Algebra Canonical Forms: Jordan Form and Rational Canonical Form 5.

2
Chapter 5 (i) p1 ( x ) p ( x ) , p2 ( x ) p ( x )

(ii) if p1 ( x ) q ( x ) , p2 ( x ) q ( x ) , then p ( x ) q ( x ) .

Canonical Forms: Jordan Form and Rational By Lemma 4.1.5, p1 ( x ) p ( x ) , p2 ( x ) p ( x ) .

Canonical Form (ii) Since p1 ( x ) q ( x ) ,

In this chapter, the two types of canonical forms, viz., the Jordan canonical q ( x ) = m ( x ) p1 ( x ) ⇒ q (T1 ) = m (T1 ) p1 (T1 ) = 0 (by(1)) → (3)
form and the rational canonical form are discussed.
Also since
§5.1 CANONICAL FORMS: DECOMPOSITION OF V: Jordan Form
p2 ( x ) q ( x ) , q ( x ) = n ( x ) p2 ( x ) ⇒ q (T2 ) = n (T2 ) p2 (T2 ) = 0 (by(2)) → (4)
The Jordan Canonical form improves a condition that all the characteristics
roots of the transformation T should lie in F. Otherwise we can pass to a splitting Consider v ∈ V . Then vq (T ) = ( v1 + v2 ) q (T ) , as V = V1 ⊕ V2
field K containing F and obtain the canonical form there.
Throughout this section V is assumed as a finite dimensional vector space = v1q ( T1 ) + v2 q (T2 ) = v1 .0 + v2 0 = 0 (by (3) and (4))
over F.
Lemma 5 .1.1 Since vq (T ) = 0, ∀v ∈ V , q ( T ) = 0. Since p ( x ) is the minimal polynomial for

Suppose that V = V1 ⊕ V2 , where V1 and V2 are subspaces of V invariant T and q ( x ) is any other polynomial which T satisfies, p ( x ) q ( x ) , proving (ii).

under T ∈ A (V ) . Let T1 and T2 be the linear transformations induced by T on V1 Corollary 5.1.2

and V2 respectively. If the minimal polynomial of T1 over F is p1 ( x ) while that of If V = V1 ⊕ V2 ⊕ ... ⊕ Vk where each Vi is invariant under T and if p0 ( x ) is

T2 is p2 ( x ) , then the minimal polynomial for T over F is the least common the minimal polynomial over F for Ti , the linear transformation induced by T on
Vi , then the minimal polynomial of T over F is the least common multiple of
multiple of p1 ( x ) and p2 ( x ) .
p1 ( x ) , p2 ( x ) ,..., pk ( x ) .
Proof :
Proof :
Let V = V1 ⊕ V2 and let T ∈ A (V ) . Then since V1 and V2 are invariant under
T, they induce linear transformations T1 and T2 on V1 and V2 respectively. Now, (
Since lcm ( p ( x ) , q ( x ) , γ ( x ) ) = lcm p ( x ) , lcm ( q ( x ) , γ ( x ) ) ) and

let p ( x ) , p1 ( x ) , p2 ( x ) be the minimal polynomials for T , T1 , T2 respectively. in general lcm ( p1 ( x ) , p 2 ( x ) ,..., p k ( x ) )

Then p (T ) = 0, p1 (T1 ) = 0 → (1)

( ( ( ( ( )))))
= l.c.m. p1 ( x ) , l.c.m. p2 ( x ) , l.c.m. p3 ( x ) ,... l.c.m.( pk −1 ( x ) , pk ( x ) ) ,...
p2 (T2 ) = 0 → ( 2)
by extending the concept used in the above lemma to V = V1 ⊕ ... ⊕Vk , the
Now to prove that the l.c.m. of p1 ( x ) and p2 ( x ) we prove that corollary follows.
5.3 Algebra Canonical Forms: Jordan Form and Rational Canonical Form 5.4
In the following theorem, V is decomposed as a direct sum of subspaces
Consider hi (T ) ≠ p ( T ) = 0 and hence hi ( T ) ≠ 0 and therefore there exists
invariant under T such that on each of these subspaces the linear transformation
induced by T has as minimal polynomial, a power of an irreducible polynomial. v ∈ V such that vhi (T ) ≠ 0 .
Theorem 5.1.3
l l
Let w = vhi (T ) ≠ 0 and consider w ( qi (T ) ) = ( vhi (T ) ) ( qi (T ) )
i i

For each i = 1, 2,...k , Vi ≠ {0} and V = V1 ⊕ V2 ⊕ ... ⊕ Vk and the minimal

li
polynomial of Ti is ( qi ( x ) ) where Ti is the linear transformation induced by (
= v hi ( T ) qi (T )
il
)
T on Vi which are invariant under T ∈ A (V ) . = v ( p (T ) )
Proof :
= v0 = 0 .
We first introduce Vi ’s for i = 1, 2,...k and claim that (1) Vi ≠ {0} i = 1 to k,
Thus w ∈ Vi and hence Vi ≠ {0} , proving (i).
li
(2) V = V1 ⊕ V2 ⊕ ... ⊕ Vk and (3) ( qi ( x ) ) is the minimal polynomial of Ti where
(ii) Now we claim that Vhi (T ) ⊂ Vi , ∀i = 1, 2,...k .
li
( q ( x))
i ’s are factors of the minimal polynomial p ( x ) of T ∈ A (V ) . Suppose
Let x ∈ Vhi (T ) ⇒ x = vhi (T ) ⇒ x = w ∈ Vi and thus Vhi (T ) ⊂ Vi , proving (ii).
that p ( x ) in F [ x ] is the minimal polynomial of T over F. Since a polynomial in We claim that

