-

UNIT 14 LINEAR EQUATIONS AND

EUCLIDEAN SPACES
Structure
14.1 Introduction
Objectives
14.2 System of Linear Equations
14.2.1 Canonical Systems
14.2.2 Gaussian Elimination Method
14.3 Euclidean Spaces
L
14.3.1 Subspaces
14.3.2 Linear Span
14.4 Linear Independence
14.4.1 Basis
14.4.2 Dimension of a Subspace
14.5 Inner Products
14.5.1 Orthonormal Basis
14.6 Summary
14.1 INTRODUCTION
You are familiar with systems of linear equations in two or three variables. Linear
systems of equations involving more than three variables are encountered in almost all
fields. In this unit, we first give a general procedure to solve any system of linear
equations. The method, in principle, can be used to determine all the solutions of a
system of linear equations.
Next, we discuss some important aspects of Euclidean Spaces. These are generalisations of
the very useful concepts of two or three dimensional vectors. Systems of linear equations
also play an important role in the study of Euclidean Spaces. The concepts introduced in
this unit will be used in subsequent units to understand clearly the structure of solutions of a
system of linear equations. In many situations, an understanding of this structure is more
important than the ability to obtain various solutiom.
t
Objectives
k
The main objectives of this unit are
*
to intrnduce the Gaussian Elimination Method to find solutions of any system
of linear equations,
I
*
to introduce some important aspects of Euclidean spaces which are essential, and
*
to understand the structure of solutions of a system of linear equations.
i 14.2 SYSTEM OF LINEAR EQUATIONS
I
You are familiar with an equation of the type
1
in which a, b and c are fixed real numbers. Such an equation is called a linear
1 equation in two variablesx and y . In this equation, a and b are referred to as the
1
coefficients of x and y, respectively. The number c is referred to ds the right hand side
i
constant. Any values of x andy for which ax + by equals c is called a solution of the
equation. To find a solution of this'equation, we can fix any values of x , say xo , and
then solve the resulting equation for the corresponding value of y , i.e.
1
y =
( c - ax0 ). Thus, we can obtain a solution of the given equation corresponding
to each choice of% . We also know that all these solutions of a linear equation in 2
variables can be represented as points ( x , y ) on a straight line in a plane.
You are also familiar with a linear equation in 3 variables. These equations are of the
t Y Pe
and the solutions of this equation can be represented by points ( x , y , z) on a plane in
a3- dimensional space.
You have also encountered with systems of linear equations in 2 or 3 variables. In
such systems, more than one linear equation is considered simultaneously. For
example, the following is a system of 2 linear equations in 3 variables :
The solutions of this systein are those values of ( x , y , z ) which satisfy both these
equations simultaneously. Thus, we can find the solutions of each equation separately
and then the common solutions, if any, are the solutions of the system. Since two
intersecting planes have a line in common, the solutions of a system of 2 linear
equations in 3 unknowns can be represented by a line in a 3- dimensional space. It may
be noted that a system of 2 linear equations in 3 variables does not have any solution if
the two planes, representing the system, do not i~ltersect.
The foregoing discussion shows that a single non-trivial linear equation in 2 or 3
unkowils always has an infinite number of solutions. However, when we consider
systems of such equatioils, then the system may fail to have any solution.
In general, a linear equation in n variables is of the type
where n is a fixed natural number and al, a2, . . . , an and b are fixed real numbers. In
this equation, n denotes the number of variables, ai is referred to as the coefficient of the
variables xi, and b is referred to a right hand side constant. Any set of n numbers
(XI , x2 , . . . , x,, ) for which a1 xl + a2 xz + . . . + a, x,, equals b is called a solution
of the equation. To find a solution, we can fix the values of any n - 1 variables and then
determine the value of the nth variable (not fixed so far) from the resulting equation.
0 0 0
Thus, if xl , xz , . . . , x,, - 1 are fixed at values XI, x2 , . . . , x,, -1 then
1 0 0
x,, = - ( I ) - a1 x? - a? xz - . . . - a,, -1 x,, -1 ), if a, P 0. Thus, we get a solutioil
an
of the given equation for each choice of w l , xz , . . . , xn - 1 . We observe that a single
non- trivial linear equation in rt variables, n 2 2, always has an infinite number of
solutions and that all the solutions can indeed be obtained. However, the geometrical
analogue seems to be lost for the moment.
In several situations, we are required to consider more than one linear equation
simnulta~leously. If m linear equations in n variables have to be considered
simultaneously, we say that we have a system of rn linear equations in n variables.
Such a systein can be written as follows :
al l xl + 012x2 + . . . + nl,, x,, = bl
The system can be compactly written as :
In the above system, aij is the coefficient of the variable xj in the i th equation of the
system and bi is the right hand side constant of the i th equation of the system. A set of
numbers rl , r2, . . . , r,, is a solution of the system if xi = ri , for i = 1 , 2 , , n,
satisfy each equation of the system seperately. We can identify the solutions of each
individual equation of the system. How can we identify the common solutions, which
are solutions of the system ? Recall that the system may have no solution at all. We now
proceed to describe one method which can be used to systematically identify the
solutions of the system, if these exist.
