Page 177 - Basic Structured Grid Generation
P. 177

166  Basic Structured Grid Generation

                        and this is the Winslow model, eqns (5.1). Thus transformations satisfying the Winslow
                        model are extremals of the functional (6.41), which may be written, using (1.158),
                        (1.162), and (1.163), as
                                                  1      11    22
                                                       (g  + g ) dx dy.                    (6.43)
                                                  2   R
                        Transforming the double integral to the computational domain gives
                                1         g 22 + g 11     √  1         g 11 + g 22
                                                 ( g dξ dη) =          √       dξ dη.      (6.44)
                                2   R      g                 2   R       g
                          The Winslow transformation, mapping a square region to R, always gives an unfolded
                        grid, except for problems which can arise due to the nature of the discretization process
                        (i.e. truncation errors). In particular, this can happen when R is not convex.

                        6.4.2 The weighted L-functional


                        The concept of a weight function in one dimension can be generalized to higher
                        dimensions. Suppose that we wish the lengths of grid cells in the ξ and η directions
                        at a point corresponding to (ξ, η) in the computational plane to be proportional to the
                        positive weight functions ϕ(ξ, η), ψ(ξ, η), respectively. Then a functional analogous
                        to (6.13) is given by


                                                      2
                                                                    2
                                           1      (x ξ ) + (y ξ ) 2  (x η ) + (y η ) 2
                                      I L =                   +               dξ dη
                                           2   R     ϕ(ξ, η)       ψ(ξ, η)
                                           1         g 11    g 22
                                        =                +          dξ dη.                 (6.45)
                                           2   R     ϕ(ξ, η)  ψ(ξ, η)
                          The two Euler-Lagrange equations for the variational problem δI = 0are
                                               ∂     1 ∂x     ∂     1 ∂x
                                                         +            = 0
                                               ∂ξ  ϕ ∂ξ    ∂η   ψ ∂η
                                               ∂     1 ∂y     ∂     1 ∂y
                                                         +            = 0,                 (6.46)
                                               ∂ξ  ϕ ∂ξ    ∂η   ψ ∂η
                        which we can write as the single vector equation

                                               ∂    1 ∂r     ∂     1 ∂r
                                                         +            = 0,                 (6.47)
                                               ∂ξ  ϕ ∂ξ    ∂η   ψ ∂η
                                   x

                        where r =   .
                                   y
                          Carrying out the partial differentiations gives the set of equations
                                             A 11 r ξξ + A 12 r ξη + A 22 r ηη + S = 0,    (6.48)
                        where A 11 , A 12 , A 22 are the symmetric 2 × 2 matrices
                                    1     10               00             1     10
                              A 11 =           ,  A 12 =         ,  A 22 =           ,     (6.49)
                                    ϕ   01                00              ψ    01
   172   173   174   175   176   177   178   179   180   181   182