(iii) v j hi (T ) = 0 for i ≠ j , v j ∈ V j .
l1 l2 lk
F [ x ] can be uniquely factorized, p ( x ) = ( q1 ( x ) ) ( q2 ( x ) ) ...( qk ( x ) ) , where

q1 ( x ) , q2 ( x ) ,..., qk ( x ) are irreducible factors of p ( x ) . Now, let lj

Now since ( q j ( x ) ) hi ( x ) for i ≠ j

{ li
}
Vi = v ∈ V v ( qi ( x ) ) = 0 , i = 1, 2,...k . lj
hi ( x ) = m ( x ) ( q j ( x ) ) and therefore for v j ∈ V j
Claim 1

(i) Vi ≠ {0} i = 1, 2,...k .

(
v j hi ( T ) = v j m (T ) ( q j (T ) )
lj
)
Proof :
( l
)
= v j ( q j (T ) ) j m (T ) = 0 m (T ) = 0 , as v j ∈ V j , proving (iii).

Consider h1 ( x ) = ( q2 ( x ) )
l2 l3
( q ( x ) ) ...( q ( x ) )
lk

3 k
 (iv) Now we claim that Vi is a subspace of V .
h2
l1 l
( x ) = ( q ( x ) ) ( q ( x ) ) ...( q ( x ) )
lk 
1 3 k
 (I ) li li
Let x, y ∈ Vi and consider (α x + β y ) ( qi ( T ) ) = α x ( qi (T ) ) + β y ( qi (T ) )
li

⋮ 
lk −1 
l1 l2
hk ( x ) = ( q1 ( x ) ) ( q 2 ( x ) ) ... ( qk −1 ( x ) )  = α .0 + β .0 = 0

Hence α x + β y ∈ V , proving (iv). ( Vi is invariant under T)

5.5 Algebra Canonical Forms: Jordan Form and Rational Canonical Form 5.6
(v) We now prove that if x ∈ ViT , then x = vi T , vi ∈ Vi and hence Thus v1 = 0 . But by assumption vi ≠ 0 for any i = 1 to k. This contradiction
li li
( li
)
x ( qi ( T ) ) = ( viT ) ( qi (T ) ) = vi ( qi (T ) ) T = 0T = 0 and thus Vi T ⊆ Vi .
leads us to conclude that V = V1 ⊕ ... ⊕ Vk , proving Claim 2.
Claim 3
Since Vi is invariant under T, T induces a linear transformation Ti on Vi . li

Claim 2 ( q ( x))
i is the minimal polynomial of Ti .

V = V1 ⊕ V2 ⊕ ... ⊕ Vk Proof:

Proof : By Lemma 4.1.5 (iii), the minimal polynomial of Ti is a divisor of the

l l l
From (I) we see that h1 ( x ) ,....., hk ( x ) are relatively prime and hence there minimal polynomial p (T ) and p ( x ) = ( q1 ( x ) ) ( q2 ( x ) ) ...( qk ( x ) ) .
1 2 k

exists polynomials a1 ( x ) , a2 ( x ) ..., a k ( x ) such that Therefore the minimal polynomial of Ti should be of the form

a1 ( x ) h1 ( x ) + a2 ( x ) h2 ( x ) + ... + ak ( x ) hk ( x ) = 1 (qi ( x )) f i
with f i ≤ li . By Corollary 5.1.2. p ( x ) = l.c.m.

⇒ a1 (T ) h1 ( T ) + a2 (T ) h2 (T ) + ... + ak (T ) hk (T ) = 1 ( ( q ( x) )
1
f1
,...,( qk ( x) )
fk
) = ( q ( x) ) ( q ( x) )
1
f1
2
f2 fk
...( qk ( x) ) .

Let v ∈V . Then v = v.1 = v ( a1 ( T ) h1 (T ) + ... + ak (T ) hk (T ) ) = v1 + v2 + ... + vk . But p ( x ) should have unique representation and hence f i = li . Thus
Now, suppose v1 + ... + vk = 0 we prove that vi = 0 for i = 1, 2,...k . Suppose li
( q ( x))
i is the minimal polynomial of Ti . This completes the theorem.
v i ≠ 0 for any i = 1 to k . Consider, ( v1 + v 2 + ... + v k ) h1 ( T ) Corollary 5.1.4
= 0.h1 (T ) = 0 ⇒ v1 h1 (T ) + v2 h1 (T ) + ..... + vk h1 (T ) = 0 ⇒ v1 h1 (T ) = 0 as If all the distinct characteristic roots λ1 ,...,λ k of T lie in F, then V = V1 ⊕ ... ⊕Vk
l
v j hi (T ) = 0 for i ≠ j . Now as v1 ∈ V1 , v1 ( q1 (T ) ) 1 = 0 and since h1 ( x ) and { l
where each Vi = v ∈V v (T − λ i ) = 0 and Ti has only one characteristic root
i
}
l1
( q ( x )) are relatively prime, there exists g ( x ) and l ( x ) in F [ x ] such that λ i on Vi .
1

l1
Proof :
h1 ( x ) g ( x ) + ( q1 ( x ) ) l ( x ) = 1 .
If all the distinct characteristic roots λ 1 , λ 2 ,..., λ k of T lie in F, then
l1
⇒ h1 (T ) g (T ) + ( q1 (T ) ) l (T ) = 1
l l
the minimal polynomial of T takes the special form. q ( x ) = ( x − λ1 ) 1 ...( x − λk ) k .