14.2.1 Canonical Systems
In some linear systems of equations, coefficients in some positions are zero such that
the system can be trivially solved. Such systems are called Canonical Systems. In this
section, we discuss two such canonical syste-and also the method to solve these. In
the next sedion, we shall show that any general linear system can be reduced to one of
the canonical systems discussed in this section.
We first consider a system of linear equations of the following form
A system of the above type can, easily, be solved as follows:
Substituting z = 3 in the first two equations, we get
Now substitutingy = 1 in the first equation, we get
Hence, the only solution of the system is x = 1, y = 1, z = 3.
I
The form of the system which can be solved by the above method, called back
1
substitution process, can be described as follows :
a) Order the variables as first variable, second variable, etc. In the example above, x is
the first variable, y is the second variable and z is the third variable.
b) The coefficient of the i th variable is not zero in the i th equation of the system and
is zero in all the subsequent equations ( i + 1 onwards) of the system.
A system of n linear equations in n variables is called a triangular system if the
variables and equations can be ordered so that condition b), given above, is satisfied.
To be more precise, a system of n linear equations in n variables
is triangular if the coefficients aij of the system satisfy the conditions : For i = 1 to n ,
aii rr 0 and aij = 0 for j = i + 1 to n . riangular system of the type considend above
Linear ~ u . b .nd
Eudidemn S p a
has a unique solution. This unique solution can be obtained by using the back
substitution meth6d.
As a second canpnical system, consider a system of the following type :
2 x + y + 4 z + t = 1 5
2 y + z + 4 t = 5
z + 2 t = 3 .
If we fix the value of t in the above systein, then the resulting system is a triangular
system. The example of triangular systein considered earlier corresponds to t = 0. Let
us fix t E 10 in the above system. If we use the back substituting method in the
resulting triangular system, verify that
x = 1 + 4 t o
y = l - t o
z = 3 - 210
t = 10
is a solution of the system. Thus the system has a unique solution, as given above,
correspoi~ding to each value of to . Clearly, the solution changes when to changes.
-
In the above example, we can retain the triangular character of the coefficient of x, y, z
and introduce more than one variable also with arbitrary coefficient. Such a systein can
also be solved by fixing the values of the additional variables and then solving the
resulting triangular system inx, 11, z.
To be more precise, consider a system of m linear equations in n variables, m < n ;
n
1 Ui j Xj = bi , i = 1 to m,
j -1
where the coefficient aij satisfy the conditioits : For i = 1 ta m , aii * 0 and ai j = 0
for j = i + 1 torn . In other words, the part of the system consisting of the variables
v XI, x2, . . . , x, is triangular. The coefficient ai j for i = 1 to m , and j = m + 1 to
ti of the non-triangular part of the systein are arbitrary. The system reduces to a
triaiigular system for every choice ofx, + 1 , x,'+ 2 , . . . , x, (non-triangular variables).
Thus all the solutioils of the systein can be obtained by considering all possible choices
of the non-triangular variables. Let us call such systems, which are not triangular, as
trapezoidal systems. In such a systein, we have atleast one non-triangular variable. We
know that arbitrary values can be assigned to the non-triangular variables.
Corresponding to each choice of the non-triangular variables, we obtain a distinctive
solution of the systein which is associated with the particular choice of the
non-triangular variables. Since there are an illfinite number of choices of the
non-triangular variables, every trapezoidal system has an infinite nuinber of solutioils.
14.2.2 Gaussian Elimination Method
Consider two systems of linear equations, say
System A :
ai j xj = bi , i = 1 to rn.
j -1
System B :
i
1
Note that the systemA and B involve the same number of variables. However, system A
consists of m equations and system B consists of r equations. We say that system A is
equivalent to systernB if every solution ofA is a solution of B and every solution of B
is a solution of A. Verify that the following two systems are equivalent :
System A System B
Suppose that we want to solve a given systemA. Does there exist a trapezoidal (or
triangular) system B which is equivalent to systemA ? Interestingly, the answer to-this
question is yes, provided the systemA has a solution.
In this section, we describe the Gaussian Elimination Method which identifies a
trapezoidal system equivalent to a given linear system. Since we know how to solve a
trapezoidal system, the Gaussian Elimination Method can be used to solve any linear
system. We first make the following intuitive obse,mations : The solutions of a system
do not change if
a)
any one equation of the system is multiplied by a non-zero number ; or
b)
any two equations of the system are interchanged ; or
c)
any equation of the system is added to any other equation of the system.
The reader is invited toprove the above statement. The Gaussian Elimination Method
uses the above operations, in a sequential manner, starting with the given system until a
trapezoidal equivalent system is obtained. We illustrate the method through examples
only.
Example 1 :
Let us consider the system
To obtain a trapezoidal system, we can eliminatex from the second and third
equations and the variable y from the third equation.
Adding the first two equations, we get an equivalent system
Multiplying the first equation by - 2 end then adding to the third equation gives an
equivalent system
x + y + z + t = 2 0
Adding the second equation to the third equ.ation, we get an equivalent system
This system is a trapezoidal system whose solution is given by
t = to,
where to can be assigned any value.
Example 2 :
Let us consider the system
2 x + y + z +, t = 15
Since the coefficient of x is zero in the second and third equations, x can be taken
as the first variable in the triangular part.
If we decidey as the second variable in the triangular part, then we must
interchange the second and third equations. Thus we get an equivalent system
2 x + y + z + t = 1 5
3 y + 5 z + 6 t = 10
2 2 + 3 t = 20.
