Page 69 -
P. 69

52    2. Finite Element Method for Poisson Equation

                     $                     % 1/2
                          1          1
                         n  2           2         √
                  =        n dx +      n dx     =   2n →∞ for n →∞ .
            v n   a
                        0          1−  1
                                     n
        Therefore, there exists no constant C> 0 such that  v  a ≤ C v  0 for all
        v ∈ V .
          However, as we will show in Theorem 2.18, there exists a constant C> 0
        such that the estimate
                                          for all v ∈ V
                             v  0 ≤ C v  a
        holds; i.e.,  ·   a is the stronger norm.
          It is possible to enlarge the basic space V without violating the previous
        statements. The enlargement is also necessary because, for instance, the
        proof of the existence of a solution of the variational equation (2.13) or
        the minimization problem (2.14) requires in general the completeness of V.
        However, the actual definition of V does not imply the completeness, as
        the following example shows:
        Example 2.8 Let Ω = (0, 1) again and therefore

                                           1
                               a(u, v):=   u v dx .

                                         0

                    α     α          1
        For u(x):= x (1−x) with α ∈   , 1 we consider the sequence of functions
                                     2
                                                    & 1    1  '
                           u(x)              for  x ∈  , 1 −  ,
                                                     n     n
                                                    &   '
                               1
                 u n(x):=  nu( ) x           for  x ∈ 0,  1  ,
                               n                        n
                                                    &       '
                                  1                      1
                           nu(1 − )(1 − x)   for  x ∈ 1 − , 1 .
                                   n                     n
        Then
                          u n − u m   a → 0  for n, m →∞ ,
                           u n − u  a → 0  for n →∞ ,
        but u/∈ V ,where V is defined analogously to (2.7) with d =1.
                                                  ˜
          In Section 3.1 we will see that a vector space V normed with  ·   a exists
                                 ˜
                      ˜
        such that u ∈ V and V ⊂ V . Therefore, V is not complete with respect
        to  ·   a ;otherwise, u ∈ V must be valid. In fact, there exists a (unique)
        completion of V with respect to  ·  a (see Appendix A.4, especially (A4.26)),
        but we have to describe the new “functions” added by this process. Besides,
        integration by parts must be valid such that a classical solution continues to
        be also a weak solution (compare with Lemma 2.1). Therefore, the following
        idea is unsuitable.
        Attempt of a correct definition of V :
        Let V be the set of all u with the property that ∂ i u exists for all x ∈ Ω
        without any requirements on ∂ i u in the sense of a function.
   64   65   66   67   68   69   70   71   72   73   74