UNIT - III

CANONICAL FORMS

1) Triangular Form
Definition :

The linear transformations S , T ∈ A (V ) are said to be similar if there exists an invertible element

C ∈ A ( V ) such that T = CSC −1 .

Definition :
The relation on A (V) defined by similarity is an equivalence relation, the equivalence class of
an element will be called its similarity class.

Note :
To check the two linear transformations are similar or not is difficult. Therefore, we can use
similarity class which matrix in some basis. These matrices will be called the Canonical Forms.

Definition :

The subspace W of V is invarient under T ∈ A (V ) if T (W ) ⊂ W .

Lemma :

V
If W ⊂ V is invarient under T, then T induces a linear transformation T on defined by
W
T ( v + W ) = T ( v ) + W . If T satisfies the polynomial q ( x ) ∈ F [ x ] then so does T . If p1 ( x ) is the
minimal polynomial for T over F and if p ( x ) is that for T then p1 ( x) | p ( x ) .

54
 V
Proof : Let W ⊂ V is inveriant under T. T is linear transformation and V = be vector space which
w
contain the element as v = v + W for v ∈ V .

Define, T ( v ) = T ( v ) + W

Claim T is well defined and linear transform

v1 + W = v1, v2 = v 2 +W for v1, v2 ∈ V claim T ( v1 ) = T ( v2 ) .

∴ v1 +W = v2 +W ⇒ ( v1 − v2 )+ W = 0 ⇒ v1 − v2 ∈ W

T ( v1 − v2 ) ∈W ∴ W is invarient under T..

∴T ( v1 ) −T ( v2 ) ∈ W = T ( v1 ) + W = T ( v2 ) + W

⇒ T ( v1 ) = T ( v2 ) , hence T is well defined.

α , β ,∈ F , α v1 + β v2 ∈V for v1, v2 ∈ V .

∴T (α v1 + β v 2 ) = T ( α v1 + β v 2 ) = T ( α v1 + βv 2 ) +W

= ( α T ( v1 ) + β T ( v2 ) ) + W = α T ( v1 ) + W + β T ( v2 ) + W

= α T ( v1 ) + W  + β T ( v2 ) + W  = α T ( v1 ) + β T ( v2 )

Hence, T is linear..

Now if v = v + W ∈V then T 2 ( v ) = T 2 ( v ) + W = T ( T ( v ) ) + W

⇒ T 2 ( v ) = T (T ( v ) + W ) = T ( T ( v + W ) ) = ( T ) ( v )
2

Thus T 2 = T 2 . Similarly ( T k ) = ( T ) k for any k ≥ 0 ... (1)

Consequently, for any polynomial q ( x ) ∈ F [ x ] , q ( T ) = q ( T ) .

Q for q ( x ) = α 0 + α1 x + .... + α n x n

∴ q ( T ) = α 0 + α1T + .... + α nT m

( )
⇒ q ( T ) = α 0 + α1T + .... + α nT m = α 0 + α1T + .... + α nT m

Qα i ∈ F
n
= α 0 + α1T + .... + α n T

= α 0 + α1T + .... + α n ( T ) = q (T )
n
Q (1)

55
 For any q ( x ) ∈ F [ x ] with q ( T ) = 0 Q 0 is the zero transformation on V .

0 = q (T ) = q (T ) .... (2)

Let p1 ( x ) be the minimal polynomial over F satisfied by T .

If q ( T ) = 0 for q ( x ) ∈ F [ x ] then p1 ( x ) | q ( x ) ... (3)

If p ( x ) is the minimal polynomial for T over F then p ( T ) = 0 .

∴ p (T ) = 0 by (2)

⇒ p1 ( x) | p ( x ) by (3)

Hence proof.

Theorem :

If T ∈ A (V ) has all its charactristics roots in F, then there is a basis of V in which the matrix of
T is triangular.
Proof :
We prove this theorem by using induction on the dimensions of V over F.
If dimV = 1 then every element in A (V) is a scalar and also the theorem is true here.
Suppose that the theorem is true for all vectr spaces over F of dimension n–1 and let V be of
dimension n over F.

The linear transformation T on V has all its characteristic roots in F let λ1 ∈ F be a characteristic
root of T. There exists a non-zero vector v1 in V such that T ( v1 ) = λ1v1 .

Let W = {α v1 | α ∈ F } ; W is a one-dimensional subspace of V and is invarient under T..

