is

Example 2
The distance from a point in to a subspace W is defined as the distance from to the
nearest point in W. Find the distance from , to W = span where

By the Best Approximation Theorem, the distance from to W is where


. Since is an orthogonal basis for W we have

26
 5  1   1 
15 21 1  7   
y  u1  u2   2    2    8 
30 6 2  2   
 1  1  4 

 1   1   0 

y  y   5    8    3 
 10   4   6 
     

The distance from y to W is 45  3 5


Exercise
Find the bast approximation to z by vectors of the form

(a)

(b) Let and .

Find the distance from to the plane in spanned by and

2-4.1 Method of Least Squares


If we have systems equations of the form which are inconsistent, the best way to
solve such systems is to find an that makes closest to b, thus an approximation of to
b. The smaller the distance between and given by , the better the
approximation. We are therefore to find an that makes very small.

Definition
If A is and b is in , a least-squares solution of is an in s.t
for all in
The term least-squares arises from the fact that is the square root of a sum of
squares. We therefore look for an that makes the closest point in to be.

2-4.2 Solution of the general least-squares problem


Given A and b in we apply the Best Approximation Threorm to the subspace ColA .
(Let ). Because is in the column space of A, the equation is
consistent and there is an in such that

Since is the closest point in to b, a vector is a least-squares solution of iff


satisfies
27
2-4.3 Orthogonal Decomposition Theorem
Let W be a subspace of that has an orthogonal basis. Then each in can be within
uniquely in the form where is in W and z is in . If is any
orthogonal basis of W then and

The vector is called the orthogonal projection of onto W and often is within as .
Let then is a unit vector of u. Suppose that satisfies , then
by the above theorem, the projection has the property that is orthogonal to . So
is orthogonal to each column of A. If is any column of A then
and since each is a row of .

normal equations of

Theorem
The set of least-square solutions of coincides with the nonempty solutions of the
normal equations

Example
Find a least-squares solution of the in consistent system for

Solution

Then becomes

We can use elementary row operation to solve but is ivertible hence

So

28
Exercise
Find a least-square solution of for

Theorem
The matrix is invertible if and only if the columns of A are linearly independent. In this
case, the equation has only one least-squares solution and is given by
When a least-square solution is used to produce as an approximation
to b, the distance from b to is called the least-squares error of this approx.

Example

Given and determine the least-square error in the least-square

solution of

Solution

error

2-4.4 Alternative calculations

Example

Find a least-square solution for for

Solution
29
Let and be the column vectors of A and they are also orthogonal. The orthogonal
projections of b onto is given by

in is a list of weights that will build out of the columns of A. Hence the weight to
place on the columns of A to produce

Exercise
Find a least-square solution for

1.

2.

Determine the least-squares error in the least-squares solution

Method of Least squares (continuation)


In curve fitting we are given n points (pairs of numbers); (x1, y1), (x2, y2), …………., (xn, yn)
and we want to determine a function f(x) such that f(x j) ~ yj, j = 1, 2, …………, n. This type
of function can be a polynomial, exponential, sine and cosine functions. If we require strict
equality f(x1) = y1, f(x2) = y2, …………, f(xn) = yn and use polynomials of sufficiently high
degree, we may apply Lagrange interpolation or Newtons interpolation. But in some
situations, these may not be appropriate solution of the actual problem.

For instance if we consider the Lagrange polynomial f(x) = x3 – x + 1 with the following four
points; (-3. 3, 0.103), (-0.1, 1.099), (0.2, 0.808), (1.3, 1.897) and graph the points, we see that
they lie nearly on a straight line. A widely used principle for fitting straight lines is the
Method of Least Squares by Gauss.
In the Method of Least Squares, the straight line y = a + bx should be fitted through the given
points (x1, y1), (x2, y2), …………, (xn, yn) so that the sum of the squares of the distances of

30
those points from the straight line is minimum, where distance is measured in the vertical
direction (the y-direction).