(
Now v1 = v1 .1 = v1 h1 (T ) g (T ) + ( q1 (T ) ) l (T1 )
l1
) Thus qi ( x ) = x − λ and hence { li
Vi = v ∈V v ( qi (T ) ) = 0 } becomes

l1
= v1h1 (T ) g (T ) + v1 ( q1 (T ) ) l (T ) = 0 + 0 = 0 {
Vi = vi ∈ V v (T − λ i ) i = 0
l
} and obviously on Vi , Ti has only λ i as a
characteristic root.
5.7 Algebra Canonical Forms: Jordan Form and Rational Canonical Form 5.8
Definition 5.1.5 Now, as Ti = λ i I + Ti − λ i I

λ 1 0 … 0 0 m (Ti ) = m ( λi I ) + m ( Si )
0 λ 1 … 0 0 

⋮ ⋮ ⋮ ⋮ ⋮  λ1 I ⋯ 0   M i1 .. 0 
The matrix  0     
 0 0 … λ 1  where λ ’s are on the diagonal and 1’s on the  ⋮ ⋱ ⋮ + ⋮ ⋱ ⋮ 
 0 0 0 … 0 λ  
 ⋯ λ i I   0 ⋯ M ir 
 i 

super diagonal and 0’s elsewhere is a basic Jordon block belonging to λ .

Theorem 5.1.6  λi 0 0 ⋯ 0  0 1 0 ⋯ 0
   
0 λi 0 ⋯ 0  0 0 1 ⋯ 0
Let T ∈ A (V ) have all its characteristic roots λ1 ,λ 2 ,...,λ k in F. Then a basis
Now λi I + M i1 =  0 0 λi ⋯ 0  +  0 0 0 ⋯ 0
of V can be found in which the matrix of T is of the form.    
 ⋱ 0  ⋱ 1
 J1 0 … 0
0
 0 0 ⋯ λi   0 0 0 ⋯ 0 
   Bi 1 
 0 J2 ⋯ ⋮ 
 
⋮ ⋮ ⋱ ⋮  where each J i =  Bi 2  where Bi1 ,..., Biri are  λi 1 0 ⋯ 0
   
0 0 … J k  ⋱ Bi ri  
λi 1 ⋯

 0 0
0 0 λi  = Bi
basic Jordan blocks belonging to λ i . =  1

 ⋱ 1
Proof : 0
 0 0 ⋯ λi 
Since λ1 ,...,λ k are the characteristic roots of T in F, by the above corollary,
V = V1 ⊕ ... ⊕Vk where each Vi is invariant under T and T induces a linear  Bi1 0 
 
{ i l
transformation Ti on Vi and also each Vi = vi ∈ V v (T − λ i I ) = 0 . Since on } ∴ m (Ti ) = 
 0
⋱  = Ji
Bir 
l  1 

Vi , (Ti − λ i I ) i = 0 we see that Si = Ti − λ i I is nilpotent.

Since V = V1 ⊕ V2 ⊕ ... ⊕ Vk , where each Vi is invariant under T, by
By Theorem 4.2.6, there is a basis of Vi such that the matrix of Si in that basis
 J1 0⋯ 0 
 M i1 .. 0   
  Lemma 4.2.4. there is a basis of V in which ( )  ⋮ ⋱ ⋮  where J i′ s
m T =
is m ( Si ) =  . ⋱ ⋮  . 0 ⋯ J 
 k 
 0 ⋯ M 
 iri 
are matrices of the induced transformations Ti on Vi .
5.9 Algebra Canonical Forms: Jordan Form and Rational Canonical Form 5.10
Note 5.1.7. Proof :
In the above theorem, we could arrange the block matrices so that in each J i Let A be the matrix of T in the standard basis of F ( ) . Let K be the splitting
n

the size of Bi1 ≥ size of Bi 2 ≥ ... size of Bi1 ≥ Bi2 ≥ ... when this has been done field of the minimal polynomial of A over F. Since A ∈ Fn and Fn ⊆ K n , A is in
K n . Thus A ∈ K n has all its characteristic roots (as roots of minimal polynomial
 Ji 0⋯ 0
  of A are characteristic roots of A) in K by Theorem 5.1.6, there is a basis v1 ,...,vn
then the matrix  ⋮ ⋱ ⋮  is called the Jordan form of T.
0 ⋯ J k  of V such that matrix of A in v1 ,...,vn is in Jordon form. But then the matrices of A