Let V = V / W we know “If V is finite dimensional vector space and W be subspace of V then
dim V = dimV − dimW ”

∴ dim V = dimV − dimW = n −1

Also we know ‘If W ⊂ V is invariant under T then T induces a linear transformation T on
V/W defined by

T ( v + W ) = T (v) + W

Also, we know that, If T satisfies the polynomial q ( x ) ∈ F [ x ] then so does T . If p1 ( x ) is

the minimal polynomial for T over F and if p ( x ) is that for T then p1 ( x) | p ( x ) ”

56
 T induces a linear transformation T on V whose minimal polynomial over F divides the
minimal polynomial of T over F.
Thus all the roots of the minimal polynomial of T , being roots of the minimal polynomial of T,,
must lie in F.

The linear transformation T in its action on V satisfies the hypothesis of the theorem.

Since V is (n–1) dimensional over F, by our induction hypothesis there is a basis v2 ,..., vn of
over F such that

T ( v2 ) = α 22v2 , T ( v3 ) = α32v2 + α 33v3.......

T ( vi ) = α i 2v2 + α i3v3 + ..... + α ii vi .....T ( vn ) = α n2v2 + α n3v3 + ..... + α nnvn

Let v2 , v3 ,..., vn be elements of V mapping into v2 ,..., vn respectively..

Then v1, v2 ,...., vn form a basis of V..

[ Q v2 ,..., vn be a basis if they are lineary independent and v2 , v3 ,...., vn maps into these
elementry ∴ v2 , v3 ,...., vn linearly independent. Therefore we have if v1, v2 ,...., vn linearly
independent then Tv1, Tv2 ,...., Tvn linearly independent Now let v1, v2 ,...., vn and
α1,...., α n ∈ F such that α1v1 + .... + α nvn = 0 . i.e. α1 = 0 the this linearly independent we are throw

if not then v1 = −α1−1 (α 2v2 + .... + α nvn ) ⇒ v1 is the linear combination of

T ( v1 ) = v1 = −α1−1 (α 2v2 + .... + α nvn ) ∈W a contradiction to W is invariant under T and dim W = 1]

Since T ( v2 ) = α 22v2 ⇒ T ( v2 ) − α 22v2 = 0 ⇒ T ( v2 )− α 22v2 ∈ W

Thus T ( v2 ) − α 22 v2 is a multiple of v1 say α 21v1 yielding, after transforming

T ( v2 ) = α 21v1 + α 22 v2

Similalry, Tvi − α i 2 v2 − α i3v3......α ii vi ∈W

∴T ( vi ) − α i1v1 + αi 2 v2 + ..... + α ii vi

The basis v1,...., vn of V over F provides us with a basis where every T ( vi ) is a linear
combination of vi and its precedessors in the basis. Therefore the matrix of T in the basis is triangular..

57
Theorem :

If V is n-dimensional over F and if T ∈ A (V ) has all its characteristic roots in F, then TT

satisfies a polynomial of degree n over F.

Proof : By previous theorem we can find a basis v1,...., vn of V over F such that

T ( v1 ) = λ1v1 , T ( v2 ) = α 21v1 + λ2 v2 , ... T ( vi ) = α i1v1 +α i 2v2 + ..... + α ii vi for i = 1, 2, ..., n

Equivalenty ( T − λ1 ) v1 = 0 , ( T − λ2 ) v2 = α 2 1v1....

( T − λi ) vi = α i1v1 + .... + α ii −1vi −1 for i = 1, 2, ...., n

Now ( T − λ1 )( T − λ2 )( v2 ) = ( T − λ1 )(α 21v1 ) = α 21 (T − λ1 ) v1 = 0

also ( T − λ1 )(T − λ2 ) = ( T − λ2 )( T − λ1 )
Continuing this type of computation yields.

( T − λ1 )( T − λ2 ) ....( T − λi ) v1 = 0 , ( T − λ1 )(T − λ2 ) .... (T − λi ) v2 = 0

.... ( T − λ1 )(T − λ2 ) .... (T − λi ) vi = 0 for i = n

The matrix S = (T − λ1 )( T − λ2 ) ∈ ....( T − λn ) satisfies S ( v1 ) = S ( v2 ) = .... = S ( vn ) = 0 .

Then, since S annihilates a basis of V, S must annihilate all of V. Therefore S = 0. Consequently

T satisfies the polynomial ( x − λ1 )( x − λ2 ) .... ( x − λn ) in F [x] of degree n proves the theorem.