The point on the line with abscissa xj has the ordinate a + bxj. Hence its distance from (xj, yj)
n
is |y – a – bxj| and that sum of square is q = 
j 1
(yj – a – bxj)2 where q depends on a and b

y j  a  bx j
y = a + bx
a + bxj
1 x
xj

A necessary condition for q to be minimum is


q
= - 2 (yj – a – bxj) = 0
a
q
= -2  xj(yj – a – bxj) = 0
b
Where we sum over j from 1 to n. Writing each sum as three sums and taking one of them to
the right, we obtain the following;
an + bxj = yj
axj + bxj2 = xjyj

The above equations are called the normal equations of our problem.

Example
Using the method of least squares, fit a straight line to the points; (-1.3, 0.103), (-0.1, 1.099),
(0.2, 0.808), (1.3, 1.897).
Solution

31
n = 4, xj = 0.1, xj2 = 3.43, yj = 3.907, xjyj = 2.3839.
Hence the normal equations are
4a + 0.10b = 3.9070
0.1a + 3.43b = 2.3839.
Solving the above equations simultaneously, we have a = 0.9601, b = 0.6670 and obtain y =
0.9601 + 0.6670x.

Question
Fit a straight line to the points (0,3), (2, 1), (3, -1), (5, -2) by the method of least squares.

Curve fitting by polynomial of degree m.


Our method of curve fitting can be generalized from a polynomial y = a + bx to a polynomial
degree m
P(x) = b0 + b1 x + ………… + bmxm where m  n – 1, parameters b0, ……….., bm.
We have m + 1 conditions,
q q
 0,......... .., = 0 which give a system of m + 1 normal equations.
b0 bm
In the case of a quadratic polynomial, p(x) = b0 + b1 x + b2x2, the normal equations are
b0n + b1 xj + b2 xj2 = yj
b0xj + b1 xj2 + b2 xj3 = xjyi
b0 xj2 + b1 xj3 + b2xj4 = xj2yi

Exercise
1. Fit a parabola p(x) = b0 + b1x + b2x2 to the points (-1, 3), (1, 1), (2, 2), (3, 6) by the method
of least squares.
2. Derive a normal equations to a cubic parabola p(x) = b0 + b1x + b2x2 + b3x3 and hence fit a
cubic parabola by least squares to (-2, -8), (-1, 0), (0, 1), (1, 2), (2, 12), (4, 80).

32
UNIT 4

QUADRATIC FORMS AND DIAGONALIZATION


OF MATRICES

Introduction
In this unit we introduce you to quadratic forms, similar matrices and how to diagonalize
matrices and quadratic symmetric matrices

SESSION 1-5 Diagonalization of Matrices


1-5.1 Quadratic Forms
A quadratic form on is a function Q defined on whose value at a vector in can be
computed by an expressing of the form , where A is an symmetric
matrix. The matrix A is called the matrix of the quadratic form. The simplest example of a
quadratic form is .

A quadratic form Q per definition;


1. Positive definite if Q(x)>0, x  0
2. Negative definite if Q ( x )  0, x  0
3. Positive semi-definite if Q ( x ) , x  0
4. Negative semi-definite if Q ( x )  0, x  0
5. Indefinite if Q(x) is both positive and negative.

Theorems:
1. If A is a symmetric, then any two eigenvalues from different eigenspaces are
orthogonal.
2. An nxn matrix A is orthogonally diagonalizable if and only if A is a symmetric
matrix.

Quadratic forms and eigenvalues: Let A be an nxn symmetric matrix. Then a quadratic form
x t Ax is;
a. Positive definite if and only if the eigenvalues of A are all positive.
b. Negative definite if and only if the eigenvalues of A are all negative.
c. Positive semi-definite if and only if one of the eigenvalues of A is 0, and the others
are positive.
d. Negative semi-definite if and only if one of the eigenvalues of A is 0, and the others
are negative.
33
e. Indefinite if and only if A has both positive and negative eigenvalues.

