Manual Page for Command >>> FORWARD
>>> FORWARD (: iswitch)
With this command the elements of Jacobian matrix are calculated by means of
a forward finite difference quotient:
The evaluation of the Jacobian thus requires n+1 TOUGH2 simulations, where n
is the number of parameters. The size of the perturbation dp can be controlled
using command >>> PERTURB.
The Jacobian is used in both the minimization algorithm and the a posteriori
error analysis. For the Levenberg-Marquardt minimization algorithm, the
accuracy obtained by using a forward (as opposed to centered) finite difference
quotient is usually sufficient, especially during the first few iterations far
away from the minimum. However, if approaching the minimum and especially for
the subsequent error analysis one might want to use a more accurate
approximation of the Jacobian. If a colon is given on the command line
followed by an integer iswitch, iTOUGH2 switches from forward to centered
finite differences after iswitch iterations.
>>> perform a total of : 6 iTOUGH2 ITERATIONS
>>> use FORWARD finite differences for : 5 iterations, i.e.
a centered finite difference quotient is used for the final
iteration and error analysis.
>>> PERTURBation factor is : 0.005 times the parameter value
>>> CENTERED |
Back to Command Index
Page updated: July 29, 1997