next up previous
Next: 8. Grid & Element Up: AMIGOS: Analytical Model Interface Previous: 6. ESA - Equation

7. Solving Methods

Solver

The analysis by the equation system analyzer leads to a set of linear equation systems that are usually filled sparsely. To solve these systems AMIGOS is equipped with a direct and iterative solver taken from the development libraries [5].

The direct solver uses the Gaussian elimination method [6] and should be used for badly conditioned systems. The disadvantage of this method is the immense amount of memory ($ O(N^2)$) and operation count for inversion ($ O(N^3)$) that is unacceptable for simulations in three dimensions. Therefore it is advisable to use the implemented iterative solver which is based on the stabilized biconjugate gradient method (BiCGStab [7]) using an incomplete LU factorization [8] and an auto scaling algorithm[9].

Time Discretization

To solve time dependent problems AMIGOS is equipped with the Backward Euler and the Cranck-Nicholson iteration scheme combined with a temporal error estimation and timestep control.

If AMIGOS detects a time dependence of a quantity it checks weather the last calculated time step was accurate enough or if the last calculation has to be rejected restarting the Newton iteration but now with a user defined `rejection-ratio' times smaller time step. The time-estimation algorithm is used to adapt the time-step in dependence of the calculated deviation of the chosen time interpolation (linear or logarithmic).


next up previous
Next: 8. Grid & Element Up: AMIGOS: Analytical Model Interface Previous: 6. ESA - Equation
M. Radi, E.Leitner, and S. Selberherr