Page 287 - Matrix Analysis & Applied Linear Algebra
P. 287

5.2 Matrix Norms                                                                   283





                                                       Properties of the 2-Norm
                                       In addition to the properties shared by all induced norms, the 2-norm
                                       enjoys the following special properties.
                                                               ∗
                                       •    A  = max     max |y Ax|.                            (5.2.9)
                                               2
                                                   x  =1  y  =1
                                                     2     2
                                                     ∗
                                       •    A  =  A   2 .                                      (5.2.10)
                                               2
                                                         2
                                              ∗
                                       •    A A  2 =  A  .                                     (5.2.11)
                                                         2

                                           
 A
                                       •         0 
                      .                    (5.2.12)
                                                    
 = max  A  ,  B

                                              0  B                2     2
                                                     2
                                                                     ∗
                                              ∗
                                                                                 ∗
                                       •    U AV  2 =  A  2 when UU = I and V V = I.           (5.2.13)
                                        You are asked to verify the validity of these properties in Exercise 5.2.6
                                    on p. 285. Furthermore, some additional properties of the matrix 2-norm are
                                    developed in Exercise 5.6.9 and on pp. 414 and 417.
                                        Now that we understand how the euclidean vector norm induces the matrix
                                    2-norm, let’s investigate the nature of the matrix norms that are induced by the
                                    vector 1-norm and the vector ∞-norm.
                                                Matrix 1-Norm and Matrix ∞-Norm
                                       The matrix norms induced by the vector 1-norm and ∞-norm are as
                                       follows.

                                       •    A  = max  Ax  = max         |a ij |
                                               1             1
                                                   x  =1          j                            (5.2.14)
                                                     1
                                                                      i
                                                = the largest absolute column sum.

                                       •    A    = max  Ax       = max    |a ij |
                                               ∞       =1     ∞     i
                                                    x                                          (5.2.15)
                                                      ∞
                                                                        j
                                                 = the largest absolute row sum.
                                    Proof of (5.2.14).  For all x with  x  =1, the scalar triangle inequality yields
                                                                     1


                                      Ax  =       A i∗ x =      a ij x j  ≤   |a ij ||x j | =  |x j |  |a ij |


                                         1
                                              i           i   j          i  j            j       i



                                           ≤      |x j |  max  |a ij |  = max  |a ij | .
                                                         j              j
                                                j            i              i
   282   283   284   285   286   287   288   289   290   291   292