The above theorem reduces to Theorem 4.2.5. if T is nilpotent, as its in the basis e1 ,...,en and in the basis v1 ,...,vn are similar i.e., there exists an
characteristic roots are zero. invertible matrix C in K n such that CAC −1 is in Jordan form.
Proposition 5.1.8
Solved Problems
Two transformations in A (V ) which have all their characteristic roots in
F are similar iff they can be brought to the same Jordan form.  1 1 1
 
Proof : 1. Find the Jordan form of  −1 −1 −1  .
 1 1 0
Let S and T be any two transformations such that their characteristic roots are  
in F. Let the matrices of S and T have their Jordan forms as given below.
Solution:
 J1 0  J '1 0   Bi1 0 
       1 1 1
m (T ) =  ⋱  and m ( S ) =  ⋱ ⋱  
 where J i =   Let A =  −1 −1 −1  . Then its characteristic equation is A − λI = 0
0 J r   0
  J 's  0 Bir   1 1 0
 i 
 

 B'i1 ⋯  i.e., λ 3 = 0 ⇒ λ = 0 .
 
and Bir = λi I + M ir & J 'i =  ⋮ ⋱ ⋮  and Bi′m = λi I + M im .
i i i i 0 1 0 
 ⋯ Bim 
 i  ∴ J ( A ) = 0 0 1 
.
0 0 0 
Then n1 ≥ n2 ≥ ... ≥ nr and m1 ≥ m2 ≥ ... ≥ m8 and by Remark 4.2.12, S and T are
similar iff r = s and n1 = m1 ,...,nr = mr . Thus m ( S ) and m (T ) are similar iff they  1 1 1  1 1 1
have the same Jordan form.    
2. Check whether A =  − 1 − 1 −1 and B =  −1 −1 −1 are similar or not.
Remark 5.1.9  1 1 0   1 0 0
  
Let A ∈ Fn and suppose that K is the splitting field (The smallest field
containing all the roots of f ( x ) ∈ F [ x] ) of the minimal polynomial of A over F.

Then there is an invertible matrix C in K n such that CAC −1 is in Jordan form.

5.11 Algebra Canonical Forms: Jordan Form and Rational Canonical Form 5.12
Solution:
 J1 
  0 1  1 1
 0 1 0 (ii) ( )  J = J = = J3
J A = J2  , where 1  0 0  ; 2  0 1 
   
By Problem (1) , J ( A ) =  0 0 1  . Now the characteristic equation of  J3 
 0 0 0
 
 J1 
B is B − λI = 0 i.e., λ 3 = λ = 0 ⇒ 0 , 1, − 1 are the characteristic roots of A.   0 1  1 1
J
(iii) J ( A) =   , where J1 =   J 2 = [ 0] = J 3 J 4 =  
2

0 0 0   J3  0 0  0 1
   
∴ J ( B) =  0 1 0  . Since J ( A ) and J ( B ) are not the same A and B  J4 
0 0 −1 

are not similar.  J1 
 
(iv) J ( A) = 
J 2  , where J =  0 1 J = 1 1 J = 1 ; J = 1
1   2   3 [ ] 4 [ ]
2
3. Write down all the possible Jordan forms for a 6 × 6 matrix A with x 2 (1 − x ) as  J3 
   0 0  0 1
minimal polynomial.  J4 
Solution:
Since the given matrix A is a 6 × 6 matrix, its characteristic polynomial should  J1 
2  
be of degree 6. Here x 2 (1 − x ) is given as the minimal polynomial of A and hence J  , where J =  0 1  J = 0 , J =  1 1 J = [ 1 ]
(v) J ( A) = 
2
 J3  1   2 [ ] 3   4
the characteristic polynomial C ( x ) of A can be in one of the following possible    0 0  0 1
 J4 
forms viz.
2
(i) C ( x ) = x 2 x 2 (1 − x ) Exercise

2 2 1. Find the Jordan form of the matrix

(ii) C ( x ) = x 2 (1 − x ) (1 − x )
 3 −1 1 
(iii) C ( x ) = x 2 xx (1 − x )
2
0 3  
(i)   (ii)  −1 5 −1
1 2  1 −1 3 
2
(iv) C ( x ) = x (1 − x ) (1 − x )(1 − x )
2  

2
(v) C ( x ) = x 2 x (1 − x ) (1 − x )  2 0 −1  2 −2 2 
   
Thus the 5 possible Jordan forms of A are (iii)  0 2 0
(iv) 1 1 1
 −1 0 2   1 3 −1 
  
 J1 
  0 1  1 1 2. Find all of the possible Jordan forms for all 10 × 10 matrices over ℚ having
(i) ( ) 
J A = J2  , where J1 =  0 0  = J 2 ; J 3 =  0 1 
    