The above system is an equivalent trapezoidal system and can be easily solved.
Example 3 :
Let us consider the system
3 x + 5 y + 2 z + 3;- 10
42 + S t = 15
Here, either x ory can be taken as the first variable of the triangular part. Let us
choose x as the first variable. Theny carmot be the second variable of the
triangular part
(because the wefficient of y in all the subsequent equations is zero). Thus y must
be in the non-triangular part only. Multiplying the second equation by - 3 and
adding to the third equation, we get an equivalent system
This is an equivalent trapezoidal system in whichx ory is the variable of the non-
triangular palt.
Consider the system
x + y + z = 4
- x + 2y + 3 2 - 5
6 y + 8 z = 20.
Adding the first equation to the second, we get
x + y + z = 4
3y + 4 2 = 9
6 y + 8 z = 20.
Multiplying the second equation by - 2 and adding to the third equation, we get
x + y + z = 4
3 y + 4 2 = 9
oy + oz = 2.
We'now observe that the third equation cannot be satisfied for any choice of X, JJ
and z. Hence, this system has no solution. Thus, the original system also has no
solution.
Consider the system
x + y + z = 4
- x + 2y + 3 z = 5
6 y + 8 z = 18.
Performing the sequence of operations as in Example 4, we get an equivalent sys-
x + y + z = 4
3 y + 4 2 = 9
This is a trapezoidal system and the solutioii is given by
1
x = l + - 2 0
3
4
y = 3 - - n ,
3
z = zo
where any value can be assigned to 20 .
Solve the following canonical syStem of linear equations:
2x+y- z+3t = 4
2y + 3z + 5t = 6
z-2t = 3
4t = 12
Linear Equations and
Euclidcnn Spec
Use the Gauss elimination method to solve the following systems of linear
equations:
a) x + 2 y + z - 1 9
E3
Show that the following system of linear equations has no solution:
2 x - y + z - 4
~ + 2 y - 3 2 - 8
4x + 3y - 5z - 10.
-
14.3 EUCLIDEAN SPACES
A linear equation in n variables involves a string of n + 1 numbers (al, a2, . . . , an , b ).
A system of m linear in n variables can be &fined by m such stdqp as shown
a11 a u ... sin bi
. .
. .
. .
A close look at the Gaussian elimination method shows that we are required to
(i)
multiply a string by a non-zero number
(ii) add two strings
(iii) interchange two strings.
We also observed that the geometrical analogue, available for n = 2 or 3, is lost when
n > 3 . We can indeed introduce some arithmetic in strings of arbitrary number of
components and retrieve the geometrical analogue also. We introduce the relevant ideas
in this section.
Let (al, az, . . . , a, ) be a string, called n-tuple, of n components, where al, a& . . . , a,
are real numbers. Let us consider the collection of such n-tuples. This collection is
denoted by
R n = {( a l , az, .. ., a, ), I al , a2, .. ., an are real numbers},
i.e., R " is the set of all n-tuples of real numbers. Two n-tuples
A = ( a l , az, ..., a,) and B = ( bi , h, ..., bn)
are said to be equal, denoted by A = B, if
ai = bi for i = 1 to n.
Thus any two n-tuples are equal if their corresponding components are equal.
The sum ofA and B, denoted byA + B , is defined as
A + B - (a1 + bl , a z . + h , ..., an + b,),
i.e., the sum of any two n-tuples can be obtained by adding the corresponding
components of the given n-tuples. The following properties of the sum of n-tuples, as
defined above, can be proved easily :
L
(i)
I f A Rn and B Rn, t henA+B ER"
(ii) A + B = (a1 + bl , a2 + h, . .., a, + b,),
B + A - (bi + a l , h + a 2 , ..., bn+an).
:. B + A - A + B
(iii) Let C = ( CL , c2, . . . , C, ), Then
(A + B) + C - ((a1 + bi ) + CL, (a2 + h ) + c 2 , ..., (an + bn) + 4)
a (a1 + ( b ~ + CL) , a2 + ( h ~ ) , . - . , a n + ( b ~ + cn)
- A + ( B + C )
)
(iv) If we denote the n-tuple ( 0,0, ... , 0 ) by 0, then
A+O=O +A=A.
- A + A = A + ( - A ) = 0 ,
where -A - = ( - al , -a2,. . . ,-a,).
Let A = ( al, ;a2, . . . , a, ) be any n-tuple and a be any real number. Then, we
define the scalar multiple a A by
a A = ( a a l , a a t , ..., a a , ) ,
i.e., a A is obtained by multiplying every component of A by a.
This operation is called multiplication of a n-tuple by a scalar. The following
properties of scalar multiplication are also easy to observe.
(vi) I f A E R" and a E R then a A E Rn.
(vii) a ( A + B ) = a ( a l + b l , az +bt , ..., an+bn)
= ( a a l + a b l , a a t + a h , ..., aa, + a b , )
= a A + a B .
(vi i i )(a + P ) A = ( ( a + B ) a l , ( a + P) a z , ..., ( a + P) a n )
= ( a a l + p a l , a a z + Baz , ..., a a n + Pan)
= ( a a l , a a z , . . . , a a , ) + ( P a l , paz , . . . , pan>
= a A + PA.
