M304T: Linear Algebra
Canonical Form
                  Triangular form
                Dr. R. Bhuvaneswari
                 BMS College for Women
                   October 2020
RB (BMSCW)               M304T           October 2020   1 / 35
Lemma
If W Ă V is invariant under T , then T induces a linear
transformation T on V {W defined by
                   pv ` W qT “ vT ` W              for v P V .
If T satisfies a polynomial qpx q in F rx s then so does T .
((i.e) T also satisfies qpx q).
If p1 px q is a minimal polynomial for T over F and ppx q is a minimal
polynomial for T over F , then
                                  p1 px q|ppx q.
      RB (BMSCW)                      M304T                      October 2020   2 / 35
Proof:
Let V “ V {W .
The elements of V are the cosets v ` W of W in V .
Given v P V , v “ v ` W for some v P V .
Define T : V Ñ V by
                          v T “ vT ` W
for v “ v ` W P V .
Now we prove that T is a linear transformation on V .
     RB (BMSCW)                 M304T                   October 2020   3 / 35
T is well-defined:
             Let v 1 “ v1 ` W and v 2 “ v2 ` W be two elements in
             V such that v 1 “ v 2 .
              ùñ v1 ` W “ v2 ` W
              ùñ v1 ´ v2 P W
              ùñ pv1 ´ v2 qT P WT Ă W      p7 W is invariant under T q
              ùñ v1 T ´ v2 T P W
              ùñ v1 T ` W “ v2 T ` W
              ùñ v 1 T “ v 2 T .
             Thus T is well-defined.
     RB (BMSCW)                    M304T               October 2020   4 / 35
T is a linear transformation:
             Let v 1 “ v1 ` W and v 2 “ v2 ` W be two elements in
             V . Then
                  pv 1 ` v 2 qT “ pv1 ` W ` v2 ` W qT
                             “ pv1 ` v2 ` W qT
                             “ pv1 ` v2 qT ` W
                             “ v1 T ` v2 T ` W p7 T is linearq
                             “ v1 T ` W ` v2 T ` W
                             “ v 1T ` v 2T .
             Thus T preserves addition.
     RB (BMSCW)                 M304T                   October 2020   5 / 35
             For α P F and Let v “ v ` W P V ,
                       pαv qT “ pαv ` W qT
                               “ pαv qT ` W
                               “ vT α ` W p7 T is linearq
                               “ pvT ` W qα
                               “ v T α.
             Thus T preserves scalar multiplication.
Hence T is a linear transformation on V .
     RB (BMSCW)                  M304T                 October 2020   6 / 35
Now we prove that for any polynomial qpx q P F rx s, qpT q “ qpT̄ q.
Let v “ v ` W P V . Then
                         v̄ T 2 “ vT 2 ` W
                              “ pvT qT ` W
                              “ pvT ` W qT
                              “ pv T qT
                                    2
                              “ vT .
Therefore,
                             T 2 “ pT q2 .
     RB (BMSCW)                  M304T                    October 2020   7 / 35
Similarly, we can get
                                   T k “ pT qk
for any integer k ě 0.
Let qpx q “ a0 ` a1 x ` a1 x 2 ` . . . ` ak x k P F rx s. Then
                   qpT q “ a0 ` a1 T ` a1 xT 2 ` . . . ` ak T k
            ùñ qpT̄ q “ a0 ` a1 T̄ ` a1 pT̄ q2 ` . . . ` ak pT̄ qk
              ùñ qpT̄ q “ a0 ` a1 T̄ ` a1 T 2 ` . . . ` ak T k
                              ùñ qpT̄ q “ qpT q.
      RB (BMSCW)                      M304T                       October 2020   8 / 35
If T satisfies a polynomial qpx q P F rx s then qpT q “ 0 implies that
                         qpT̄ q “ qpT q “ 0 “ 0.
Thus T̄ also satisfies qpx q.
Given : p1 px q is the minimal polynomial for T over F and ppx q is the
minimal polynomial for T over F .
Then p1 pT q “ 0 and ppT q “ 0.
      RB (BMSCW)                  M304T                    October 2020   9 / 35
Since ppT q “ 0, we have ppT̄ q “ 0.
