This preview shows page 1. Sign up to view the full content.
Unformatted text preview: f the Picard iteration.
The grid has 1000×1×1 internal nodes. The results show that convergence is slow until the 999th
iteration is reached; then convergence occurs abruptly at the 999th iteration, within machine
precision. This behaviour was also observed by Ababou (1996) for linear saturated flow problems.
Fig. 3(b) shows the numerical results for all interlooped iterations (Picard and DSCG). There
is an increase in error at the beginning of each new Picard iteration, due to updates of the matrix
coefficients. However, this error decreases with iterations, i.e., there is a global convergence of the
nonlinear Picard iterations. Fig. 3(a) Error norm of the DSCG matrix solver within a single Picard step; notice the abrupt
convergence occuring at the 999th iteration (left).
Fig. 3(b) Error norm of the DSCG matrix solver for all Picard iterations (right). 4 NUMERICAL RESULTS AND ANALYSES We now analyse the spatial statistics of the seawater wedge (interface elevation ZSALT(x,y))
for the 10 different levels of aquifer variability, as specified in the numerical continuation
method. This is shown in Fig. 4(a) and Fig. 4(b). Modélisation stochastique de l'intrusion saline en 2D plan 76 Fig. 4(a) Saltwater/freshwater interface Zsalt(x,y) for σlnK = 4.0 (left).
Fig. 4(b) Transverse profile of σZsalt versus distance (x) from seashore, for 10 different values of
σlnK ranging from 1.0 to 4.0 (right). Due to the randomness of the vertically integrated permeability K(x,y), the res...
View Full Document
This document was uploaded on 01/19/2014.
- Winter '14
- The Land, In French