Unit 10 Eigenvalctes and Eigenvectors: Structure
Unit 10 Eigenvalctes and Eigenvectors: Structure
EIGENVECTORS
Structure
10.1 Introduction
Objectives
10.2 The Algebraic Eigenvalue Problem
10,3 Obtaining Eigenvalues and Eigenvectors
Characteristic Polynomial
Eigenvaiues of Linear Transformation
10.4 Diagonalisation
10.5 Summary
10.6 Solutions/Answers
10.1 INTRODUCTION
In Unit 7 you have studied about the matrix of a linear transformation. You have had
several opportunities, in earlier units, to observe that the matrix of a linear
transformation depends on the choice of the bases of the concerned vector spaces.
Let V be an n-dimensional vector space over F, and let T : V -,V be a linear
transformation. In this unit we will consider the problem of finding a suitable basis B,
of thevector space V, such that the n x n matrix[T], is a diagonal matrix. This problem
can also be seen as: given an n x n matrix A, find a suitable n x n non-singular matrix P
such that P-' AP is a diagonal matrix (see Unit 7, Cor. to Theorem 10). It is in this
context that the study of eigenvalues and eigenvectors plays a central role. This will be
seen in Section 10.4.
The eigenvalue problem involves the evaluation of all the eigenvalues and eigenvectors
of a linear transformation or a matrix. The solution of this problem has basic
applications in almost all branches of the sciences, technology and the social sciences,
besides its fundamental role in various branches of pure and applied mathematics. The
emergence of computers and the availability of modern computing facilities has further
strengthened this study, since they can handle very large systems of equations.
In Section 10.2 we define eigenvalues and eigenvectors. We go on to discuss a method
of obtaining them, in Section 10.3. In this section we will also define the characteristic
polynomial, of which you will study more in the next unit.
Objectives
After studying this unit, you should be able to
obtain the characteristic polynomial of a linear transformation or a matrix;
obtain the eigenvalues, eigenvectors and eigenspaces of a linear transformation or a
matrix;
obtain a basis of a vector space V with respect to which the matrix of a linear
transformation T : V -t V is in diagonal form;
obtain a non-singular matrix P which diagonalises a given diagonalisable matrix A.
The fundamental algebraic eigenvalue problem deals with the determination of all the
us look at some examples of how we can find
,
Warning: The z r o vector can never be an eigenvector. But, 0 E F can be an eigenvalue.
For example, 0 is an eigenvalue of the linear operator in E 1, a corresponding
eigenvector being (0,l).
@x € Ker (T-).I)
.'., W, = Ker (T - &I), and hence, W, is a subspace of V (ref. Unit 5. Theorem 4).
Since h is an eigenvalue of T, it has an eigenvector, which must be non-zero. Thus, W,
is non-zero.
Definition: For an eigenvalue h of T, the non-zero subspace W,. is called the ei,genspace
of'? associated with the eigenvaloe h.
Let us look at an example.
Example 3: Obtain W, for the linear operator given in Example 1.
Solution: W, = {(x,y,z) E R3 1 T(x,y,z) = 2(x,y ,z))
= {(x,y.z) E R" (2x,2y,2z) = 2(x, y ,z))
= R'.
rnlues and Elgenvectom Now, try the following exercise.
E E2) For T in Example 2, obtain the complex vector spaces W,, W-, and W,.
0
0 0 1
- - -
is the standard ordered basis of V,(F). That is, the matrix of A, regarded as a h e a r
transformation from Vn(F) to V,(F), with respect to the standard basis B,, is A ~tself.
This is why we denote the linear transformation A by A itself.
Looking at matrices as linear transformations in the above manner will help you in the
understand~ngof eigenvalues and eigenvectors for matrices.
Definition: A scalar i..is an eigenvalue of an n x n matrix A over F if there exists X € Vn(F),
X # 0, such that AX = AX.
If A is an eigenvalue of A , then all the non-zero vectors in Vn(F) which are solutions of
the matrix equatlon AX = i,X are eigenvectors of the matrix A corresponding to the
eigenvalue A.