(i x) ( a $ ) A = ( ( a P ) a l , ( a $ ) a 2 , ..., ( a f J ) a n )
= ( a ( f l a ~ ) , a ( B a t ) , ..., a ( B a n ) )
= a ( P a 1 , Pa*, . . . , b a n >
= a ( P A ) .
(XI
1 A = la^, l a 2 , ..., l a n )
= ( a ~ ; az , ..., a, )
The set R n together with the operation of addition and scalar multiplication, as defined
above, is called a Euclidean Space. For n = 2 or 3 , the operations that we have
defined above coincide with the corresponding operations of vector in 2 or 3
dimensional space. In view of this, the n-tuples ( a1 , a2 , . . . , a, ) are also called
vectors. Although, we are restricting our attention to n-tuples of real numbers, the
generalization to n-tuples of complex numbe~s is almost identical.
E4
Use the properties of addition and scalar multiplication to prove the following :
a) a A - 0 - a = O o r A = O
b) ( - l ) A = - A
C) A - ( B - C ) = A - B + C .
14.3.1 Subspaces
Let us consider an arbitrary subset S of R " . Some subset have the following
properties :
(a) A ES- a A E S forall a E R
(b) A E S a n d B E S = s A + B E S
Such subsets are called subspaces.
Example 6 :
Let
Si = { ( x , Y , 0 ) 1 x E R, y E R }
be a subset of R 3, i.e., S1 consists of all points in the ( x , y ) - plane. Then if
A = ( a l , a t , 0 ) and B = ( b l , 62, O)areinSl,wehave
a A = (aa!, a a z , 0 ) E S1 for all a E R and
A + B = ( a1 + bl , a2 + b2, 0 ) E S l .
Hence the set of all points in the ( x , y ) - plane is a subspace of R 3.
Example 7 :
Let
s z = { ( x , y , 2 ) 1 x + 2y + 32 = 01,
i.e., Sz is the set of all points lying in the plane x + 2y + 3 z = 0, which passes
through the origin.
If A = ( a l , a2, a3) and B = ( b l , bz , b3) arein S2,thenwemusthave
a1 + 2a2 + 3a3 = 0 , and bl + 2b2 + 3b3 = 0 .
Adding these equations, we get
( a1 + b l ) + 2( a2 + b2) + 3( a3 + b3) = 0 ,
i.e., A + B - ( a1 + bl , a2 + 62, a3 + b3 ) E Sz . Similarly, verify that
a A E S2 for all a E R . Hence Sz , which is a plane passing through the
origin, is also a subspace of R 3.
Exarnple 8 :
Let
S3= { ( x , y , 2 ) ( x + 2y + 3 2 = 4 ) .
I=
Note that S3 is also a plane, but it does not contain the origin. Hence S3 cannot be
sub space of R 3. Give reasons.
The above example illustrate that some subsets of R " are subspaces of Rn whereas
some other subsets are not subspaces of R ".
Which of the following subsets of R " are subspaces :
(a) { ( XI , ~ 2 , ..., xn) I XI = ~ 3 }
(b) { ( XI , x2, ..., xn) 1 x2 = 11
(c) ( ( XI , 12, ..., x,,) I XI + x i = 0 and XI = x4)
(d) ((XI. , x2, ..., x,,) 1 + d = 1, .
14.3.2 Linear Span
. Let S - { A ~ , A2, ...., Ap} beanysubsetof~~andchooseany
a1 , a2, . . . , a, in R . Then, using the operations of addition and scalar
multiplication, we construct a vector
in R ". For a fixed choice of a1 , a 2 , . . . , a,, the vector A , as defined above, is
called a linear combination of the vectorsA1 , A2 , . . . , A, . Let us consider
i.e., L ( S ) is the set of all possible linear combinations of the vectors in S . We call
L,( S ) as the linear a p a i o f ~ . We now observe that L ( S ) is a subspa& of R " for
every non-empty subset S . Let
and
B = PlAi + $2A2 + ... + PpAp E L( S )
Then clearly
y A- y a i Al + y a 2 Az + . . . + y a p Ap EL( S ) f o r a l l y ER
and
A+ B = ( al +$l ) Al +( az+$2) A2+ ... t ( ap+Pp) Ap E L( S )
Hence, L ( S ) is,a subspace of R " .
Example 9 :
Let
s = { ( I , l , O ) , ( 0 , 2 , o ) } .
Hence,L { ( I , 1, 0 ) , ( 0 , 2, 0)}isthexy-planein R ~ .
E6
Prove the following :
(a) SI G S2 L( S1) C L( S2)
(b) Si sasubspacee L ( S ) I S .
If P and Q are subspaces of Rn, prove that
P + Q = { A + B ( A E P and B E Q }
is also a subspace of R ".
I f P = L { ( O , 1 , 1 , I ) , ( 0 , 1 , 0 , 1 ) ) a n d '
Q = L { ( 1 , 0 , 0 , O) , ( 0 , 0 , 1 , 0))leterinineP + Q
14.4 LINEAR INDEPENDENCE
Let S = {AI , A2, . . . , Ap ] be any subset of Rn. Consider the equation
Obviously, a1 = a 2 = : . . - - a, = 0 satisfies this equation irrespective of our choice
ofA1, A2, . . . , A, . In case eachAiis non-zero and a1 = a 2 = . . . = a, = 0 is the
only solution of the above equation, then the set S of non-zero vectors
A1 , A2, . . . , A, is said to be linearly independent. Otherwise, S is called lineally
dependent. Thus the vectorsAl , A2, . . . , Ap are linearly dependent if atleast one of
the a's is not zero and a1 A1 + a 2 A2 + . . . + ap Ap is the zero vector. It may be
noted that any subset of Rn is either linearly independent or linearly dependent.
