next up previous contents index
Next: Tutorial Up: Fitting of Data Previous: The Fitting Process.

Outputs

 

To check any typing error or missing specification, first are displayed the options, the required precision, the maximum number of function evaluations and the method parameters.

The frequency of the intermediate displays are controlled by the SET/FIT PRINT=iter. It includes the display of the iteration number, the actual number of function evaluations, the sum of the squares of the residuals, the so-called reduced chi, the percentage of decrease of the reduced chi since last iteration, and, except for the NR method, the norm of the gradient and the dimension of the space spanned by the Jacobian. The reduced chi square is the

In any case, this is followed by the value of the parameters. Except for the NR method, the value of the gradient and the singular values of the Jacobian matrix are added.

At the end, a diagnostic message telling you if the convergence was reached or if any numerical failure occurred during the algorithm. The different messages are:

  --> METHOD : Convergence achieved <--
*** ERR-1-METHOD : Bad initializations ... Aborting ***
*** ERR-METHOD : Likely an error in forming the derivatives ***
*** ERR-NR  : Problems in inverting matrix  ***
*** WARN-2-METHOD : No convergence reached ***
*** WARN-3-METHOD : Final parameters not really satisfactory ***
*** ERR-4-METHOD : No convergence in singular value decomposition ***
*** WARN-5-METHOD : parameters only a good estimation ***
*** ERR-i-METHOD : Final parameters are not satisfactory ***
In the last message i varies from 6 to 8 and the greater i is, the less reliable are the final value of the parameters. For warnings and errors numbered more than 3, it is recommended to perform another FITting with different initial guesses. If warning 2 is displayed, do again the FITting starting with the last computed value of the parameters ( nfeval < 0 in the FIT/... command). If an error in the derivatives is reported, check your user functions code.

The diagnostic will be followed by the covariance matrix if you set iter to a negative value in SET/FIT PRINT=.

Finally, the found optimal value of the parameters with their estimated standard deviation are listed.


Pascal Ballester
Fri Mar 24 17:17:52 MET 1995