Example4: Let A
eigenvector of A.
w u b n : NOW A
[[ k a] : I:[ ;1.
= 0 2 0
=
. Obtain an eigenvalue and a corresponding
1 1
T h ~shows
s that 1is an
eigenvectors [[ [ ]
]and ,respectively. I l e eigenvalues oE diag (dl ,......, dn
are d l ,........,dn
= [:I # I:]
Solution: Suppose A E R is an eigenvalue of A. Then
i
For examp1e;the eigenspace W,, in the situation of Example 4, is
-
Elpnvducs and Elgenncton E E4) Find W, for the matrix E3. .
The algebraic eigenvalue problem for matrices is to determine all the eigenvalues and
eigenvectors of a given matrix. In fact, the eigenvalues and eigenvectors of an n x n
matrix A are precisely the eigenvalues and eigenvectors of A regarded as a linear
transformation from Vn(F)to Vn(F).
We end this section with the following iemark:
-
A s a h A is an eigenvalue of the matrix A if and only U (A AI) X = O has a non-zero
-
solution, i.e., If and only if det (A A I) = 0 (by Unit 9, Theorem 5).
Similarly, A is an eigenvalue of the linear transformation T if and only if det (T - AI) = 0
(ref. Section 9.4).
So far we have been obtaining eigenvalues by observation, or by some calculations that
may not give us all the eigenvalues of a given matrix or linear transformation. The
remark above suggests where to look for all the eigenvalues In the next section we
determine eigenvalues and eigenvectors explicitly.
la:
Once we know that A is an eigenvalue of a matrix A, the eigenvectors can easily be
-':1
obtained by finding non-zero solutions of the system of equations given by AX = A X.
all a12 ... atn
This homogeneous system of linear equations has a non-trivial solution if arid only if
the determinant of the coefficient matrix is equal to O (by Unit 9, Theorem 5). Thus, h
I is an eigenvalue of A if and only if
.....
a a ...... t-a,,
where the coefficients c;, c,, ......c, depend on the entries aijof the matrix A.
The equation fA(t)= 0 is the characteristic equation of A.
When no confusion arises, we shall simply write f(t) in place of fA(t).
Consider the following example.
= (t-l)(t+l) = t2- 1.
Now try this exercise.
[P0 I:
E E5) Obtain the characteristic polynomial of the matrix
:1 - 2 .
+
Eigenvdua snd E ~ ~ W V ~ C ( O R Note that h is an eigenvalue of A if?det(hI - A) = fA(h)= 0, that is, iff h is a root of the
. characteristic polynomial f,(t), defined above. Due to this fact, eigenvalues are also
called characteristic roots, and eigenvectors are called characteristic vectors.
The roots of the characteristic For example, the eigenvaluesof the matrix in Example 6 are the roots of the polynomial
pdynomial of a matrix A form the
set ofeigenvalues of A.
tz-1, namely, 1 and (- 1).
Now, the characteristic polynomial fA(t)is a polynomial of degree n. Hence, it can have
n roots, at the most. Thus, an n x n matrix has n eigenvalues, at the most. .
For example, the matrix in Example 6 has two eigenvalues, 1and-1, and the matrix in
E5 has 3 eigenvalues.
Now we will prove a theorem that will help us in Section 10.4.
Solution: In solving E6 you found that the eigenvaluesof A are h, = 1,h, = -1, A, = -2
Now we obtain the eigenvectors of A.
The eigenvectors of A with respect to the eigenvalue h, = 1are the non-trivial solutions
of
which gives
0 1 -2
which gives
Thus, in this example, the eigenspaces W,, W-, and W-2 are 1- dimensional spaces,
generated over R by
1
J Obtain its eigenvalues and eigenvectors.
which gives x, + x, = 0
-xl-x2=0
-2x, -2x, + 2x, + x4 = 0
X1 + X2-X3 = 0
The first and last equations give x3 = 0. Then, the third equation gives x4 = 0. The first
'equation gives x, = -x,.
which gives x, + x, = x,
-XI - x2 = X2
-2x1 - 2x, + 2x, + x4 = x,
XI + x2-Xg = X4
The first two equations give x2 = 0 and x, = 0. Then the last equation gives x, = -x,.
