Page 38 - Matrices theory and applications
P. 38

2.3. Alternate Matrices and the Pfaffian

                                                                                             =
                                                          ∈ M n (A) are similar (that is, M
                              Proposition 2.2.2 If M, M
                               −1
                                 MP with P ∈ GL n (A)), then
                              P

                                                       det M =det M.
                              2.3 Alternate Matrices and the Pfaffian                           21
                              The very simple structure of alternate forms is described in the following
                              statement.
                              Proposition 2.3.1 Let B be an alternate bilinear form on a vector space
                              E,of dimension n. Then there exists a basis
                                                {x 1 ,y 1 ,... ,x k ,y k ,z 1,... ,z n−2k }
                              such that the matrix of B in this basis is block-diagonal, equal to
                              diag(J,... ,J, 0,... , 0),with k blocks J defined by

                                                              0  1
                                                      J =            .
                                                             −10
                                Proof
                                We proceed by induction on the dimension n.If B = 0, there is nothing to
                              prove. If B is nonzero, there exist two vectors x 1 ,y 1 such that B(x 1 ,y 1 )  =0.
                              Multiplying one of them by B(x 1 ,y 1) −1 , one may assume that B(x 1 ,y 1 )=
                              1. Since B is alternate, {x 1 ,y 1 } is free. Let N be the plane spanned by x 1 ,y 1 .
                              The set of vectors x satisfying B(x, v) = 0 (or equivalently B(v, x)= 0,
                              since B must be skew-symmetric) for every v in N is denoted by N .The
                                                                                        ⊥
                              formulas
                                          B(ax 1 + by 1 ,x 1 )= −b,  B(ax 1 + by 1,y 1 )= a

                              show that N ∩ N  ⊥  = {0}. Additionally, every vector x ∈ E can be written
                              as x = y + n,where n ∈ N and y ∈ N  ⊥  are given by
                                            n = B(x, y 1 )x 1 − B(x, x 1 )y 1 ,  y := x − n.

                              Therefore, E = N ⊕ N . We now consider the restriction of B to the
                                                   ⊥
                              subspace N  ⊥  and apply the induction hypothesis. There exists a basis
                              {x 2 ,y 2 ,... ,x k ,y k ,z 1 ,... ,z n−2k } such that the matrix of the restriction of
                              B in this basis is block-diagonal, equal to diag(J,... ,J, 0,... , 0), with k−1
                              blocks J, which means that B(x j ,y j )= 1= −B(y j ,x j )and B(u, v)= 0
                              forevery otherchoiceof u, v in the basis. Obviously, this property extends
                              to the form B itself and the basis {x 1 ,y 1 ,... ,x k ,y k ,z 1 ,... ,z n−2k }.

                                We now choose an alternate matrix M ∈ M n (K) and apply Proposition
                              2.3.1 to the form defined by M. In view of Section 1.2.3, we have the
                              following.
   33   34   35   36   37   38   39   40   41   42   43