2) Nilpotent Transformations
Lemma :

If V = V1 ⊕ V2 ⊕ .... ⊕ Vn where each subspace Vi is of dimension ni and is invariant under T,

T,
an element of A (V), then a basis of V can be found so that the matrix of T in this basis is of the form

 A1 0 L 0
0 A2 L 0 

 M 
 
0 0 L An 

where each Ai is an ni × ni matrix and is the matrix of the linear transformation induced by T on Vi .

58
 (1) (1) (1) (2) (2) (2)
Choose a basis of V as follows : v1 , v2 ,...., vn1 is a basis of V1 , v1 , v2 ,...., vn2 is a basis
(n ) ( n ) (n )
of V2 and so on .... v1 , v2 ,...., vnn is a basis of Vn . Since each Vi is invariant under T.
T.

( )
T v j (i ) ∈ Vi so is a linear combination of v1 , v2 ,...., vni and only these. Thus the matrix of
(i ) (i ) (i )

T in the basis so chosen is of the desired form. That each Ai is the matrix of Ti , the linear transformation
induced on Vi by T is clear from the very definition of the matrix of a linear transformation.

Lemma :

If T ∈ A (V ) is nilpotent, then α 0 + α1T + .... + α mT m , when the α i ∈ F is invertible if α 0 ≠ 0 .

Proof :

If S is nilpotent and α 0 ≠ 0 ∈ F a simple computation shows that

 1 r −1 
S S2 r −1 S
(α 0 + S )  − 2 + 3 + .... + ( −1)  =1
α 0r 
if S r = 0
 α0 α0 α0

Now if T r = 0 , S = α1T + α 2T 2 + .... + α mT m also must satisfy S r = 0 .

S r = (α1T +α 2T 2 + .... + α mT m ) = 0
r
QT r = 0

Thus for α 0 ≠ 0 in F, α 0 + S is invertible.

Definition :

If T ∈ A (V ) is nilpotent then k is called index of nilpotent of T if T k = 0 but T k −1 ≠ 0 .

Theorem :

If T ∈ A (V ) is nilpotent of index of nilpotence n1 then a basis of V can be found such that

the matrix of T in this basis has of the form

 M n1 0 L 0 
 
 0 M n2 0 
 
 M O 
 0 0 L M nr 


where n1 ≥ n2 ≥ ..... ≥ nr and where n1 + n2 + ..... + nr = dimV .

59
Proof : T ∈ A (V ) is nilpotent with index of nilpotence n1 .

∴T n1 = 0 but T n1−1 ≠ 0
We can find a vector v ∈ V such that vT n1−1 ≠ 0 .

Let v, T ( v ) , T 2 ( v ) ,....,T n1 −1 (v ) be n1 vectors we claim that these are linearly independent

over F.

Suppose that these are linearly dependent i.e. there are scalars α1, α 2 ,...., α ni ∈ F not all
zero such that

α1v + α 2T ( v ) + .... + α n T n1−1 ( v ) = 0

i

Let α s be the first non-zero scalar, hence

( )
T s −1 α s + α s +1T + .... + α n1T n1 −s ( v ) = 0

We know “If T ∈ A (V ) is nilpotent then α 0 + α1T + .... + α mT m when the α i ∈ F is invertible

if α 0 ≠ 0 ”

∴α S ≠ 0 we have α s + α s +1T + .... + α n1T n1− s is invertible, and therefore T n1−1 ( v ) ≠ 0 .

Thus no such non-zero α s exists and v, T ( v ) , T 2 ( v ) ,....,T n1 −1 (v ) Linearly independent over

F.

Let V1 be the subspace of V spanned by v1 = v, v2 = T ( v ) ,...., vn1 = T n1−1 ( v ) .

∴V1 is invariant under T and have basis v1, v2 ,...., vn1 , T can be induces the linear transformation

of V1 . The matrix representation of T1 w.r.t. the above basis is M n1 .

So far we have produced the upper left-hand corner of the matrix of the theorem.

Now n2 ≤ n1 , ∴T n2 ≠ 0 for n1 ≠ n2 (if n1 = n2 then do above process)

We can find u ∈ V such that T n2 ( u ) ≠ 0 .

∴ claim, u, T ( u ) ,...., T n2 −1 (u ) linearly independent and spans V2 subspace of V and V2 is

invariant under T. Therefore T induce a linear map on V2 whose matrix representation is M n2 and so
on.
Similarly we can find other.
We can get basis for V and the matrix representation of the required form.

60
Lemma :

If u0 ∈ V1 is such that T n1− k ( u ) = 0 when 0 < K ≤ n1 then u = T k ( u ) for some u0 ∈ V1 .

