Page 51 - 6099
P. 51

you would not expect well-posedness of the backward-in-time problem for the diffusion
                   equation.
                       As a third example, consider solving a matrix equation instead of a PDE: namely,
                   Au = b, where A is an m × n matrix and b is a given m-vector. The ”data” of this
                   problem comprise the vector b. If m > n, there are more rows than columns and the
                   system is overdetermined. This means that no solution can exist for certain vectors b;
                   that is, you don’t necessarily have existence. If, on the other hand, n > m, there are
                   more columns than rows and the system is underdetermined. This means that there are
                   lots of solutions for certain vectors b; that is, you can’t have uniqueness. Now suppose
                   that m = n but A is a singular matrix; that is, det A = 0 or A has no inverse. Then the
                   problem is still ill-posed (neither existence nor uniqueness). It is also unstable. To illustrate
                   the instability further, consider a nonsingular matrix A with one very small eigenvalue.
                   The solution is unique but if b is slightly perturbed, then the error will be greatly magnified
                   in the solution u. Such a matrix, in the context of scientific computation, is called ill-
                   conditioned. The ill-conditioning comes from the instability of the matrix equation with
                   a singular matrix.
                       As a fourth example, consider Laplace’s equation u xx + u yy = 0 in the region D =
                   (−∞ < x < ∞, 0 < y < ∞). It is not a well-posed problem to specify both u and u y on
                   the boundary of D, for the following reason. It has the solutions

                                                          1   √
                                               u n (x, y) =  e − n  sin nx sin ny.                 (5.10)
                                                          n
                                                                                                √
                   Notice that they have boundary data u n (x, 0) = 0 and ∂u n /∂y(x, 0) = e  − n  sin nx,
                   which tends to zero as n → ∞. But for y 6= 0 the solutions u n (x, y) do not tend to zero
                   as n → ∞. Thus the stability condition (iii) is violated.
                       We present the slightly more realistic examples.

                   Example 5.1 A string with fixed endpoints. Consider a string fixed at x = 0 and
                   x = a, as in Fig. 5.4








                                            x = 0                          x = a


                                         Figure 5.4: A string with fixed endpoints


                       It satisfies the wave equation
                                                     2
                                                            2
                                                  1 ∂ u    ∂ u
                                                        =      , 0 > x > w
                                                  2
                                                 c ∂t 2    ∂x 2
                   with boundary conditions

                                                u(0, t) = u(a, t) = 0, t > 0,



                                                             44
   46   47   48   49   50   51   52   53   54   55   56