Example 10 :
LetuscheckwhetherthesetS = { ( I , 0 , 0 ) , ( 1 , 1 , O) , ( 1 , 1 , l ) } i s
linearly independent or linearly dependent. Consider the equation
a ( l , 0 , 0 ) + P ( 1 , 1 , 0 ) + ~ ( 1 , 1 , 1 ) = ( 0 , 0 , O) ,
~.e.,
( a + P + * / , P * Y , Y ) = ( O, O, 0 )
The equality of two vectors requires :
a + p + y = O
p + y = o
y = 0
This is a triangular system with the unique solutioil a = P = y = 0. Hence S is a
linearly independent subset of R 3.
Example 11 :
Let
s = { ( I , 0 , - 1 1 , ( 1 , 1 , I ) , ( 2 , 1 , 011.
Then
a ( l , O , - l ) + P ( l , 1 , l ) + y ( 2 , 1 , 0 ) = ( 0 , 0 , 0 )
Linear Equations m d
Euclidcm Space
gives
a + p + 2 y = O
Check that a = 1 , 8 = 1 , y = - 1 is one of the solutions of the above system.
Hence, { ( 1 , 0 , - 1 ) , ( 1 , 1 , 1 ) , ( 2 , 1 , 0 ) 1 is linearly dependent.
We now make some useful observations about linearly independenttdependent sets.
1. The zero vector cannot belong to a linearly independent set.
2. If S is a linearly independent set, then every subset of S is also linearly
independent.
3. If S is a linearly dependent set, then every set containing S is also liilearly
dependent.
These observations follow immediately from the definition itself. Next, we prove two
important results.
Theorem 1 :
If { A1 , A2 , . . . , Ap ) is linearly independent and {A1 , A2 , . . . , Ap , Ap + 1 )
is linearly dependent, thenAp+ 1 is a linear combination ofAl , A2, . . . , Ap .
Proof:
Since { A1 , A2 , . . . , Ap , Ap + 1 ) is linearly dependent, we can find scalars
81, 82, . .. , 8 p + l ~ ~ ~ h t h a t
where atleast one of the 8' s is not zero. We observe that pp + i z 0,
because if $, + 1 = 0, then we must have
where atleast one of the Pi, 1 5 i s p , is non-zero. This means that
{A1 , A2 , . . . , Ap ) is linearly dependent, which is contrary to what is given.
Hence, 8, + 1 z 0. Thus, we have
i.e., Ap+ 1 is a linear combination of A1 , A2, . . . , Ap .
Theorem 2 :
If {A1 , A?, . . . , A, ) is linearly independent and A L { A1 , A2, . . . , A, ),
then { A1 , A2, . . . , A, , A ) is liilearly independent.
Proof:
If possible, let {A1 , A2, . . . , Ap , A ) be linearly dependent. Then by theorem
l , A can be expressed as a linear combination of Al , A2, . . . , Ap . This is not
possible
becauseA @ L {A1 , A2, . . . , A, ). Hence { A1 , A2, . . . , A, A ) must be
linearly independent.
E 8
Which of the following sets are linearly independent ?
(a) { ( I , 1 , 01, ( 1 , 2 , 0 1 7 ( 2 , 1 , 0 1 1
(b) [ ( l , 1 , I ) , ( 2 9 1 , I ) , ( 1 , 2 , 2 ) ]
(c) [ ( 2 , 2 , 21, ( 3 , 1 , 1 1 , ( 1 , 3 , 3 1 )
(d) [ ( L - 1 , 19- 11, ( - 1 , - 1 7 - 1 9 1 1 , ( - 1 , 1 , 1 , I ) ]
(el [ ( I , 0 , 01, ( 0 , 3 , 0 1 7 ( 1 , 1 , 1 1 , ( 19 2 , 311
E9
I f {A, B , C)islinearindependent,provethat{A + B , B + C, C + A) i s
also linearly independent.
Linear tio om M d
Euclidean Space
14.4.1 Basis
Let S be a subspace of R ". A subset T of S is called a basis of S if
(a)
T is linearly independent, and
(b) L ( T ) = S.
For example, let
S = I ( x , y , O ) I ~ E R , Y E R } .
Example9showsthat S - ~ { ( l , 1 , O) , ( 0 , 2 , o) }.
Verify that [ ( 1 , 1 , 0 ) , ( 0 , 2 , 0 ) } is linearly independent also. Hence,
{ ( I , 1 , O) , ( 0 , 2 , 0 ) ) isabasisof S.
We invite the reader to show that the following set is a basis of Rn
T = { EI , Ez, . . . , En] where for i = 1 to n
Ei = ( 0 , 0 , ..., 0 , 1 , 0 , ..., 0 )
is a vector whose i th component is 1 and all the other components are zero. The set T ,
given above, is called the standard basis of R ".