Proof : Since u ∈ V ,

u = α1v + α 2T ( v ) + .... + α k T k −1 ( v ) + α k +1T k ( v ) + .... + α n T n1 −1 ( v )

1

Thus 0 = T n1− k ( u ) = α1T n1−k ( v ) + .... + α k T n1 −1 ( v )

However T n1− k ( v ) ,..., T n1 −1 ( v ) are linearly independent over F

Whence α1 = α 2 = .... = α k = 0 and so u = α k +1T k ( v ) + .... + α n1T n1 −1 ( v ) =T k (u 0 )

When u0 = α k +1v + .... + α n1T n1− k −1 ( v ) ∈ V . Hence the proof.

Lemma :

There exists a subspace W of V, invariant under T, such that V = V1 ⊕ W .

Proof :
Let W be a subspace of V of largest possible dimension such that

(1) V1 I W = ( 0 ) (2) W is invariant under T.

First we show that V = V1 + W .

Suppose not, then there exist an element z ∈ V such that z ∉ V1 + W .

Since T n1 = 0 , there exists an integer k, 0 < k < n, such that T k ( z ) ∈V 1 +W and such that

T i ( z) ∈V1 + W for i < k.

Thus T k ( z) = u + w when u ∈ V1 and where w ∈ W . But then

0 = T k ( z) = T
n1 − k
(T k ( z ) ) = T n1− k ( u ) + T n1− k ( w)
However, since both V1 and W are invariant under T, T n1− k ( u ) ∈ V1 and T ni − k ( w ) ∈ W .

Now since V1 I W = ( 0 ) this leads to

T n1− k (u ) = − Tn1− k ( w) V1 I W = ( 0) ⇒ T ni −k ( u ) = 0

We know “If u ∈ V1 is nsuch that T n1− k u = 0 where 0 < k ≤ n1 then u = T k u0 for some
u0 ∈ V1 .”

61
 We have T k ( u0 ) = u for some u0 ∈ V1 .

Therefore T k ( z) = u + w = T k u0 + w

Let z1 = z − u0 then T k ( z1 ) = T k ( z ) −T k ( u0 ) = w ∈ W and since W is invariant under T

this yields T m ( z1 ) ∈W for all m ≥ k .

On the other hand if i < k, T i ( z1 ) = T i ( z ) − T i ( u0 ) ∉ V1 + W for otherwise T i ( z ) must fall

in V1 + W comtradicting the choice of k.

Let W1 be the subspace of V spanned by W and z1 , Tz1,...., T k −1z1 , Q z1 ∉ W and since

W ⊂ W1 , the dimension of W1 must be larger than that of W, moreover, since T k ( z1 ) ∈W and since
W is invariant under T, W1 must be invariant under T..

By the maximal nature of W, there must be an elements of the form

w0 + α1z1 + α2T ( z1 ) + .... + α k T k −1 ( z1 ) ≠ 0 ∈W1 I V1

Where w0 ∈ W . Not all of α1,...., α k can be 0, otherwise we would have

0 ≠ w0 ∈ W I V1 = ( 0) , a contradiction.

Let α s be the first non-zero α s , then w0 + T s −1 (α s + α s +1T + ... + α kT k − s ) ( z1 ) ∈ V1 .

Since α s ≠ 0 ⇒ α s + α s +1T + .... + α k T k − s is invertible and its inverse say R is a polynomial

in T.

Thus W and V1 are invariant under R.

However from the above R ( w0 ) + Tk −1 ( z1 ) ∈ R (V1 ) ⊂ V1

Forcing T s −1 ( z1 ) ∈V1 + R (W ) ⊂ V1 + W since s − 1 < k this is impossible, therefore

V = V1 + W and Because V1 I W = ( 0 ) , V = V1 ⊕ W .

Hence the proof.

Problem : Let V = V1 ⊕ W where W is invariant under T when we can find a basis of V, so that

 M n1 0
matrix representation of T in this basis is of the form  .
 0 A2 

62
Definition : The integers n1, n2 ,....., nr are called the invariants of T..

Definition : If T ∈ A (V ) is nilpotent the subspace M of V, if dimension m which is invariant under T,,

is called cyclic with respect to T if

1) T m ( M ) = ( 0 ) , T m−1 ( M ) ≠ ( 0 )

2) There is an element z ∈ M such that z , T ( z ) ,....,T m−1 ( z ) form a basis of M.

Lemma : If M of dimension m, is cyclic with respect to T, then the dimension of T k ( M ) is m–k for
all k ≤ m .

Proof : Let M of dimension m is cyclic w.r.t. T consider z , T ( z ) , T 2 ( z ) ,....,T m −1 (z ) be basis of M.

