Page 126 - Matrices theory and applications
P. 126

109
                                                 6.3. Similarity Invariants and Jordan Reduction
                              Definition 6.3.2 The elementary divisors of the matrix M ∈ M n (k) are
                                             α(j,k)
                                                  for which the exponent α(j, k) is nonzero. The mul-
                              the polynomials p
                                             k
                                                             m
                                                               is the number of solutions j of the
                              tiplicity of an elementary divisor p
                                                             k
                              equation α(j, k)= m.The list of elementary divisors is the sequence of
                              these polynomials, repeated with their multiplicities.
                                Let us begin with the case of the companion matrix N of some polynomial
                              P. Its similarity invariants are (1,... , 1,P)(see above). Let Q 1 ,... ,Q t be
                              its elementary divisors (we observe that each has multiplicity one). We then
                              have P = Q 1 ··· Q t, while the Q l ’s are pairwise coprime. To each Q l we
                              associate its companion matrix N l , and we form a block-diagonal matrix
                              N := diag(N 1 ,... ,N t ). Since each N l − XI l is equivalent to a diagonal

                              matrix
                                                                    
                                                        1
                                                           .
                                                           .        
                                                           .        
                                                                    
                                                              1     
                                                                  Q l
                              in M n(l) (k[X]), the whole matrix N − XI n is equivalent to

                                                                           
                                                     1
                                                        .
                                                        .                  
                                                         .         O       
                                                                           
                                                             1
                                                                           
                                              Q :=                           .
                                                                Q 1
                                                                           
                                                                           
                                                                    .
                                                                    .      
                                                        O            .     
                                                                        Q t

                              Let us now compute the similarity invariants of N , that is, the invariant
                              factors of Q. It will be enough to compute the greatest common divisor
                              D n−1 of the minors of size n − 1. Taking into account the principal minors
                              of Q,we see that D n−1 must divide every product of the form

                                                         Q l ,  1 ≤ k ≤ t.
                                                      l=k

                              Since the Q l ’s are pairwise coprime, this implies that D n−1 =1. This

                              means that the list of similarity invariants of N has the form (1,... , 1, ·),
                              where the last polynomial must be the characteristic polynomial of N .

                              This polynomial is the product of the characteristic polynomials of the
                              N l ’s. These being equal to the Q l ’s, the characteristic polynomial of N is

                              P. Finally, N and N have the same similarity invariants and are therefore

                              similar.
                                Now let M be a general matrix in M n (k). We apply the former reduction
                              to every diagonal block M j of its Frobenius canonical form. Each M j is
                              similar to a block-diagonal matrix whose diagonal blocks are companion
                              matrices corresponding to the elementary divisors of M entering into the
   121   122   123   124   125   126   127   128   129   130   131