Thus, the eigenvectors are
[n s s]
Example 9 :Obtain the eigenvalues and eigenvectors of
A=
which is equivalent to
X2 = -x 1
XI = -x2
X3 = -x3
The last equation gives x, = 0. Thus, the eigenvectors are Eipenvalues and Eigei~vectors
which gives x, = x,
X1 = X2
X3 = X3
L 1
[::]
ain the eigenvalues of T.
Hence, the linear transformation T has no real eigenvalues. But, it has two complex
eigenvalues i and -i.
Try the following exercises now.
E E9) Obtain the eigenvalues and eigenvectors of the differential operator D : P,-+ P, :
D(a, + alx + a2x2)= a, + 2a2x, for a,, a,, a, E R.
E E10) Show that the eigenvalues of a square matrix A coincide with those of At.
Now that we have discussed a method of obtaining the eigenvalues and eigenvectors of
a matrix, let us see how they help in transforming any square matrix into a diagonal
10.4 DIAGONALISATION
In this section we start with proving a theorem that discusses the linear independence
nding to different eigenvalues.
Elgenvducs and Elgenvedors Theorem 2 :Let T : V + V he a linear transformation on a finite-dimensional vector
space V over the field F. Let A,, A ,,.., A, be the distinct eigenvalues of T and v,, v, .....
v, be eigenvectors of T corresponding to A,, A, ,.... A, respectively. Then v,, v,,. ......
v, are linearly independent over F.
Proof: We know that
Tvi=Aivi,AiE F,O # v i E V f o r i = 1 , 2 , .... m,andAi#Ajfori# j.
Suppose, if possible, that {v,,v,,. ... v,) is a linearly dependent set. Now, the single
non-zero vector v, is linearly independent. We choose r(Sm) such that {v,,~,,.... v,)
. ,
is linearly independent and {v, v,, ...., v, ,v,) is linearly dependent. Then
v, = dlvl + a2v2+ ..... + a,,vF1 .........(1)
for some a,, a,, ....... a, in F.
Applying T, we get
Tv,= a,T vl + a,Tv, + ... + a,,T vr-,. This gives
A?, = a, Alvl + a2A2v2+ ..... + a,-, A,-, V,-, ....... (2)
Now, we multiply (1) by A, and subtract it from (2), to get
0 = a,(h, - A,) V, + a2(A2-A,)v, + ..... + a,, (A,l-A,)v,l
Since the set {v,, v,, .... v,- ,) is linearly independent, each of the coefficients in the
above equation must be 0. Thus, we have ai(Ai-A,) = 0 for i = 1 , 2 , ........ r-1.
Buthi#A,fori= 1 , 2.......r-l.Hence(Ai-A,)#Ofori= 1 , 2 ....... r-1,andwemust
have ai = 0 for i = 1, 2, ..... r - 1. However, this is not possible since (1) would imply
that v, = 0, and, being an eigenvector, v, can never be 0. Thus, we reach a contradiction.
Hence, the assumption we started with must be wrong. Thus, {v,, v,, ...... v,) must be
linearly independent, and the theorem is proved,
We will use Theorem 2 to choose a basis for a vector space V so that the matrix [TI, is
a diagonal matrix.
where A,, A,, ..., A, are scalars which need not be distinct.
The next theorem tells us under what conditions a linear transformation is
diagonalisable.
Theorem 3 : A linear transformation T, on a finite-dimensional vector space V, is
diagonalisable if and only if there exists a basis of V consisting of eigenvectors of T.
Proof: Suppose that T is diagonalisable. By definition, there exists a basis B = {v,,
v, ,...., v,) of V, such that
- a ! 0 ... 0
0 a! ... 0
[TIB= : . .
. :. ,which means that T is diagonalisable.