Note that p1 px q is the minimal polynomial for T̄ and ppT̄ q “ 0.
Thus p1 px q|ppx q.
      RB (BMSCW)                 M304T                    October 2020   10 / 35
Note
All the characteristic roots of T which lie in F are roots of the
minimal polynomial of T over F .
Therefore, We say that all the characteristic roots of T are in F if all
the roots of the minimal polynomial of T over F lie in F .
       RB (BMSCW)                M304T                    October 2020   11 / 35
Theorem
If T P ApV q has all its characteristic roots in F , then there is a basis
of V in which the matrix of T is triangular.
Proof: We prove this theorem by induction on the dimension of V
over F .
If dimpV q “ 1, then every element in ApV q is a scalar, and so the
theorem is true.
Assume that the theorem is true for all vector spaces over F of
dimension n ´ 1.
      RB (BMSCW)                   M304T                    October 2020   12 / 35
Let V be a vector space of dimension n over F .
Let the linear transformation T on V has all its characteristic roots
in F .
We will prove that there is a basis of V in which matrix of T is
triangluar.
         RB (BMSCW)              M304T                   October 2020   13 / 35
Let λ1 P F be a characteristic root of T .
Then there exists a nonzero vector v1 in V such that
                             v1 T “ λ1 v1 .
Let W “ tαv1 : α P F u.
Then W is a one-dimensional subspace of V generated by v1 .
Note that WT Ă W .
Therefore, W is invariant under T .
     RB (BMSCW)                  M304T                 October 2020   14 / 35
Let V “ V {W . Then
                  dimpV q “ dimpV q ´ dimpW q “ n ´ 1.
By the above lemma, the linear transformation T on V induces a
linear transformation T on V̄ defined by
                         pv ` W qT̄ “ vT ` W
such that the minimal polynomial of T̄ divides the minimal
polynomial of T .
     RB (BMSCW)                   M304T                  October 2020   15 / 35
Given all the characteristic roots of T are in F .
Then all the roots of minimal polynomial for T lie in F .
Thus all the roots of minimal polynomial for T̄ (which are roots of
minimal polynomial of T ) lie in F and hence all the characteristic
roots of T̄ also lie in F .
Since dimpV̄ q “ n ´ 1 and the linear transformation T̄ on V̄ has all
its roots in F , by induction hypothesis, there exists a basis
v 2 , v 3 , ¨ ¨ ¨ , v n of V̄ over F such that the matrix of T̄ in the basis is
triangular.
      RB (BMSCW)                      M304T                      October 2020   16 / 35
Further
                  v 2 T “ α22 v 2
                  v 2 T “ α32 v 2 ` α33 v 3
                       ..
                        .
                  v i T “ αi2 v 2 ` αi3 v 3 ` ¨ ¨ ¨ ` αii v i
                       ..
                        .
                  v n T “ αn2 v 2 ` αn3 v 3 ` ¨ ¨ ¨ ` αnn v n
     RB (BMSCW)                       M304T                     October 2020   17 / 35
Since v i P V̄ “ V {W , D vi P V such that v i “ vi ` W .
Let v2 , v3 , . . . , vn be the elements of V mapping into v 2 , v 3 , ¨ ¨ ¨ , v n
respectively.
claim: v1 , v2 , . . . , vn is a basis for V .
Let α1 v1 ` α2 v2 ` . . . ` αn vn “ 0 where αi P F .
Then
            W “ 0 ` W “ pα1 v1 ` α2 v2 ` . . . ` αn vn q ` W
       ùñ α1 pv1 ` W q ` α2 pv2 ` W q ` . . . ` αn pvn ` W q “ W
(ie) α2 v 2 ` . . . ` αn v n “ 0 “ W , the zero element of V {W .
( Note that α1 v1 P W .)
       RB (BMSCW)                       M304T                        October 2020   18 / 35
Since v 2 , v 3 , ¨ ¨ ¨ , v n form a basis for V over F , we have
                        α2 “ 0, α3 “ 0, . . . , αn “ 0.
Now α1 v1 ` α2 v2 ` . . . ` αn vn “ 0 implies that
                                   α1 v1 “ 0
Since v1 ‰ 0, we get α1 “ 0.