∴T k ( z ) ,T k +1 ( z ) ,T k + 2 ( z ) ,..., T m+ k −1 ( z ) ∈T k ( M ) (be basis of ∈ )

But M is cyclic w.r.t. T i.e. T m ( m ) = 0 i.e. T m ( z ) = 0 and T m−1 ( z ) ≠ 0

∴T k ( z ) , T k +1 ( z ) ,..., T m−1 ( z ) be basis of T k ( M ) .

Therefore m – k elements ∴ dim (T k ( M ) ) is m – k, ∀k ≤ m .

Example : For a nilpotent T in A (V). Find integers n1 ≥ n2 ≥ ..... ≥ nr and subspaces V1, V2 ,...., Vr of
V cyclic with respect to T and of dimensions n1, n2 ,....., nr respectively such that V = V1 ⊕ ..... ⊕ V2 .
Show that these are unique integers.

Solution : Let T ∈ A (V ) is nilpotent, we suppose index of nilpotent n1.

∴T n1 = 0 and T n1−1 ≠ 0

Then v ∈ V , v, T ( v ) , T 2 ( v ) ,....,T n1 −1 (v ) is Linearly independent set in V form a subspace

of V generated by these element say V1 and dim V1 = n1.

Not let u ∈ V and u ∉ V1 and n2 ≤ n1 be integer ∴T n2 −1 ≠ 0 .

And u, T ( u ) , T 2 ( u ) ,...., T n 2 −1 (u ) is linearly independent set form a subspace V2 of V

generated by these elements.
Continue this process until we cover V.

63
 Suppose at the last we get V1, V2 ,...., Vr be subspace each invariant under T and

Vi I V j = ( 0) , ∀i ≠ j and V = V1 + V2 + .... + Vr ⇒ V = V1 ⊕ ..... ⊕ Vr

Now we shows these integer are unique.

Suppose there are other integers m1 ≥ m2 ≥ .... ≥ ms and subspace U1 ,....,U s of V cyclic
w.r.t. to T and of dimensions m1 ,...., ms respectively such that V = U1 ⊕ ..... ⊕ Us .

Claim :

s = r and m1 = n1,...., mr = nr

Suppose that this were not the case then there is a first integer i such that mi ≠ ni we may
assume that mi < ni .

Consider T mV , therefore for V = V1 ⊕ ..... ⊕ Vr we have

T mi V = T miV1 ⊕ .... ⊕ T mi Vr , ∴ dim T mi V j = n j..... mi , j = 1 ..... r

Therefore above

∴ dim (T i V ) ≥ ( n1 − mi ) + ( n2 − mi ) + .... +( ni − mi )
m
.... (1)

and for V = U1 ⊕ ..... ⊕ Us and since T mi (U j ) = ( 0 ) for j ≥ i

∴T mi (V ) = T miU1 ⊕T miU 2 ⊕ ...... ⊕ T mi Ui −1

Thus dim T
mi
(V ) = ( m1 − mi ) + ( m2 − mi ) + .... + ( mi −1 − mi )

= ( n1 − mi ) + ( n2 − mi ) + .... + ( ni−1 − mi ) Q n j = m j for j < i

Contradict to the equation (1) Q ni − mi > 0 thus there is unique set of integers
n1 ≥ n2 ≥ ..... ≥ nr such that V = V1 ⊕ ..... ⊕ Vr .
Equivalently we have shown that the invariants of T are unique.

Theorem : Two nilpotent linear transformation are similar if and only if they have the same invariants.
Proof : The above Example has proved that if the two nilpotent linear transformations have different
invariants, then they cannot be similar for their respective matrices.

64
  M n1   M m1 
   
 M n2   M m2 
  and  
 O   O  cannot be similar.
 M nr   M mr 
 

In the other direction, if the two nilpotent linear transform S and T have the same invariants
n1 ≥ ..... ≥ nr .

We know the result “If T ∈ A (V ) is nilpotent of index of nilpotent n1 then basis of V can be

 M n1 
 
 O  when n ≥ ..... ≥ n and
found such that the matrix of T in this basis has of the form
 M nr 
1 r

n1 + n2 + ..... + nr = dimV .”

Therefore there are basis v1, v2 ,.....vn and w1,.....wn of V such that the matrix of S in v1,.....vn

 M n1 
 
 O .
and that of T in w1 ,....., wn are each equal to
 M nr 


But of A is the linear transformation defined on V by A ( vi ) = wi then S = ATA−1 (Prove).