0 0 ... a ,-
Proof :Let A,, A,,. .,A, be the n distinct eigenvalues of T. Then there exist eigenvectors
v,, v,,. ..,v, corresponding to the eigenvalues A,, A,,. .,An,respectively. By Theorem 2,
the set ,v,, v,,. .,v,, is linearly independent and has n vectors, where n = dim V. Thus,
from Unit 5 (corollary to Theorem 5), B = {v,, v,, -......,v,) is a basis of V consisting of
eigenvectors of T. Thus, by Theorem 3, T is diagonalisable.
Just as we have reached the conclusions of Theorem 4 for linear transformations, we
define diagonalisability of a matrix, and reach a similar conclusion for matrices.
Definition: An n x n matrix A is said to be diagonalisable if A is similar to a diagonal
matrix, that is, F'AP is diagonal for some non-singular n x n matrix P.
Note that the matrix A is diagonalisable if and only if the matrix A, regarded as a linear
transformation A: V,(F) + V,(F) : A(X) = AX, is diagonalisable.
1 Thus, Theorems 2,3 and 4 are true for the matrix A regarded as alinear transformation
from Vn(F) to V,(F). Therefore, given an n x n matrix A , we know that it is
diagonalisable if it has n distinct eigenvalues.
i; We now give a practical method of diagonalising a matrix.
Theorem 5: Let A be an n x n matrix having n distinct eigenvalues A,, A,,. ..,A,. Let X,,
X,,.......,XnEVn(F) be eigenvectors of A corresponding to A,, A,,. ..,A,, respectively.
Let P = (XI, X,, ......., X,) be the n x n matrix having X,, X, ,...,X, as its column
vectors. Then
P-'AP = diag(A,, A,,. ..,An).
Proof: By actual multiplication, you can see that
AP = A(Xl, X,,.. ..,X,,)
:h ! = (AX,, AX, ,....,AX,)
j
$11
: = (l,X1 A,%, " h,X,)
r A, 0 .... 0 1
.... .
0 0 ... A,'
= Pdiag (XI, ,.., A,).
! Now, by Theorem 2, the column vectors of P are linearly independent. This means that
t
P is invertible (Unit 9, Theorem 6). Therefore, we can pre-multiply both sides of the
b matrix equation AP = P diag (A,, A,,. ..., 1,) by P-' to get P-'AP = diag (A;, A,,.. .,A,).
Let us see how this theorem works in practice.
-81
Example 11: Diagonalise the matrix
.=[: 2
2 -21
P=
L-:2 1
-:I
2 . Check, by actual multiplication, that
P-'AP = [ 5 0 0
0 3 01 ,which is in diagonal form.
0 0 -3
The following exercise will give you some practice in diagonalising matrices
E E12) Are the matrices in Examples 7 , 8 and 9 diagonalisable? If so, diagonalise them.
E 3) If 3 is an eigenvalve, then 3 11 1
rxi roi
f such that
b
corresponding to 3.
[
2 1
0 21
0
0
4
[ i] [i] = .This gives us the equations
t-a10 ... 0
0 t-a, ... 0
m) f ~ ( t j =. ... . = (t-a,)(t-aJ ... (t- a,)
... .
0 0 ... t-a,
.'. ,its eigenvalues are a,, a,,....,a,.
The eigenvectors corresponding to a, are given by
[:::I.
Then[D], = 0 0 2
t -1
.'. ,the characteristic polynomial of D is 0
0
t -2 = t3
0 0 t,
r]
E12) Since the matrix in Example 7 has distinct eigenvalues 1, -1 and -2, it is
diagonalisable. Eigenvectors corresponding to
0 0 2 1 0 0
..,ifP=[:r-:1 1 1 thenP-'[I 0 01 -2l]p=[o 0 10 - O]
2.
The matrix in Example 8 is not diagonalisable. This is because it only has two distinct
eigenvalues and, corresponding to each, it has only one linearly independent
eigenvector. :., we cannot find a basis of V,(F) consisting of eigenvectors. And now
apply Theorem 3.
The matrix in Example 9 is diagonalisable though it only has two distinct eigenvalues.
This is because corresponding to A, = -1 there is one linearly independent eigenvector,
but corresponding to A, = 1 there exist two linearly independent eigenvectors.
Therefore, we can form a basis of V3(R) consistingof the eigenvectors