      RB (BMSCW)                      M304T                         October 2020   19 / 35
Thus α1 v1 ` α2 v2 ` . . . ` αn vn “ 0 implies
                        α1 “ 0, α2 “ 0, . . . , αn “ 0.
Hence v1 , v2 , . . . , vn is linearly independent vectors in V .
Since dimpV q “ n, v1 , v2 , . . . , vn forms a basis for V over F .
Now we prove that in this basis the matrix of T is triangular.
      RB (BMSCW)                     M304T                          October 2020   20 / 35
Since v 2 T “ α22 v 2 , we have
                     v2 T ` W ´ α22 pv2 ` W q “ 0
                         ùñ v2 T ´ α22 v2 P W
Since the elements of W are scalar multiples of v1 , we have
                            v2 T ´ α22 v2 “ α21 v1
for some α21 P F .
6, v2 T “ α21 v1 ` α22 v2
      RB (BMSCW)                    M304T               October 2020   21 / 35
Similarly,in general, for any i ě 2,
                      v i T “ αi2 v 2 ` αi3 v 3 ` ¨ ¨ ¨ ` αii v i
implies that
                    vi T ´ αi2 v2 ´ αi3 v3 ´ . . . ´ αii vi P W
                ùñ vi T ´ αi2 v2 ´ αi3 v3 ´ . . . ´ αii vi “ αi1 v1
for αi1 P F .
                ùñ vi T “ αi1 v1 ` αi2 v2 ` αi3 v3 ` . . . ` αii vi
      RB (BMSCW)                         M304T                      October 2020   22 / 35
Since W is invariant under T , v1 P W ùñ v1 T P W
                              ùñ v1 T “ α11 v1
Thus v1 , v2 , . . . , vn is a basis of V such that
                   v1 T “ α11 v1
                   v2 T “ α21 v1 ` α22 v2
                       ..
                        .
                    vi T “ αi1 v1 ` αi2 v2 ` ¨ ¨ ¨ ` αii vi
                        ..
                         .
                   vn T “ αn1 v1 ` αn2 v2 ` ¨ ¨ ¨ ` αnn vn
      RB (BMSCW)                      M304T                   October 2020   23 / 35
That is, vi T is a linear combination only of vi and its predecessors in
the basis.
Thus the matrix of T in this basis v1 , v2 , . . . , vn is triangular.
Therefore, the theorem is true for any vector space V of dimension n.
Hence the theorem is true for any finite dimensional vector space.
      RB (BMSCW)                     M304T                       October 2020   24 / 35
Above theorem in terms of matrices.
Theorem
Suppose the matrix A P Fn has all its characteristic roots in F . Then
there is a matrix C P Fn such that CAC ´1 is a triangular matrix.
Proof: Assume the matrix A P Fn has all its characteristic roots in F .
Then A defines a linear transformation T in F pnq whose matrix in the
basis
 v1 “ p1, 0, 0, ¨ ¨ ¨ , 0q, v2 “ p0, 1, 0, ¨ ¨ ¨ , 0q, ¨ ¨ ¨ , vn “ p0, 0, 0, ¨ ¨ ¨ , 1q
is precisely A.
        RB (BMSCW)                       M304T                           October 2020   25 / 35
Note that the characteristic roots of T are the characteristic root of
A.
Therefore, all the characteristic roots of T are in F .
By the above theorem, there is a basis of F pnq in which the matrix of
T , mpT q is triangular.
Thus A is a matrix of T in one basis and mpT q is a matrix of T in
another basis.
By change of basis theorem, there exists an invertible matrix C P Fn
such that
                           mpT q “ CAC ´1 .
Thus there exists C P Fn such that CAC ´1 is triangular.
      RB (BMSCW)                  M304T                    October 2020   26 / 35
Note
If a linear transformation T (or a matrix A) is brought to triangular
form over F , then the characteristic roots of T (or A) are the main
diagonal entries on the triangular form.
       RB (BMSCW)                M304T                   October 2020   27 / 35
Caylety Hamilton Theorem
Theorem
If V is an n´ dimensional vector space over F and if T P ApV q has
all its characteristic roots in F , then T satisfies a polynomial of
degree n over F .