J 3 
2 3
 x 2 ( x − 1) ( x + 1) as minimal polynomial.
5.13 Algebra Canonical Forms: Jordan Form and Rational Canonical Form 5.14
3. Find all of the possible Jordan forms for all 6 × 6 matrices over ℚ having (ii) If V is a finite dimensional vector space, then by Note (i) and
3 Theorem 3.3.23. V = V1 ⊕V2 ⊕ ... ⊕ Vk where each Vi is a cyclic submodule
( x − 2 )( x + 2 ) as minimal polynomial.
as F [ x ] is a Euclidean ring.
§5.2. CANONICAL FORM: Rational Canonical Form
Definition 5.2.3
To obtain the Jordan Form of T ∈ A (V ) , T must have its characteristic roots V is said to be cyclic relative to T if for every w ∈ V there exists v ∈ V such
in F. In rational canonical form the location of characteristic roots is not assumed. that w = vf (T ) .
Proposition 5.2.1 Lemma 5.2.4
If V is a vector space over a field F and T ∈ A (V ) , then V can be made into a Suppose V is cyclic relative to T ∈A(V) and p( x) = γ0 +γ1x +γ2 x2 + ...+γr−1xr−1 + xr
module over F [ x ] , the ring of polynomials in x over F. is the minimal polynomial of T. Then there exists a basis of V in which

Proof :  0 1 0 0 
⋯
 0 0 
⋯
We define for v ∈ V and f ( x ) ∈ F [ x ] , f ( x ) v = vf ( T ) where T ∈ A (V ) .  0 1
Then as V is a vector space, it is an abelian group under addition. The verification m (T ) =  ⋱ .
 
of the three axioms m − 1,m − 2,m − 3 are given below.  0 0 0 ⋯ 1 
 −γ 0 −γ1 … −γ r −1 
m − 1 : f ( x )( u + v ) = f ( x ) u + f ( x ) v
Proof :
L.H.S. = ( u + v) f (T ) = uf (T ) + vf (T ) = f ( x) u + f ( x) v = R.H.S.
Since V is cyclic with respect to T, for every w ∈ V , there exists v ∈V such
m−2: ( f ( x ) + g ( x ) ) v = f ( x ) .v + g ( x ) .v that w = vf (T ) ,T ∈ A (V ) → (1).

L.H.S.= v ( f (T ) + g (T ) ) = v ( f (T ) ) + v ( g ( T ) ) = f ( x ) v + g ( x ) .v = R.H .S. Since p ( x ) = γ 0 + γ 1 x + γ 2 x 2 + ... + γ r −1 x r −1 + x r is the minimal polynomial of

T, p ( T ) = 0 i.e., T r = −γ 0 − γ 1T − ... − γ r −1T r −1 → (2) .

m − 3 : g ( x ) ( f ( x ) v ) = ( g ( x ) f ( x )) v
We claim that if for some polynomial s ( x ) ∈ F [ x ] vs (T ) = 0 , then
R.H.S. = v ( g (T ) f (T ) ) = v ( f (T ) g (T ) ) = ( vf (T ) ) g (T ) = g ( x) ( f ( x) v) = LHS
s (T ) =0 for, w ( s (T ) ) = ( vf (T ) ) s (T ) .
Thus V is an F [ x ] – module.
Note 5.2.2 = v ( f ( T ) s ( T ) ) = v ( s (T ) f (T ) ) = ( vs ( T ) ) f (T ) = 0. f (T ) = 0 for every
(i) If V is a finite dimensional vector space, then V becomes a finitely generated
w ∈ V . Hence s (T ) = 0 .
F [ x ] - module.
5.15 Algebra Canonical Forms: Jordan Form and Rational Canonical Form 5.16

We now claim that v, vT , vT 2 ,..., vT r −1 are linearly independent. Suppose Definition 5.2.5

α1v + α 2 vT + ... + α r vT r −1 = 0 with not all scalars zero. Then If f ( x ) = α 0 + α1 x + ... + α r −1 x r −1 + x r is in F [ x ] , then the r × r matrix

v (α1 + α 2T + ... + α r T r −1 ) = 0 ⇒ vg ( T ) = 0 where g ( x) = α1 +α2 x + ... +αr xr−1 .  0 1 0 0 

⋯
 0 0 1 0 
⋯
By the previous claim, this implies that g (T ) = 0 which is not possible as the 
 ⋱ 
degree g ( x ) is r – 1 which is less than the degree of the minimal polynomial  
1  is called the companion matrix of ( ) and is
0 ⋯
f x
 0 0
p ( x ) . Thus, v, vT , vT 2 ,..., vT r −1 are linearly independent. Also from (2), T r − k is  −α 0 −α1 … −α r −1 

a linear combination of 1, T,...,T r −1 ,k ≥ 0 and hence w = vf (T ) .

written as C ( f ( x ) ) .
= v (linear combination of 1,T,...,T r −1 ) In view of the above definition, Lemma 5.2.4. may be restated as “if V is

= a linear combination of v , vT ,..., vT r −1 . Thus v , vT ,..., vT r −1 span V cyclic relative to T and if the minimal polynomial of T in F [ x ] is p ( x ) , then in
and hence is a basis of V. some basis of V the matrix of T is C ( p ( x ) ) ”.
Now consider,
Preposition 5.2.6
v1T = vT = 0.v1 + 0.v2 + ... + 0.vr
The characteristic polynomial of the companion matrix C ( p ( x ) ) is
v2T = ( vT ) T = 0.v1 + 0.v2 + ... + 0.vr
p ( x ) = γ 0 + γ 1 x + ... + γ r −1 x r −1 + x n .
.
. Proof :

r −1
vrT = ( vT ) T = vT r = γ 0 v1 + ( −γ 1 ) v2 + ... + ( −γ r −1 ) vr  0 1 0 ⋯ 0 
 0 0 1 ⋯ 0 
where v1 = v, v2 = vT ,..., vr = vT r −1 v1 = v , v2 = vT ,..., vr = vT r −1 
 ⋱ 
 
