Page 337 -
P. 337

320    7. Discretization of Parabolic Problems


        for i =1,... ,M. Then the following statements are equivalent:
          (1) The one-step method is nonexpansive for the model problem (7.80)
             and all eigenvalues λ i of A h .
          (2) The one-step method is nonexpansive for the problem (7.41),with
             respect to the norm  ·   h induced by  ·, ·  h .
          (3)  E h,τ   h ≤ 1 in the matrix norm  ·   h induced by the vector norm
              ·   h.

        Proof: We prove (1) ⇒ (3) ⇒ (2) ⇒ (1):
          (1) ⇒ (3): According to (7.83) (1) is characterized by

                                  |R(−λ i τ)|≤ 1 ,                  (7.92)
        for the eigenvalues λ i .
          For the eigenvector, w i with eigenvalue λ i we have (7.91), and thus, for
                          M
                             c
        an arbitrary u 0 =  i=1 i w i ,
                             M
                     2                  2
             E h,τ u 0    =     c i E h,τ w i
                     h                  h
                            i=1
                             M                  M
                                            2      2         2    2
                        =      c i R(−λ i τ)w i   =  c |R(−λ i τ)|  w i   ,
                                            h      i              h
                            i=1                 i=1
        because of the orthogonality of the w i , and analogously,
                                       M

                                   2
                                           2
                                                2
                                u 0   =   c  w i   ,
                                   h
                                                h
                                           i
                                       i=1
        and finally, because of (7.92),
                                     M

                                 2       2    2       2
                          E h,τ u 0   ≤  c  w i   =  u 0   ,
                                 h       i    h       h
                                     i=1
        which is assertion (3).
          (3) ⇒ (2): is obvious.
          (2) ⇒ (3):
               |R(−λ i τ)| w i   h =  R(−λ i τ)w i   h =  E h,τ w i   h ≤ w i   h .

          Thus, nonexpansiveness is often identical to what is (vaguely) called
        stability:
        Definition 7.25 A one-step method with a solution representation E h,τ
        for q h = 0 is called stable with respect to the vector norm  ·   h if
                                    E h,τ   h ≤ 1
        in the induced matrix norm  ·   h .
   332   333   334   335   336   337   338   339   340   341   342