Page 319 - DMTH502_LINEAR_ALGEBRA
P. 319

Unit 29: Spectral Theory and Properties of Normal Operators




          Since                                                                                 Notes


                                      I =   E 1ji  E 2  2 j  E mj m
                                           1 j    2 j    jm
          each vector   in V may be written in the form

                                                 E
                                       =      E 1j 1 2 j 2  ...E mj m  .          … (3)
                                         j 1 ..., j m
          Suppose j , ..., j , are indices for which   = E E , ... E      0. Let
                  1   m                      1j1  2j2  mjm
                                       =   II E nj  .
                                      i        n
                                          n  i
          Then   = E  ; hence there is a scalar c  such that
                   ij i                 i
                                   T   = c ,  1   i   m.
                                     1   i
          For each T in , there exist unique scalars b  such that
                                             i
                                         m
                                            i i
                                     T =   b T
                                         i  1
          Thus

                                    T  =   b T
                                            i i
                                          i
                                       =    b c  .
                                             i i
                                          i
          The function T       b c                        i                 i
                            i i , is evidently one of the roots, say r  or , and   lies in V(r ). Therefore,
                          i
          each non-zero term in (3) belongs to one of the spaces  V(r ), ..., V(r ). It follows that V is the
                                                          1       k
          orthogonal direct sum of V(r ), ...,V(r ).
                                 l      k
          Corollary: Under the assumptions of the theorem, let P  be the orthogonal projection of V on
                                                        j
          V(r ) (1   j   k). Then P P  = 0 when i   j,
             j              i  j
                                      I = P  + ... + P ,
                                          1     k
          and every T in  may be written in the form
                                               P
                                             T
                                     T =   r j ( ) .                              … (4)
                                                j
                                          i
          Definition: The family of orthogonal projections {P , ..., P } is called the resolution of the identity
                                                   1   k
          determined by , and (4) is the spectral resolution of T in terms of this family.
          Although the projections P , ..., P , in the preceding corollary are canonically associated with the
                                1   k
          family , they are generally not in  nor even linear combinations of operators in ; however,
          we shall show that they may be obtained by forming certain products of polynomials in elements
          of .
          In the study of any family of linear operators on an inner product space, it is usually profitable
          to consider the self-adjoint algebra generated by the family.
          Definition:  A  self-adjoint algebra  of  operators  on  an  inner  product space  V  is  a  linear
          sub-algebra of L(V, V) which contains the adjoint of each of its members.





                                           LOVELY PROFESSIONAL UNIVERSITY                                   313
   314   315   316   317   318   319   320   321   322   323   324