We now observe that theorem 2 can be used to generate a basis set of any subspace S
of Rn. Choose any A1 # 0 in S . If L ( {A1 ) ) = S, then {A1 ) is a basis set of S .
k
Otherwise, whenL ( {A1 ) ) # S , we can choose A2 E S such that
A2 4 L ( {A1 ) ). Then Theorem 2 shows that {A1, A2 ) is linearly independent.
Clearly,this iterative process of enlarging a linearly independent subset Tof S can be
repeated till we obtain a linearly independent subset T of S satisfying L ( T ) = S . Note
that we can also generate a basis of S which contains any given linearly independent
subset of S . This is so because the iterative process can be initiated with the given
linearly independent subset. Thus any linearly independent subset of R " can be
extended to a basis of R ".
If { A1 , A2 , . . . , A, ) is a basis of subspace S of R ", then it is easy to see that any
X E S can be expressed as a unique linear combination of the basis, that is, there exist
unique scalars a1 , a 2 , . . . , a, E R such thatX = a1 A1 + u2 A2 + . . . + a, Ap .
Thus ( a1 , a 2 , . . . , a, ) are called coordinates of X with respect to the basis
{ Al , A2, -.., Ap).
Show that { (
{ ( x , Y , 2 )
basis of the subspace
14.4.2 Dimension of a Subspace
We know that every subspace S of Rn has several basis sets. Is it possible that two
different bases of the same subspace contain different numbers of vectors ? The
following theorem, whose proof is not in the course, shows that this is not possible.
Theorem 3 :
Any two bases of a subspace of R " contain the same number of vectors.
The above theorem shows that with every subspace of Rn we can associate a unique
number, namely, the number of vectors in any basis of the subspace. This unique
number is called the dimension of the subspace. Thus, to find the dimension of a
subspace, we can determine any one basis of the subspace and count the number of
vectors in that basis.
Since we know that
is one of the basis of Rn, we can now say that the dimension of R " is n.
We invite the reader to prove the following obvious results :
1. If S is a p dimensional subspace of R ", then every set of p linearly
independent vectors in S is a basis of S .
2. Every set of p + 1 or more vectors in a p dimensional subspace of Rn is
linearly dependent.
3. The dimension of any subspace of Rn cannot exceed n .
Example 10 :
Let us compute the dimension of the subspace
Clearly the vector ( 1 , - 1 ,' 0 , 0 ) is i ns . Also
= { ( a , - a , 0 , 0 ) I a C R) .
Sincethevector(0, 1 , - 1 , 0 ) i s i n ~ butnotin ~ ( ( 1 , - 1 , 0 , 0) }, we
conclude, using theorem 2, that { ( 1 , - 1 , 0 , 0 ) , ( 0 , 1 , - 1 , 0 ) ) is a
linearly independent subset of S . Now
L { ( l , - 1 , 0 , 0 ) , ( 0 , 1 , - l , O) } = { a ( l , - l , O, O)
+ j 3 ( 0 , 1 , - 1 , 0 ) l a ~ , f 3 ~ }
~i ncet hevect or ( 0, 0 , 1, - 1) i si nSbut not i n L { ( 1 , - I , 0 , o ) ,
( 0 , 1, -1, o.)), weconclydethat{(l, -1, 0 , 01, ( 0 , 1, -1, o ) ,
( 0 , 0 , 1 , - 1 ) ) is a linearly independent subset of S . Since the vector
( 1 , 1 , 1 , 1 ) 4 S , the dimension of S cannot equal the dimension of R ' , i.e.,
4 . ~ e n c e { ( l , - 1 , 0 , o ) , ( 0 , 1, -1, 0 ) , ( 0 , 0 , 1 , - 1 ) ) i s a b a s i s o f ~
and, therefore, the dimension of S is 3.
The example that we have given so far point towards the fact that a study of subspaces
of Rn is intimately CO~eCted witiithe solutions of a system of linear equations in n
variables. The dimension of the subspace is concerned with the number of independent
equations in the system. We shall unfold this relationship between subspaces of R" and
c
the solutions of systems of linear equations in n variables in the subsequent sections.
+
Showt hat ( ( 1, 0 , O), ( 1 , 1, O), ( 1 , 1, l ) ) i s a ba s i s o f ~ ~ . ~ ho wt ha t
{ ( 4 , 5 , O) , ( 1 , 1 , o ) , ( 1 , 1 , 1 ) ) and ( 1 , 0 , 0 ) , ( 4 , 5, O),
( 1 , 1, 1 ) ) a r e a l s o b a s i s o f ~ ~ . ~ s { ( l , 0, o ) , ( 1 , 1, O), ( 4 , 5 , 0 ) ) a
basis of R ?
E 12
Determine bases Sl and S2 of R satisfying the following conditions :
(a) Sl n S2 - cp
(b) ( ( l , O , O , 0 ) , ( 1 , i , O , o ) J G S l
(c) { ~ 1 , 1 , 1 , 0 ~ , ~ 1 , 1 , 1 , 1 ~ ) G s 2 - .
E 13
.i
If A, B, and CinRnsatisfy A + B + C = 0, showthat
L ( { A, B ) ) = L ( { B , c ) ) = L ( { A , c ) ) .
Liur Equations md
Euclidean Space
Show that { ( XI , x2, x3 , x4 ) I xi 5 x2 and x3 = .u } is a subspace of R
Determine the dimension of this subspace.
