Page 65 - Basic Structured Grid Generation
P. 65

54  Basic Structured Grid Generation

                           3.4 Geodesic curves


                        Another aspect of the intrinsic geometry of surfaces arises from the problem of deter-
                        mining the curve of minimum length which joins two given points on the surface.
                                                                 3
                        For example, when the surface is a plane in E , the shortest distance between two
                        points will be a straight line. If the general problem is approached through the usual
                        calculus of variations, differential equations are obtained which any solution must sat-
                        isfy. Curves which satisfy these equations are called geodesics, although in fact not all
                        solutions are necessarily curves of minimum length.
                                                                                            1
                          According to eqn (3.22) the length of a curve C on a surface with co-ordinates u ,u 2
                        and covariant metric tensor a αβ is given by
                                             t 2               t 2

                                                     α β            1  2  1  2
                                       L =       a αβ ˙u ˙u dt =  f(u ,u , ˙u , ˙u ) dt,   (3.66)
                                             t 1              t 1
                                                                                         1
                                                                                               2
                        where t is used to parametrize the curve, each a αβ in general is a function of u and u ,
                        and we assume that t takes the fixed values t 1 ,t 2 at the given end-points. Moreover, f
                        is formally regarded as a function of four independent variables. Neighbouring curves,
                                                  α
                                                                          α
                                                          α
                                                                                 α
                        having first-order variations δu and δ ˙u in the values of u and ˙u at corresponding
                        values of t, but still having the same end-points, have a first-order variation in length

                                               t 2        t 2  ∂f  α   ∂f   α

                                        δL =     δf dt =        δu +      δ ˙u  dt
                                                             ∂u α     ∂ ˙u α
                                              t 1        t 1
                        with summation over α. Integration by parts on the last term gives
                                            ∂f    α      t 2  ∂f   d   ∂f      α
                                                     t 2
                                     δL =      δu    +           −           δu dt,
                                            ∂ ˙u α           ∂u α  dt  ∂ ˙u α
                                                    t 1  t 1
                        and the integrated part vanishes because of the fixed end-point requirement.
                          If L is to be a minimum for C, the first-order variation δL must be zero for arbitrary
                                   α
                        variations δu . Thus
                                               t 2
                                                   ∂f    d   ∂f      α

                                                      −            δu dt = 0
                                                  ∂u α  dt  ∂ ˙u α
                                              t 1
                                                             2
                                                         1
                        for arbitrary (first-order) variations δu ,δu . A standard argument of the calculus of
                        variations then leads to the conclusion that the two differential equations
                                              ∂f    d     ∂f
                                                  −          = 0,   α = 1, 2,              (3.67)
                                              ∂u α  dt  ∂ ˙u α
                        must hold everywhere on C. Curves for which these equations hold are geodesics.
                        Equations (3.67) are the Euler-Lagrange equations (or just the Euler equations)for
                        the variational problem δL = 0 with L given by (3.66).

                                                  2
                                               1
                                                    α β
                          Now since f =   a αβ (u ,u )˙u ˙u ,wehave
                                          ∂f     1 ∂a αβ  α β     ∂f    1     β
                                              =        ˙ u ˙u  and    =   a γβ ˙u .
                                         ∂u γ   2f ∂u γ          ∂ ˙u γ  f
   60   61   62   63   64   65   66   67   68   69   70