Hence S and T are similar.

 0 1 1
 
Example : Let m (T ) =  0 0 0  ∈ F3 . Find similar matrix. T act on F (3) .
 0 0 0
 

i.e. Find A such that ATA−1 = S .

Basis of F (3) is u1 = (1,0,0) , u2 = ( 0,1,0) , u3 = ( 0,0,1)

Let v1 = u1 , v2 = T ( u1 ) = u2 + u3 , v3 = u3
w.r.t. this basis

 1 0 0 0 1 1  1 0 0
 1 0 0
ATA =  0 1 1 0 0 0  0 0 0
−1
A =  0 1 1  and    
   0 0 1 0 0 0  0 1 1
 0 0 1 

65
A Decomposition of V : Jordan Form
Example : Let V be finite dimensional vector space over F, T ∈ A (V ) , V subspace of V invariant
under T. T induces a linear transformation T1 on V1 defined by T1 ( u ) = T ( u ) for every u ∈ V1 . Show
that for any polynomial q ( x ) ∈ F [ x ] , the linear transformation induced by q ( T ) on V1 is precisely
q ( T1 ) . In particular q ( T ) = 0 then q ( T1 ) = 0 .

Let V be finite dimensional vector space over F. T ∈ A (V ) , V1 ⊆ V invariant under T.

T.

Therefore T induces a linear transformation T1 on V1 defined by T1 ( u ) = T ( u ) for every

u ∈ V1 .

Let q ( x ) ∈ F [ x ] be any polynomial such that q ( T ) = 0 .

p ( x ) minimal polynomial for T and p1 ( x ) is minimal for T.

p ( x ) | q ( x ) we know that p1 ( x) | p ( x ) = p1 ( x ) | q ( x ) ⇒ q (T1 ) = 0

Lemma : Suppose that V = V1 ⊕ V2 where V1 and V2 are subspaces of V, invariant under T. Let T1
and T2 be the linear transformations induced by T on V1 and V2 respectively. If the minimal polynomial
of T1 over F is p1 ( x ) while that of T2 is p2 ( x ) , then the minimal polynomial for T over F is the least
common multiple of p1 ( x ) and p2 ( x ) .

Proof : If p ( x ) is the minimal polynomial for T over F, as we know above example both p ( T1 ) and
p ( T2 ) are zero, whence p1 ( x) | p ( x ) and p2 ( x ) | p ( x ) . But then the least common multiple of
p1 ( x ) and p2 ( x ) must also divide p ( x ) .

On the other hand if q ( x ) is the least common multiple of p1 ( x ) and p2 ( x ) , consider

q ( T ) . For u1 ∈ V1 since p1 ( x ) | q ( x )

q ( T ) ( v1 ) = q ( T1 )( v1 ) = 0 similarly for v2 ∈ V 2

q ( T ) ( v2 ) = q ( T2 )( v2 ) = 0 Given any u ∈ V , v can be written as cos θ = ( v1 + v2 ) when

v1 ∈ V1 , v2 ∈ V2 in consequence of which

q ( T ) ( v ) = q (T ) ( v1 + v2 ) = q ( T ) ( v1 ) + q ( T ) ( v2 ) = 0

Thus q ( T ) = 0 and T satisfies q ( x ) . ∴ p ( x ) | q ( x ) .

⇒ p ( x) = q ( x)

66
Corollary :
If V = V1 ⊕ ..... ⊕ Vk where each Vi is invariant under K and T if pi ( x ) is the minimal
polynomial over F of Ti the linear transformation induced by T on Vi then minimal polynomial of T
over F is least common multiple of p1 ( x) , p2 ( x ) ,..., pk ( x ) .

Theorem :

For each i = 1, 2, ...K, Vi ≠ ( 0 ) and V = V1 ⊕V2 ⊕..... ⊕Vk .

The minimal polynomial of Ti is qi ( x )li .

Proof : We prove this result using induction on k.

