180 likes | 308 Views
A COMPARISON BETWEEN SEQUENTIAL ITERATIVE AND GLOBAL METHODS FOR A REACTIVE TRANSPORT NUMERICAL MODEL. J. Erhel INRIA - RENNES - FRANCE SIAM conference on Geosciences AVIGNON - June 2005. Geochemistry and transport models Numerical methods standard iterative method
E N D
A COMPARISON BETWEEN SEQUENTIAL ITERATIVE AND GLOBAL METHODS FOR A REACTIVE TRANSPORT NUMERICAL MODEL J. Erhel INRIA - RENNES - FRANCE SIAM conference on Geosciences AVIGNON - June 2005
Geochemistry and transport models Numerical methods standard iterative method global method Results and comparison Outline
Secondary species and chemical components Aqueous and fixed species No precipitation in a first step Action laws involving stoechiometric matrices Mass conservation laws Geochemistry modelequilibrium chemistry model
Convection dispersion Transport model Convection Dispersion Transport model for each component Initial condition Boundary conditions
Coupled reactive transport model System of Differential Algebraic Equations
Numerical discretization Cell-centered finite volume scheme for mass local conservation Implicit time scheme because of stiffness Set of nonlinear equations at each time step
Sequential iterative method Time step n NONINEAR TRANSPORT F T NONLINEAR CHEMISTRY Time step n + 1 Euler implicit time scheme Nonlinear Gauss-Seidel iterative method Global Newton method for chemistry
Global method Time step n LINEARIZED TRANSPORT T,c,s or C,F,c,s LINEARIZED CHEMISTRY Time step n + 1
User-defined function and Jacobian Variable time step and order Newton nonlinear method Sparse direct linear solver In Matlab, function ode15s modified to use UMFPACK Global method with DAE solver
Test cases Ion exchange Pyrite 4 components, 39 aqueous species and 13 fixed species. Somehow artificial with no precipitation.
Chemistry solver Ion exchange Global Newton method
Ion exchange Sequential iterative method 1 Newton iteration 30 Newton iterations
Pyrite – comparison between both methods Global method Sequential iterative method
In our tests, the global method is faster and more robust than the sequential iterative method. However, the direct linear solver requires to couple transport and chemistry model and is memory consuming. We plan to do more experiments and to design an iterative linear solver which can decouple both models. Conclusion