Example
Let , compute for the following matrices (a) (b)

Solution
(a)

 3 2   x1 
X t AX   x1 , x2    
 2 7   x2 
x 
(b)   3 x1  2 x2 , 2 x1  7 x2   1 
 x2 
 3 x1 x1  2 x1 x2  2 x1 x2  7 x2 x2
 3 x12  4 x1 x2  7 x22

Example
For in , let
Write this quadratic form as

Solution
Note that the coefficients and go on the diagonal of A. To make A symmetric, the
coefficient of for must be split evenly between the and entries in A.
The coefficient of

Example
Let . Compute the value of for and

Solution

34
Exercise
Let Q(x,y,z)= . Write this in the form .

Change of Variable in a Quadratic Form


If represents a variable vector in , then a change of variable is an equation of the form
, or equivalently, , where P is an invertible matrix and u is a new
variable vector in . Here u is the coordinate vector of relative to the basis of
determined by the columns of P. If the change of variable (i) is made in a quadratic form
, then and the new matrix of the quadratic
form is . If P orthogonally diagonalizes A, then and . The
matrix of the new quadratic form is diagonal.

Example
Let . Then the matrix of quadratic form is given by

Solution
First we orthogonally diagonalize A. The eigenvalues are and and the
associated unit eigenvectors are

Note that these are automatically orthogonal, hence they provide an orthonormal basis for

Let P ,D=

The matrix A above is an indefinite matrix since it has both positive and negative
eigenvalues. Hence the eigenvectors are not orthogonal.

Then and . A suitable change of variable is , where


and

Then

35
Find the value of when
We know that , we have

Hence

Exercise
Make a change of variable x  Pu that transforms the quadratic q( x)  5x12  4x1x2  5x22 into
quadratic form with no cross product term and hence find the value of Q( x) when x  (2, 2)
Remark
In some cases especially when there is repeated eigenvalue, the symmetric matrix does not
yield mutually orthogonal eigenvectors. In order to get orthogonal matrix P, we use gram
Schmidt process to construct mutually orthogonal eigenvectors from the original eigenvectors
and hence obtain the needed orthogonal matrix for PTAP = B.

1-5.2 Similar Matrices

Definition
A matrix B is said to be similar to a matrix A if there is a non-singular matrix P such that
B  P1 AP

Example 1
Let

Then

And

The following elementary properties hold for similarity.


1. A is similar to B.
2. If B is similar to A, then A is similar to B

36
3. If A is similar to B and B is similar to C, then A is similar to C
By property 2 we replace the statements “A is similar to B” and “B is similar to A” by “A and
B are similar”.

Definition
We shall say that the matrix A is diagonalizable if it is similar to a diagonal matrix. In this
case we also say that A can be diagonalized.

Example 2
If A and B are as in Example 1, then A is diagonalizable, since it is similar to B.

Theorem
Similar matrices have the same eigenvalues
Proof
Let A and B be similar. Then B  P 1 AP , for some nonsingular matrix P. We prove that A
and B have the same characteristic polynomials, f A ( ) and f B ( ) , respectively. We have
f B ( )  det ( I n  B)  det ( I n  P 1 AP)
 det( P 1 I n P  P 1 AP )  det( P 1 ( I n  A) P )
 det( P 1 ) det( I n  A) det( P) (1)
 det( P 1 ) det( P) det( I n  A)
 det( I n  A)  f A ( )
Since f A ( )  f B ( ) , it follows that A and B have the same eigenvalues.

It follows from Exercise 1, that the eigenvalues of a diagonal matrix are the entries on its
main diagonal. The following theorem establishes when a matrix is diagonalizable.

Theorem
An n x n matrix A is diagonalizable if and only if it has n linearly independent eigenvectors.
We see that the eigenvectors of the matrix in Example 1 above are linearly independent.

Theorem
If the roots of the characteristic polynomial of an nxn matrix A are all distinct, then A is
diagonalizable.