Proof: Since T has all its characteristic roots in F , we can find a
basis v1 , v2 , . . . , vn is a basis of V over F in which matrix of T is
triangular.
      RB (BMSCW)                    M304T                      October 2020   28 / 35
That is, we can find a basis v1 , v2 , . . . , vn of V over F such that
                   v1 T “ α11 v1
                   v2 T “ α21 v1 ` α22 v2
                       ..
                        .
                   vi T “ αi1 v1 ` αi2 v2 ` ¨ ¨ ¨ ` αii vi
                       ..
                        .
                   vn T “ αn1 v1 ` αn2 v2 ` ¨ ¨ ¨ ` αnn vn
We know that in triangular form, the characteristic roots are the
entries in the main diagonal.
      RB (BMSCW)                     M304T                    October 2020   29 / 35
Let λ1 , λ2 , ¨ ¨ ¨ , λn be the characteristic roots of T . Then λi “ aii for
each i “ 1 to n.
Therefore, we get
         v 1 T “ λ1 v 1
         v2 T “ α21 v1 ` λ2 v2
             ..
              .
         vi T “ αi1 v1 ` αi2 v2 ` ¨ ¨ ¨ ` αipi´1q vi´1 ` λi vi
             ..
              .
         vn T “ αn1 v1 ` αn2 v2 ` ¨ ¨ ¨ ` αn1 pn ´ 1qvn´1 ` λn vn
      RB (BMSCW)                    M304T                        October 2020   30 / 35
Which implies that
                     v1 T ´ λ1 v1 “ 0
                     pi.eqv1 pT ´ λ1 q “ 0
        v2 pT ´ λ2 q “ α21 v1
                   ..
                    .
         vi pT ´ λi q “ αi1 v1 ` αi2 v2 ` ¨ ¨ ¨ ` αipi´1q vi´1
                    ..
                     .
        vn pT ´ λn q “ αn1 v1 ` αn2 v2 ` ¨ ¨ ¨ ` αn1 pn ´ 1qvn´1
     RB (BMSCW)                     M304T                        October 2020   31 / 35
Since v1 pT ´ λ1 q “ 0, we have
               v2 pT ´ λ2 qpT ´ λ1 q “ α21 v1 pT ´ λ1 q “ 0.
We know that
                  pT ´ λ1 qpT ´ λ2 q “ pT ´ λ2 qpT ´ λ1 q
Therefore,
             v1 pT ´ λ2 qpT ´ λ1 q “ v1 pT ´ λ1 qpT ´ λ2 q “ 0
     RB (BMSCW)                    M304T                       October 2020   32 / 35
Continuing this procedure, we get for i “ 1 to n,
                   v1 pT ´ λi qpT ´ λi´1 q ¨ ¨ ¨ pT ´ λ1 q “ 0
                   v2 pT ´ λi qpT ´ λi´1 q ¨ ¨ ¨ pT ´ λ1 q “ 0
                                                         ....
                                                          ..
                   vi pT ´ λi qpT ´ λi´1 q ¨ ¨ ¨ pT ´ λ1 q “ 0
For i “ n, the matrix S “ pT ´ λn qpT ´ λn´1 q ¨ ¨ ¨ pT ´ λ1 q satisfies
v1 S “ v2 S “ . . . “ vn S “ 0.
      RB (BMSCW)                      M304T                      October 2020   33 / 35
Therefore, S annihilates all the vectors in the basis of V and so S
must annihilate all elements of V .
That is vS “ 0 for all v P V .
ùñ S “ 0.
Let ppx q “ px ´ λ1 qpx ´ λ2 q ¨ ¨ ¨ px ´ λn q P F rx s be a polynomial of
degree n.
Then ppT q “ S “ 0.
Thus T satisfies a polynomial of degree n over F .
      RB (BMSCW)                   M304T                     October 2020   34 / 35
Remark
Not every linear transformation on V over F has all its characteristic
roots in F .
It depends on the field F .
For example, if F “ R, then then minimal polynomial of
                               «         ff
                                   0    1
                                 ´1 0
over F is x 2 ` 1 has no roots in F .
So in general we can not assume all characteristic roots of T P ApV q
lie in F .
       RB (BMSCW)                  M304T                 October 2020   35 / 35