14.5 INNER PRODVCTS
You know that the geometrical notions oflength, perpendicularity and angle (inR
or R ) can be studied through the concept of dot (or scalar) product of vectors. We
introduce these concepts inRn in this section.
LetA = ( a l , a2, ..., an) andB = ( b l , b2, ..., b,)beanytwovectorsinRn.
We associate a real number (A, B ) called inner product ofA and B as follows :
Note that the inner product defined ahove corresponds to the notion of dot product,
when n = 2 or 3. We observe the following properties of this inner product :
(i) ( ~ , ~ ) = a : + a $ + ... + a : z O f o r a l l AE R n
2
(ii) ( A , A ) P O e a: +a$ + ... + a , = 0
(iii) ( A, B ) = al bl + a2b2 + ... + a, b,
(iv) ( a ~ , B ) = a a l b l + a a z b z + ... + a a , b,
(v) Let
C = ( c l , c2, . . . , G) Then
( A + C, B) = ( a l +c l ) bi + (az+cz)b;! + ... + ( an+cn) bn
= ( A, B) + ( C, B).
We could, in fact, associate a number (A , B ) with vectors A and B of Rn in many
other ways such that the properties (i) to (v) listed above are valid. Each such
association can be used to define an inner product in R ". However, we shall confine our
attention to the inner product as defined above. This particular inner product is called
the standard inner product of R ".
ForA E R ", the length ofA , to be denoted by IIA 11 , is defined by IIA 11 = m.
The length I(A I( ofA is also called the norm ofA . We now prove some properties of
the length function.
(vi) I ( aA + pB( I 2 = ( a A + PB, a A + P B)
- ( a A , a A + P B) + ( P B, a A + P B)
[use (v)l
1
- a [ a ( A, A) + fJ(B, A)] + f J [ a ( A, B) + P( B, B) [use (iv)]
I
! = a 2 ( ~ , ~ ) + ~ ~ P ( A , B ) ] + P ~ ( B , B ) [use (iii)]
- a211~1l2 + ~ ~ P ( A , B ) + f J 2 ( 1 ~ 1 1 2 .
(vii) Substituting P = 0 in (vi), we get
I l a ~l l ~ - a2ll~Il2 * IIaAIl = lal IIAII.
(viii) Using (i) and (ii), we. observe that
IlAII > 0 i f A i. 0 and )lo(( = 0 .
1
(ix) Substituting a = 1 and 6 = - - 1 ,, , ( A , B ) in (vi) and using (viii), we
i. e. , (A, B)2 s 11AI12 IlB112
HenceI(A, B)I s ((Al ( I(BI1.
This inequality is called the Schwan Inequality.
(x) Substituting a = fJ = 1 in (vi), we get
[use (ix)]
Hence
llA + B I I s llA II + IIB II
This inequality is called the triangle Inequality.
The Schwarz inequality shows that
Thus, we can define the angle 0 between the vactorsA and B as follows:
cos 0 = ( A, B)
l l A I I I1 B I I
The above definition of angle coincides with the familiar notion of angle in R or R 3.
Now we have generalised the notion of angle in R ". We can now say that vectors A and
B are perpendicular if (A , B ) = 0.
14.5.1 Orthogonal Basis
A subset of S of R " is said to be orthogonal if for allA and B in S , we have
( A, B) = 0 ,
wheneverA # B . An orthogonal set S is said to be orthonormal if, in addition, we have
11 A 11 = 1 for every A E S .
Theorem 4 :
Every orthogonal set of non-zero vectors is linearly independent.
Proof:
Let S - { A1 , Az , . . . , A, ) be any orthogonal subset of R ".
Then a l A l + a z A 2 + . . . + a p A p = O
* ( ~ I A I + a2Az + ... + a p Ap , Ai ) = ( 0 , ~ ; ) = 0
* a; ( A; , A; ) = 0 , because, ( A, , A; ) = 0 for j # i
=>a; = 0 , because A; 0 implies ( A; , A; ) * 0.
Thus a1 5 a 2 = . . . = a, = 0 . Hence S is linearly independent.
I
Theorem 5 :
If {A1, Az , . . . , Ap ) is any orthonormal set in R " and X E R ", then
P
Proof:
Choose any Aj , j = 1 to p. Then
Ix- C. ( X, Ai ) Ai , Aj I = ( x, A,) - 2 ( X , A; ) (A; , A ~ )
/
= ( X, Aj ) - ( x , ~ j ) because(A;,Aj) = O i # j
and(A;,A;) = ( ( ~ ~ ( 1 ' = 1
= 0.
We now ask the impottant question :
Is it necessary that every subspace of R" has an orthonormal basis ? The following
theorem shows the answer is 'Yes'.
Theorem 6 :
I
Every non-zero subspace of R" has an orthonormal basis.
Proof :
Let S be any subspace of Rn. Choose any arbitrary basis of S , say
{ Al , Az , ..., Ap).