Exercise
1. Let A be a 3x3 matrix whose eigenvalues are -3 4 and 4and associated eigenvectors
 1   0   0
     
are  0  ,  0  and  1  respectively. Find a diagonal matrix D that is similar to A.
 1  1 1
     
37
 5 3 1
2. Let A    , find the matrix B such that P AP  B where P is an invertible
 3 5
matrix. What can you say about the matrices A and B?

1-5.3 Diagonalization of Quadratic Symmetric Matrices

Given the quadratic form q(x, y, z) = 2x2 + 6x1x2 + 5y2 – 2yz + 2z2, we can rewrite the
quadratic form as q(x, y, z) = 2x2 + 3x1y2 + 3yx1 + 5y2 – yz – zy + 2z2
2 3 0
 
Hence A =  3 5 1 
0 1 2 

Now the characteristic polynomial is (-1) ( - 2) ( - 7)  and hence A has the eigen values 1
= 2, 2 = 7 and 3 = 0. The corresponding eigenvectors are
1  3    3
1   1   1  
V1 =  0 , V2   5  and V3  2 
10   35   14  
3   1 1 
v1v2 = 0 v1v3 = 0, v2v3 = 0  orthogonal

Hence v1, v2, v3 are orthonormal  B = PTAP or PAPT

 1 3 3 
 10 35 14 
P = 0 5 2 
 35 14 
 3 1 1 
 10 35 14 

2 0 0
 
Hence (f) = B = PTAP =  0 7 0
0 0 
 0

Example
Given the quadratic form q(x, y, t) = 5x2 + 8xy + 5y2 + 4xz + 4yz + 2z2 = 100
 q(x, y, z) = 5x2 + 4xy + 4yx + 5y2 + 2xz + 2zx + 2yz + 2zy + 2z2 = 0
Hence the matrix of the quadratic form is
5 4 2 5 4 2  x
     
A = 4 5 2  ie ( x, y, z )  4 5 2   y  = 100
2 2  2 2   z 
 2  2

Diagonalizing A

38
5 4 2
4 5 2 = 0  1 = 1 2 = 1 3 = 10
2 2 2
From (5 - ) 1 + 42 + 23 = 0
41 + (5 - )2 + 23 = 0
21 + 22 + (2 - )3 = 0
The eigenvectors are
  12   1 
   3 2  23 
 
V1 =  12 , V2    3 1 3  V3   2 3 
   
 0   4  1 
   3
 3 2
(i) They have been normalized since no repeated eigenvalues
(ii) V1 V2 = 0, V1 V3 = 0, V2 V3 = 0  orthogonal

Example
Let q be the quadratic form associated with the symmetric bilinear form f. Show that the
1
polar form of f: f(u, v) = [q(u + v) q(u) – q(v)].
2

Solution:
q(u + v) - q(u) – q(v) = f(u + v, u + v) – f(u, u) – f(v, v)
= f(u, u) + f(u, v) + f(v, u) + f(v, v) – f(u, u) – f(v, v)
= 2 f(u, v)
1
Assume the characteristic of the field f is not two  q(u + v) – q(u) – q(v) = f(u, v)
2

Exercise
Consider the quadratic forms K on a real inner product space V, find a symmetric bilinear
form H s.t. K(x) = H(x, x)  xV. Then find an orthonomal basis  for V s.t.  (H) is a
diagonal matrix.
x
 
(a) K: R  R defined by K  y  = 3x2 + 3y2 + 3z2 – 2xz
3

z 
 
x
(b) K: R2  R defied by K   = 7x2 – 8xy + y2
 y

Theorem: Diagonalization of a matrix.


If an n x n matrix A has a basis of eigenvector, then D = X-1AX is diagonal, with the
eigenvalues of A as the entries on the main diagonal. Here X is the matrix with these
eigenvectors as column vectors.