If k = 1 the V = V1 and there is nothing that needs proving suppoose then that k > 1 and

p ( x ) = q1l1 ( x ) q2l2 ( x ) .......qlk k ( x )

We first want to prove that each V1 ≠ ( 0 ) . We introduce the k polynomials

l lk
h1 ( x ) = q2 ( x ) 2 ....qk ( x )

l l lk
h2 ( x ) = q1 ( x ) 1 q3 ( x ) 3 ....qk ( x )

M
lj
hi ( x ) = ∏ q j ( x )
i≠ j

M
l l k −1
hk ( x ) = q1 ( x ) 1 ....qk −1 ( x )

Since k > 1, hi ( x ) = p ( x ) whence hi (T ) ≠ 0 , thus, given i, there is a v ∈ V such that

w = hi (T ) v ≠ 0 . But

l l
qi (T ) i ( w) = qi (T ) i ( hi (T ) v ) = p ( T ) ( v ) = 0

In consequence w ≠ 0 is in Vi and so Vi ≠ ( 0 ) . In fact, we have shown that hi (T )V ≠ 0 is

l
in Vi and if v j ∈ V j for j ≠ i since q j ( x ) j | hi ( x ) .

hi ( T) ( v j ) = 0

67
 The polynomials h1 ( x ) , h2 ( x ) ,...., hk ( x ) are relatively prime.

We know that “taken two polynomials f ( x ) , g ( x ) in F [ x ] they have a greatest common

divisor d ( x ) which can be realized as d ( x ) = λ ( x ) f ( x ) + µ ( x ) g ( x ) .”

We can find polynomials a1 ( x ) , a2 ( x ) ,...., ak ( x ) in F [ x] such that

a1 ( x ) h1 ( x ) + .... + ak ( x ) hk ( x ) = 1 .

From this we get a1 (T ) h1 (T ) + .... + ak (T ) hk (T ) = 1 whence, given v ∈ V ,

v = v ⋅ 1 = ( a1 ( T ) ) h1 (T ) + .... + ak ( T ) hk ( T ) ( v )

= a1 (T ) h1 ( T ) ( v ) + .... + ak ( T ) hk ( T) ( v )

Now, each ai (T ) hi ( v ) is in hi (T )V and since we have shown above that hi (T ) V ⊂ Vi , we

have now exhibited v as v = v1 + .... + vk when each vi = ai (T ) h1 (T ) ( v ) is in Vi .

Thus Vi = V1 + V2 + .... + Vk .

We must now verify that this sum is a direct sum. To show this, it is enough to prove that if
u1 + u2 + .... + uk = 0 with each ui = 0 . So suppose that u1 + u2 + .... + uk = 0 and that some ui say

u1 is not 0 apply h1 ( T ) we obtain h1 ( T ) ( u1 ) + .... + h1 ( T ) ( uk ) = 0 .

However h1 ( T ) ( v j ) = 0 for j ≠ i since u j ∈ V j , the equation reduced to h1 ( T ) ( u1 ) = 0 .

But q1 ( T ) l1 ( u1 ) = 0 and since h1 ( x ) and q1 ( x ) are relatively prime we are led to u1 = 0 which is
of course in consistent with assumption that u1 ≠ 0 .

⇒ V = V1 ⊕V2 ⊕..... ⊕Vk

Now prove that the minimal polynomial of Ti on Vi is qi (T ) li .

l
By definition of Vi , since qi (T ) li Vi = 0 , qi (Ti ) i = 0 whence the minimal equation of Ti

must be divisor of qi ( x ) li thus of the formqi ( x ) f i with f i ≤ l i . By “Corollary above”.

The minimal polynomial of T over F is the least common multiple of qi ( x ) fi ,...., qk ( x ) f k and

so must be qi ( x ) fi ,...., qk ( x ) f k . Since this minimal polynomial is in fact qi ( x ) li ,...., qk ( x ) l k we

must have that, f i ≥ l i ,...., f k ≥ l k ⇒ l i = fi for i = 1, 2, ....., k and so qi ( x ) li is minimal polynomial

for Ti .

Hence the proof.

68
Note : If all the characteristic roots of T should happen to lie in F, then the minimal polynomial of T
l lk
takes on the especially nice form q ( x ) = ( x − λ1 ) 1 .... ( x − λk ) where λ1 ,...., λk are the distinct

characteristic roots of T. The irreducible factors qi ( x ) are x − λi . Noe that on Vi , Ti has only λi as
a characteristic root.

Corollary :

If all the distinct characteristic roots λi ,...., λk of T lie in F, then V can be written as

{ l
}
V = V1 ⊕ ..... ⊕ Vk where Vi = u ∈ V | (T − λi ) i v = 0 and where Ti has only one characteristic

root λi on Vi .

Note :

QV = V1 ⊕ ..... ⊕ Vk if dimVi = ni then we can find a basis of V such that in this basis the

 A1 
 A2 
 
matrix of T is of the form  O  where each Ai is an ni × ni matrix and is in fact the
 
 Ak 

matrix of Ti .

