3.5.2 Fréchet Derivative (Jacobian Matrix)



next up previous contents
Next: 3.5.3 Automatic Local Scaling Up: 3.5 Solving the Nonlinear Previous: 3.5.1 Modified Newton Method

3.5.2 Fréchet Derivative (Jacobian Matrix)

 

in the iteration (3.5-2) needs not to be the exact Jacobian in order to get a convergent scheme, nevertheless an exact Jacobian is desirable since it yields quadratic convergence.

As far as possible we use exact values for the Jacobian. The Jacobian matrix is assembled in exactly the same order as the right hand side . Each entry in consists of a sum of contributions from the nine points of the discretization stencil. The subscript indicates the quantity, the subscripts identify the grid point. The coefficient consists of the physical parameters which may depend on the concentrations , and is a concentration independent function of grid spacing or time step size.

 

 

Corresponding to the nine points of the discretization stencil the Jacobian has nine block-diagonals (Figure 3.5-1), each block ( in Figure 3.5-1) consisting of a possibly full matrix (3.5-8). Following the chain rule we have . For the fluxes the coefficient may depend on concentrations at the grid points and on the concentration at a mid interval point (3.4-27), (3.4-31). The derivative is divided up among the grid points consistently to the interpolation used for .

 



Martin Stiftinger
Wed Oct 19 13:03:34 MET 1994