We shall now describe a procedure which gives us an orthonormal subset
{ B1 , BZ , . . . , Bp ) of S such that Bj is a linear combination of Al , A2, . . . , A;
Clearly Aj * 0 for any j. Let
so that { Bi ) is an orthonormal set. Using theorem 4,
cz = A Z - ( AZ, BI ) BI ,
is orthogonal to B1 . Let
Then { B1, B2 ) is an orthonormal set. Assuming that an orthonormal subset
{ B1 , BZ , . . . , B, ) of S is available, we shall, now, show how we can construct
an orthonormal subset { B1, Bz , . . . , B, + 1 ) of S , provided r <p. Using
Theorem 4,
is orthogonal to each B; for i = 1 to r . . Setting
i
I
we get { B1, Bz , . . . , B, + 1 ) as an orthonormal subset of S . The set
{ B1 , BZ , . . . , Bp ), wnstucted as above, gives an orthoilormal basis of S .
t
i
The process outlined in theorem 6 to construct an orthonormal basis of S , starting with
i an arbitrary basis, is called Gram Schmidt Orthogonalisation Process.
I'
Example 13 :
We illustrate the Gram Schmidt process with the following starting basis :
Hence, the orthonormal basis constructed by using the Gram Schmidt process is
{ Bl , Bz, B3) = ( ( 1 , 0 , 01, (0, 1 , o ) , ( 0 , 0 , 1))whichisstandard
basis of R 3.
It may be noted that if we change the order of the vectors in the starting basis, then the
Gram Schmidt process gives a different orthonormal basis. Try the above example with
thestartingbasisas{( 1, 1 , o ) , ( 1 , 0 , 01, ( 1 , 1 , 1 ) ) .
Let {A1, Az, . . . , Ap)beany basisofasubspaceSofRn.Ifthebasis
{A1 , Az , . . . , A, ) is known to be orthononnal and
then, for any i = 1 to p , we have
b e c a ~ s e ( ~ ~ , A;) = 0 when j z i and (A; , A;) = 1.
E 15.
Determine an orthogonal basis of R containing ( 1 , - 2, 1
E 16.
Use Gram Schmidt process on the basis { ( 0 , 0 , 1 ) ,
to obtain an orthonormal basis of R 3.
Determine an orthonopinal basis of the subspace L {( 1,
(0, 2, 1, 0), (1, 0, 8; 0)). ,
E 18.
If IjA 11 = ( ( B (1 , show that A + B and A - B are orthogonal.
Prove the following
(a) llA + B1I2 + llA - B1I2 = 211A1I2 + 211B1I2
(b) (-4, B ) = 0 * 11-4 + B( I 2 = I I A ~ ~ ~ + I I B ~ ~ ~
f4.6 SUMMARY
We briefly sum up what has been done in this unit.
1. Canonical systems are linear systems of equations in which coefficients in some
positions are zero and the system can be trivially solved by
Linear Equations and
Euclidean Space
(a) ordering the variable and using back substitution process or b) Gauss
Elimination Method in which solutions remain unchanged, i) multiplication of an
equation by a non-zero number ii) interchange of any G o equtionsiii)addition of
two equations.
2.
The set R " = { ( a 1, a 2 ,..., a, ) 1 al, a2, ..., a, E R together with the
operations of additions and multiplications, is called a Euclidean space.
3-
Subsets S E R " having properties, a)A E S * a A E S V a E R
(b) A E S and B E S => A + B E S, are called subspaces
4. The set L ( S ), of all possible linear coinbinations of vectors ins, is called Linear
span of S.
5-
The vectorsAl,A2, ... , Ap of SRn are linearly dependent if
a l A1 + a 2 A2 + ...+ apAp = Oandatleastoneofthea si snot zero;
otherwise the vector are linearly independent.
6) (a) A subset T of S, a subspace of R", is called a basis of S if i) T b linearly
independent ii) L (T ) = S.
(b) Any linearly independent subset of R" can be extended to a basis of Rn by
iterative process.
(c) The number of vectors in any basis of subspace of R" is called the dimension
' ofS.
7: * a) ForA = ( al, a2 ,..., a,) E R" and B = ( bl, 62 ,..., 6, ) E Rn, the inner product
a ( A, B) i sdef i nedas( A, B) = a l b l + a z 6 2 + ... +a,b,.
:; (b) The length of norm ofA is I(A I( = f i .
(c) 1 (A, B ) ( s IJA 11 . 11 B (1 (Schwarz Inequality)
(d) ( 1 A + B ( 1 s ((A ( 1 + ) I B I (
(Triangle of Inequlity).
(e) Angle 8 between vector A aild B is given by cos 8 =
(A, B
IIA I1 IIB 11.
8.
(a) A subset S of Rn is said to be orthogonal if, VAJ S,(AJ ) = 0 .
In addition, if ((A(( = 1 V A E S t he set S is called orthonormal .
(b) Every non-zero subspace of Rn has an orthonormal basis.
(c) Gram Schmidt process is outlined in 14 S.1, to construct orthonormal basis.
14.7 SOLUTIONSIANS WERS
E l . (a) t = 3, z - 9, y =- 18, x = 11.
(b) t = t o, z = 6 + t o , y = - 8 + t o , x = 39
E2. (a) x = - 4 , y = 9, z = 5.
(b) x = 4 - 2 t 0 , y = 0 , z = 2+t o, t = to.
E 5. Subspaces are given by (a) and (c).
E7. P + Q = L {( 1, 07 0, O), (0, 0, 1, O), (0, 190, 1)).
E 8. Linearly independent sets are given by (d). Others are linearly dependent.
E14. Dimension is 2.
E15- {( l i - 2, l , 3) , ( 2, 1, - 3, 1) , ( l , l , l , O) , (-191, Oil)).