** Next:** Function Specification
** Up:** Outline of the Available
** Previous:** The Quasi-Newton Method.

This method is
identical to the Gauss-Newton method where the Jacobian is estimated
by finite differences and the Hessian by second order differences.
It does not require the programming of the derivatives but makes a lot
of function computations. Its use has to be restricted to problems
where the derivatives are really too difficult to write.
It is slower and less precise than the two last algorithms.

*http://www.eso.org/midas/midas-support.html*

*1999-06-09*