39
5 4  4 1 
E.g. A =   has eigenvectors   and  
1 2  1    1

4 1 
Hence X =  
1  1
1  1  1 5 4 4 1  0.2 .2   24 1
X-1AX =  1     
5  4  1
 2 1
  1  .2  .8  6  1
6 0
=  
0 1 
Example
Diagonalize the matrix,
 7.3 0.2  3.7 
 
A =   11.5 1.0 5.5 
17.7  9.3 
 1.8

Solution
The characteristic polynomial is -3 -2 + 12  = 0. Hence the roots (eigenvalues of A) are 1
= 3, 2 = -4, 3 = 0. Eigenvalues
  1 1   2 1 1 2   .7 .2 .3 
        1  
 3 ,   1, 1  X   3 1 1  X    1.3  .2 .7 
  1 3   4 1 4   .8  .2 
       3  .2
v1 v2 v3
3 0 0
 
 D = X-1AX =  0 4 0
0 0 
 0

1-5.4 Reduction to a canonical forms


For quadratic forms we have Q = XTAX. If A is real symmetric, then A has an orthogonal
basis of n eigenvectors. Hence the matrix X with these eigenvectors as column vectors is
orthogonal, so that X-1 = XT  A = XDX-1 = XDXT.

Q = XTXDXTX. If we set XTx = y


Since XT = X-1 we get x = Xy. We know that xTX = (XTx)T = yT and XTx = y so that Q
becomes Q = yTDy = 1y12 + 2y22 + ……… + nyn2

Principal axes Theorem

40
Any quadratic form in n- variables is equivalent by means of an orthogonal
matrix P to a quadratic form

h(y)= are the eigenvalues.

Example
Find out what type of conic section the following quadratic form represents and transform it
to principal axes Q = 17x12 – 30x1x2 + 17x22 = 128.

Solution
17  15 
We have Q = xTAX. Where A =  
  15 17 
The characteristic polynomial is (17 - )2 – 152 = 0
 1 = 2, 2 = 32
 Q = 2y12 + 32y22 But Q = 128
y12 y22
 2y1 + 32y2 = 128 = 2  2 = 1
2 2
8 2
If we want to know the direction of the principal axes in the x1x2 – coordinates, we normalize
the eigenvectors from (A - I) x = 0 with  = 2 and  = 32.
 1  1 
 2  2
We get   and  
 1   1 
 2  2 
1 2 1
2  y1 
Hence x = Xy = 1   
 1 
 y2 
 2 2 

 x1 = y1/2 – y2/2
x2 = y1/2 + y2/2

Consider the matrix 3x2 + 4xy + 8xz + 4zy + 3z2 = 1. This has the matrix transformation
3 2 4  x
  
(x, y, z)  2 0 2  y  = 1
4 3   z 
 2
The eigenvalues are 1 = 8, 2 = -1, 3 = -1 and the eigenvectors are
 2 1  0 
     
v1 = 1 , v2    2 , v3    2 
 2 0  1 
     
Not orthogonal vectors because v1v2 = 0, v1v3 = 0, v2v3 = 4 hence not normalized they are not
mutually orthogonal and cannot provide an orthogonal matrix.

41
We may use C = (v1 v2 v3) and C-1 AC = D. However, we cannot get a canonical
transformation. We can however, construct a new set of mutually orthogonal eigenvectors
(u1, u2, u3) and (v1, v2, v3) such that Q = (u1 u2 u3) then QTAQ = D and we can have a
canonical transformation from XTAX = XTAD QTX = (QTX)T D(QTX) = yTDy

Exercise
1. Find an orthogonal matrix Q that diagonalizes the symmetric matrix
 5 1 1 
 
A 1 5 1 
 1 1 5 
 
2. Consider the conic section whose equation is q( x)  2 x 2  2 xy  2 y 2  9 . Find the
canonical transformation.
3. Diagonalize the matrix of quadratic form
Q( xyz )  3x 2  4 xy  8 xz  4 yz  3z 2

42

You might also like