Page 281 - Numerical Analysis and Modelling in Geomechanics
P. 281

262 THE 2-D CHEBYSHEV SPECTRAL ELEMENT METHOD
            subdomains. Then, an approximating function is defined on each subdomain as a
            truncated  expansion  of  Chebyshev  polynomials.  For  the  simpler  1-D  case,  the
            procedure  is  described  in  Priolo  and  Seriani  (1991).  In  the  case  of  two-
            dimensional  problems,  the  original  spatial  domain  Ω  is  decomposed  into  non-
            overlapping  quadrilateral  elements  Ω ,  where  e=l,…,  n ,  and  n e  is  the  total
                                                           e
                                           e
            number of elements. As approximating functions on each element Ω , functions
                                                                   e
            belonging  to  the  space  are  chosen,  i.e.,  polynomials  of  degree≥ N 1  in  x 1  and  of
            degree≥ N  in x . Then a global approximating function is built up as a sum of the
                        2
                    2
            elemental  approximating  functions.  The  resulting  function  is  a  continuous
            piecewise polynomial defined on the decomposition   of the original domain Ω.
            In  this  case,  the  polynomial  space  is  constructed  by  using  the  Chebyshev
            orthogonal polynomials and for simplicity it is assumed that N =N =N, i.e., the
                                                               1
                                                                  2
            order of the polynomials is the same in both directions x  and x .
                                                              2
                                                         1
              It can be shown (Canuto et al., 1988) that a function f(ξ)=f(ξ , ξ ), defined on
                                                                  2
                                                               1
            the  square  interval  [−1,  1]×[−1,  1],  can  be  approximated  by  a  truncated
            expansion using a tensor product of Chebyshev polynomials as follows:
                                                                        (9.11)
            where  are  the  grid  values  of  the  function  f,  and  are  Lagrangian  interpolants
            satisfying the relation within the interval [−1, 1], and identically zero outside. Here,
            δ ik  denotes  the  Kronecker-delta  symbol,  and  ζ  stands  for  ξ 1  or  ξ 2  .  The
            Lagrangian interpolants are given by

                                                                        (9.12)



            where  T p  are  the  Chebyshev  polynomials  and  ζ i  are  the  Chebyshev  Gauss-
            Lobatto quadrature points ζi=cos (π /N) for i=0,…, N. The coordinates ξ ={ξ ,
                                                                           1i
                                         i
                                                                       ij
            ξ } of the internal nodes for the discretisation of the rectangular domain [−1 , 1]
             2j
            ×  [−1,  1]  are  obtained  as  Cartesian  products  of  the  ζ i  points.  In  order  to  apply
            these interpolants and construct the approximating function space, there needs to
                                                        2
                                 (e)
                                               (e)
            be defined the mapping Λ  (x): x≥ Ωe→ξ ≥ [−1, 1]  between the points x of
            each  element  Ω =[a ,  a e+1 ]×[b ,  b e+1 ]  of  the  decomposition  Ω  in  the  physical
                         e
                                     e
                             e
            space and the local element coordinate system {ξ , ξ } by
                                                   1
                                                      2
                                                                        (9.13)
            with and dimensions of the element Ω . Then the global approximating function
                                           e
            is formed by the sum of the elemental approximating functions (9.11) defined on
            each element.
   276   277   278   279   280   281   282   283   284   285   286