Thus m (T ) in the basis v1 ,...,v r is  0 0 0 ⋯ 1 
Consider C ( p ( x )) =  −γ 0 −γ1 … −γ r −1  .
 0 1 0 ⋯ 0 
 0 0 1 ⋯ 0 
 0 −1 0 ⋯ 0 
 ⋱  0 
   x −1 ⋯ 0 
 0 0 0 ⋯ 1  xI − C ( p ( x ) ) = ⋱ 
Then
 −γ 0 −γ1 … −γ r −1   
0 0 ⋯ x −1 
γ 0 γ1 γ 2 ⋯ x + γ r −1 
5.17 Algebra Canonical Forms: Jordan Form and Rational Canonical Form 5.18
By adding to the first column, x times the second column x 2 times the third Since the minimal polynomial of Ti should divide the minimal polynomial of
r −1
column etc., and x times the last column, we get the determinant as e
T = ( q ( x ) ) , the minimal polynomial of Ti is of the form (q( x ) ) i where
e

 0 −1 0 ⋯  0 ei ≤ e → ( P ) . By suitably arranging Vi ’s we have e1 ≥ e2 ≥ ,...,≥ ek . Since Vi is

 0 x −1 ⋯  0
  a cyclic submodule relative to Ti , there is a basis of Vi in which
xI − C ( p ( x ) ) = ⋱ 
 
 0 0 ⋯ x −1  ( e
)
m (Ti ) = C q ( x ) i , by Lemma 5.2.4. Now, since V = V1 ⊕V2 ⊕ ... ⊕Vk , by
 p ( x ) γ1 γ 2 ⋯ x + γ r −1 
 Lemma 4.2.4, there exists a basis of V in which m (T ) is

−1 0 ⋯ 0
 (
C q ( x )e1 ) 

r −1 x −1 ⋯ 0  
= ( −1) p ( x) ⋱
⋮ ⋱  .
0 … x 1

 (
C q ( x) r
e
) 


e1
Now we prove that e = e1 . For vi ∈ Vi we see that vi ( q (T ) ) = 0 i = 1 to r .
r −1 r −1
= ( −1) p ( x )( −1) = p ( x)
e1 e1
Thus C ( p ( x ) ) has p ( x ) as its characteristic polynomial. This implies that v ( q (T ) ) = 0 and hence ( q (T ) ) = 0 . But q ( x ) is the minimal
e

Theorem 5.2.7 polynomial of T and hence e ≤ e1 → ( Q ) . From ( P ) and ( Q ) we have e = e1 .

e
If T ∈ A (V ) has as minimal polynomial p ( x ) = q ( x ) where q ( x ) is a monic Corollary 5.2.8

If T ∈ A (V ) has the minimal polynomial p ( x ) = q1 ( x ) 1 ...qk ( x ) k over F,

l l
irreducible polynomial in F [ x ] , then a basis of V over F can be found in which

where q1 ( x ) ,q2 ( x ) ,...qk ( x ) are irreducible distinct polynomials in F [ x ] , then a

C q ( x )e1
 ( ) 
 basis of V can be found in which the matrix of T is of the form
 ⋱ 
the matrix of T is of the form   where

 ( e
C q ( x) r ) 

 R1



C q ( x )ei1
 i ( ) 

 ⋱  where each Ri =  
⋱
e = e1 ≥ e2 ≥ ,...,≥ er .  Rk   

Proof :

 (
C qi ( x ) iri
e
) 


Since V is a finitely generated F [ x ] - module, V = V1 ⊕ ... ⊕Vr where where ei = ei1 ≥ ei 2 ≥ ,...,≥ eiri

{ ( e
) }
Vi = v ∈ V v q ( x ) i = 0 (by Theorem 5.1.3.) and Vi is cyclic submodule. Also

each Vi is invariant under T and hence induces a linear transformation Ti on Vi .

5.19 Algebra Canonical Forms: Jordan Form and Rational Canonical Form 5.20
Proof : Theorem 5.2.10
(Cayley – Hamilton Theorem)
Let T ∈ A (V ) and p ( x ) = q1 ( x ) 1 . q2 ( x ) 2 ... qk ( x ) k be the minimal
e e e

Every linear transformation T ∈ A (V ) satisfies its characteristic polynomial.

polynomial of T. Since V is a finitely generated F [ x ] - module by Note 5.2.2,
Proof :
V = V1 ⊕ V2 ⊕ ... ⊕ Vk where Vi ’s are cyclic submodules, each Vi is invariant
e e e
Let p ( x ) = q1 ( x ) 1 .q2 ( x ) 2 ...qk ( x ) k be the minimal polynomial of
under T and T induces a linear transformation Ti on Vi such that the minimal
e e e e e e
ei
T ∈ A (V ) . Let q1 ( x ) 11 ...q1 ( x ) 12 ...q1 ( x ) 1r1 ...q k ( x ) k 1 ...q k ( x ) 1 r1 ...q k ( x ) krk be
polynomial of Ti is ( qi ( x ) ) .
the elementary divisors of T where e11 = e1 , e21 = e2 ,..., ek 1 = ek .
ei