0 1 0 L 0 0
 
0 0 1 M M
= Mn
Notation  M 0 1
 0 0 
L 0  n×n


λ 1 0 L 0
0 λ 1 L 0 

M 
Definition : The matrix  0 
 0 0 L 1  with λ ' s on the diagonal, 1 is on the super diagonal
0 0 L λ 
 0

and 0’s elsewhere is a basic Jordan block belonging to λ .

69
Theorem : Let T ∈ Ap (V ) have all its distinct roots, λi ,...., λk in F. Then a basis of V can be fund in

 J1   Bi1 
 J2   
  Bi 2
which the matrix T is of the form  O  where each J i =   and
O 
 
J k   
  Bir 
where Bi1,....Bi ri are basic Jordan blocks belonging to λi .

Proof : Note that an m × m basic Jordan block belonging to λ is merely λ + M m , where,

0 1 0 L 0 0
 
0 0 1 L 0 0
Mm =  M 
 
0 0 0 L 0 1
0 0 0 L 0 0  m×m


We know that “If V = V1 ⊕V2 ⊕..... ⊕Vk where each subspace Vi is of dimension ni and is
invariant under T, an element of A (V), then a basis of V can be found so that the matrix of T in this

 A1 0 L 0 
0 A2 L 0 

basis is of the form  M .
 
0 0 L Ak 

Where each Ai is an ni × ni matrix and is the matrix of the linear transformation induced by T
on Vi .”

Also we know that “If all the distinct characteristic roots λi ,...., λk of T lie in F, then V can be

{ l
}
written as V = V1 ⊕V2 ⊕..... ⊕Vk where Vi = v ∈ V | ( T − λi ) i ( v ) = 0 and where Ti has only

one charactristic root, λi on Vi .”

Therefore, we can reduce to the case when T has only one characteristic root λ , that is T − λ
is nilpotent.

70
 Thus T = λ + (T − λ ) and since T − λ is nilpotent, there is a basis in which its matrix is of the

 M n1 0 
 
 O 
form
 0 M nr 


Therefore, “If T ∈ A (V ) is nilpotent of index of nilpotence n1 , then a basis of V can be found

 M n1 0 L 0 
 
 0 M n2 L 0 
such that the matrix of T in this basis has the form  M .

 0 0 L M nr 


Where n1 ≥ n2 ≥ ..... ≥ nr and n1 + n2 + ..... + nr = dimV .

But then the matrix of T of the form

λ  M
  n1   Bn1 
λ    
 +  O = O
 
 O    
  M nr   Bnr 
 λ  

Using the first remark made in this proof about the relation of a basic Jordan block and the
Mm’s we have the required.

Example :

 −2 5 1 0
 −2 4 1 0
A=
1)  −1 2 1 0
 
 −1 2 0 1

Find Jordan form

Solution : The characteristic equation is ( x − 1)4 .

rank of A − λ I is 2.
Therefore, geometric multiplicity of equation is 2.
Hence there are two Jordan blocks.

of the form dia ( J 2 (1), J 2 (1) ) or diag ( J 3 (1), J1 (1) )

71
 The minimal polynomial is ( x − 1) 3

Therefore Jordan form diag ( J 3 (1), J1 (1) )

Example : Let T ∈ A (V ) and F [ x ] ring of polynomials in x over F and define for any f ( x ) in

F [ x ] , v ∈ V , f ( x ) v = f (T ) v . Prove that V is a module over F [ x ] .

[Let R ring M ≠ φ is said to be an R-module if M is an abelian group under operation t such
that r ∈ R and m ∈ M there exists an element m ∈ M subject to (i) r ( a + b ) = ra + rb
(ii) r ( Sa ) = ( rS ) a (iii) ( r + S ) a = ra + Sa , ∀a, b ∈ M and r1, S ∈ R ]

Example : F [ x ] is a Eucliaden ring. V is finitely generated module are F [ x ] .

V is the direct sum of a finite number of cyclic subv modules.
(On R-module M is said to be finitely generated if there exists element a1.... an ∈ M such that
every m ∈ M is of the form m = T1a1 + .... + Tn an .

M-cyclic if there is an element m0 ∈ M such that every m ∈ M is of the form m = α m0 for

α ∈R.

Problems :

1 1 1
 
1. Prove that the matrix  −1 − 1 − 1 is nilpotent, and find its invariants and Jordan form.
1 1 0
 

2. Find all possible Jordan forms for all 8 × 8 matrices having x 2 ( x − 1) 3 as minimal polynomial.

3. If the multiplicity of each characteristic root of T is 1, and if all the characteristic roots of T are
in F, prove that T is diagonalizable over F.

❏❏❏

72