Page 122 - Matrices theory and applications
P. 122

6.3. Similarity Invariants and Jordan Reduction
                                                                                            105
                                When A 0 = A 1 = I n , Theorem 6.3.2 tells that XI n − B 0 and XI n − B 1
                              are equivalent, namely that they have the same invariant polynomials, if
                              there exists P ∈ GL n (k) such that PB 0 = B 1 P, which is the criterion
                              given by Theorem 6.3.1.
                                Proof
                                We prove Theorem 6.3.2. The condition is clearly sufficient.
                                Conversely, if XA 0 +B 0 and XA 1 +B 1 are equivalent, there exist matrices
                              P, Q ∈ GL n (A), such that P(XA 0 + B 0 )= (XA 1 + B 1 )Q.Since A 1 is
                                                                      2
                              invertible, one may perform Euclidean division of P by XA 1 + B 1 on the
                              right:
                                                   P =(XA 1 + B 1 )P 1 + G,
                              where G is a matrix whose entries are constant polynomials. We warn the
                              reader that since M n (k) is not commutative, Euclidean division may be
                              done either on the right or on the left, with distinct quotients and distinct
                              remainders. Likewise, we have Q = Q 1 (XA 0 + B 0 )+ H with H ∈ M n (k).
                              Let us write, then,
                                (XA 1 + B 1 )(P 1 − Q 1 )(XA 0 + B 0 )= (XA 1 + B 1 )H − G(XA 0 + B 0 ).
                              The left-hand side of this equality has degree (the degree is defined as the
                              supremum of the degrees of the entries of the matrix) 2 + deg(P 1 − Q 1 ),
                              while the right-hand side has degree less than or equal to one. The two
                              sides, being equal, must vanish, and we conclude that
                                                  GA 0 = A 1 H,  GB 0 = B 1 H.
                              There remains to show that G and H are invertible. To do so, let us define
                              R ∈ M n(A) as the inverse matrix of P (which exists by assumption). We
                              still have
                                             R =(XA 0 + B 0 )R 1 + K,  K ∈ M n(k).

                              Combining the equalities stated above, we obtain
                                             I n − GK =(XA 1 + B 1 )(QR 1 + P 1 K).
                              Since the left-hand side is constant and the right-hand side has degree
                              1+ deg(QR 1 + P 1 K), we must have I n = GK,so that G is invertible.
                              Likewise, H is invertible.

                                We conclude this paragraph with a remarkable statement:
                              Theorem 6.3.3 If B ∈ M n (k),then B and B T  are similar.
                                                          T
                                Indeed, XI n − B and XI n − B are transposes of each other, and hence
                              have the same list of minors, hence the same invariant factors.

                                2
                                 The fact that A 1 is invertible is essential, since the ring M n(A) is not an integral
                              domain.
   117   118   119   120   121   122   123   124   125   126   127