Since ( qi ( x ) ) is the minimal polynomial of Ti on Vi , by the above theorem e e e e e

Since q1 ( x ) 11 ...q1 ( x ) 1r1 ...qk ( x ) k 1 ...qk ( x ) k 2 ...qk ( x ) krk = p ( x ) s ( x ) where
 0 1 0 0 ⋯ s ( x ) = the product of elementary divisors other than q1 ( x ) ,...,qk ( x )
e1 ek
and
 0 0 1 0 
⋯
 since p (T ) = 0, pT (T ) = 0 . Thus T satisfies its characteristic polynomial.
there exists a basis of Vi such that m (Ti ) is  ⋱  where
  Remark 5.2.11
 0 0 0 ⋯ 1 
 −γ 0 −γ 1 ⋯ −γ r −1  Every root of pT ( x ) is called a characteristic root of T and moreover the

ei = ei1 ≥ ei 2 ≥ ... ≥ eiri . multiplicity of any root of pT ( x ) as a root of the polynomial, equals its multiplicity

as a characteristic root of T. Also every characterstic root of T is a root of pT ( x ) .

Now, since V = V1 ⊕V2 ⊕ ... ⊕ Vk by Lemma 4.2.4. there exists a basis of V in
Theorem 5.2.12
 R1 
Let V and W be any two finite dimensional vector spaces over F and suppose
which m (T ) =  ⋱ .
  that ψ is a vector space isomorphism of V onto W. Suppose that S ∈ A (V ) and
 Rk 
T ∈ A (W ) are such that for any v ∈V , ( vS ) ψ = ( vψ ) T . Then S and T have the
Definition 5.2.9
same elementary divisiors.
The matrix of T in the statement of the above corollary is called the rational
canonical form of T and the polynomials Proof :
Let V and W be two finite dimensional vector spaces over F and ψ :V → W
q1 ( x ) 11 ,q1 ( x ) 12 ,...q1 ( x ) 1r1 ,...,qk ( x ) k 1 ,qk ( x ) k 2 ,...,qk ( x ) krk in F [ x ] are called
e e e e e e

the elementary divisiors of T and the product of these elementary divisiors is be a 1 – 1, onto homomorphism. Let S ∈ A (V ) and T ∈ A (W ) be such that for

denoted by pT ( x ) which is the characteristic polynomial of T. any v ∈ V , ( vS )ψ = ( vψ ) T → (1) .

We first claim that for any polynomial f ( x ) ∈ F [ x] and v ∈ V , ( vf ( S ) )ψ =

( vψ ) f (T ) . Consider, ( vS ) ψ = ( vS ) S = ( ( vS ) ψ )T = ( vψ )T
2 2
. In general
5.21 Algebra Canonical Forms: Jordan Form and Rational Canonical Form 5.22

( vS )ψ = ( vψ ) T
m m
and hence if f ( x ) ∈ F [ x ] , then ( vf ( S ) )ψ = ( vψ ) f (T ) . (ii) The only elementary divisors of T1 , the linear transformation induced by T
e
on W1 is q1 ( x ) .
1
Also, if f ( S ) = 0 then ( vψ ) f (T ) = 0 for any v ∈ V and since ψ is onto
(iii) The other elementary divisors of T are those of the linear transformation
Vψ = W and hence Wf (T ) = 0 which in turn implies that f (T ) = 0 . Also, if
T2 induced by T on N.
g ( x ) ∈ F [ x ] is such that g (T ) = 0 , then for any v ∈ V , v ( g ( S ) )ψ = 0 . Since
Now, since M ψ = N , M is isomorphic to N under ψ 2 , the isomorphism
ψ is 1 – 1, vg ( S )ψ = 0 for every v ∈ V and therefore g( S ) = 0 . Thus S and T induced by ψ .
satisfy the same set of polynomials in F [ x ] and hence must have the same Moreover if u ∈ M , then ( uS 2 )ψ 2 = ( uS )ψ = ( uψ ) T = ( uψ 2 ) T2 , hence S2
e e e
minimal polynomial p ( x ) = q1 ( x ) 1 .q2 ( x ) 2 ...qk ( x ) k where q1 ( x ) ,...,qk ( x ) are and T2 are in the same relation with ψ 2 as that of S and T with ψ .

distinct irreducible polynomials in F [ x ] . e

Also since the elementary divisors of S are q1 ( x ) and those of S2 and the
1

Now, we claim that if U is a subspace of V invariant under S then Uψ is a e

elementary divisors of T are q1 ( x ) and those of T2 , by repeating the above
1

subspace of W invariant under T, for (U ψ ) T = (US ) ψ ⊂ U ψ . argument we must arrive at S and T having the same elementary divisors.
Theorem 5.2.13
Since U and Uψ are isomorphic, the minimal polynomial of S1 the linear
transformation induced by S on U is the same as the minimal polynomial of T1 , S, T ∈ A (V ) are similar in A (V ) iff they have the same elementary divisors.

the linear transformation induced by T on Uψ . Proof :

Now since the minimal polynomial for S on V is Let S, T ∈ A (V ) be similar in A (V ) . Then they satisfy the same minimal
e e e
p ( x ) = q1 ( x ) 1 .q2 ( x ) 2 ...qk ( x ) k by Theorem 5.2.7., and its corollary we can take e
polynomial. Without loss of generality let q ( x ) be the minimal polynomial of S
e
as the first elementary divisor of S, the polynomial q1 ( x ) 1 and we can find a and T where q ( x ) is a monic irreducible polynomial and is of degree d (say).
subspace V1 of V which is invariant under S such that

(i) V = V1 ⊕ M , where M is invariant under S.

{
By Theorem 5.2.7. V = V1 ⊕ .... ⊕ V r where Vi = v ∈V v q (T ) = 0
i
( e
) }
i = 1 to r and Vi ’s are cyclic submodules invariant under T and the induced
(ii) The only elementary divisor of S1 , the linear transformation induced on
 
i e

transformation on Vi has as companion matrix C  qi ( x )  where

e
V1 by S is q1 ( x ) 1 .
 
(iii) The other elementary divisors of S are those of the linear transformation
e1 ≥ e2 ≥,..., ≥ er .
S2 induced by S on M.
Now as q(x ) is of degree ‘d’ and companion matrix is of order dei ,
Equivalently in W we have the following assertion.
using Lemma 4.2.10.
(i) W = W1 ⊕ N where W1 = V1ψ and N = M ψ are invariant under T.
5.23 Algebra Canonical Forms: Jordan Form and Rational Canonical Form 5.24
fm elementary divisiors. Then there are bases of V such that m(S) in one basis is
dim V ( q (T ) ) = d e1 − d f m + ... + d er − d f m
(
C q ( x)
e
) and also it is the same for m(T) in another basis of V.
≥ d ( e1 − f m ) + ... + d ( em−1 − fm ) + d ( em − f m ) → (1) .
fm
dim Vq (T )
If m(T) = m(S) in two different bases, then T and S must be similar (Refer
Now suppose V has another decompositions V = U1 ⊕ ... ⊕Us where solved problems).
Corollary 5.2.14
{ ( f
) }
U i = u ∈ V u q ( T ) i = 0 and U i ' s are cyclic submodules invariant under
Suppose the two matrices A, B in Fn are similar in K n where K is an extension
T and the induced transformation on U i has as comparison matrix (i.e., K ⊃ F ) of F, then A and B are already similar in Fn .

( f
)
C q ( x ) i where f1 ≥ f 2 ≥,..., ≥ f s . Proof :
Let A, B ∈ Fn be similar in K n . Then there exists an invertible matrix C in K n
Now dim V ( q (T ) )
fm
((
= dim U1 q ( T 1)
fm
) + ... + dimU ( q (T ) ) ) . We claim
s
fm
such that B = C −1 AC ⇒ CB = AC → (1).

that r = s, e1 = f1 ,..., er = f r Suppose not i.e., suppose ei ≠ fi . Then there is a Let V = F ( n ) and W = F ( n ) C . Let us define ψ : V → W by vψ = vC for every
first integer m such that em ≠ f m while e1 = f1 ,..., em −1 = f m−1 . For definiteness we v ∈ V , then ψ is well defined and 1 – 1. For, if v = v ' ⇔ vC = v ' C ⇔ vψ zv 'ψ as

may suppose that em > fm →( P) . C is invertible . ψ is obviously onto. Also ( v + v ' )ψ = ( v + v ' ) C .

Also (
U j q (T )
fm
)=0 for j = m, m + 1,..., s so that = vC + v ' C = vψ + v 'ψ and (αν )ψ = (αν ) C

fm fm fm = α ( vC ) = α ( vψ ) and hence ψ is a homomorphism. Thus ψ is 1 – 1,

V ( q (T ) ) = U1 ( q ( T ) ) ⊕ ... ⊕ U m−1 ( q (T ) )
onto homomorphism.
fm fm fm
⇒ dim V ( q (T ) ) = dim U 1 ( q (T ) ) + ... + dim U m −1 ( q (T ) ) and hence We now claim that ( vA)ψ = ( vψ ) B . Consider, ( vA)ψ = v ( AC ) = v ( CB )
(by (1)).
fm
⇒ dim V ( q (T ) ) = d ( f1 − f m ) + ... + d ( f m −1 − f m )
= v ( CB ) = ( vψ ) B .
= d ( e1 − f m ) + ... + d ( em−1 − f m ) → (2) Since the condition for A and B to have the same elementary divisors are true
by Theorem 5.2.11, A and B have the same elementary divisors and therefore by
as U j ’s are cyclic submodules and q(x ) is of degree ‘d’. the above theorem A and B are similar.
Using (2) in (1) Remark 5.2.15
fm fm If A, B ∈ Fn are similar in K n , then the above corollary states that there
dim V ( q ( T ) ) ≥ dim V ( q ( T ) ) + d ( em − f m )
exists a D in Fn such that B = D −1 AD and it does not imply that B = C −1 AC ,
⇒ d ( em − f m ) ≤ 0 or em ≤ f m , a contradiction to (P) and hence both S and
C ∈ Kn then C ∈ Fn .
T have the same elementary divisiors. Conversely, let S and T have the same