The Function Information File

While the actual photon model functions are part of the FORTRAN code, many other properties of the photon functions are read from the file pointed to by the logical MFIT_INFO_FILE. This allows the user to change these function properties by copying the standard file /wingspan/mfit_func.info, modifying it, and modifying the logical MFIT_INFO_FILE to point to new file. Since this file is read using fixed formats, its layout should not be modified. The format of the file is described in the comments at the top of the standard file.

The first line describing each term specifies the name of the term. The next line lists the number of parameters ( ). This value should not be altered because it must match the code. The next lines describe each parameter in 10 columns of specifications. The first and second columns specify the parameter names and units. The third column contains an `F' or `V' to specify the default parameter variation status. The fourth column specifies the default parameter value.

The Levenberg-Marquardt nonlinear fitting algorithm will converge only if the parameter values are close enough to the best-fit values. A common failure mode of the algorithm is to change the parameter values excessively so that the new values are too far from the best values. The algorithm is then unable to converge. To avoid this problem, MFIT limits the change of the parameter values per iteration-this makes the program more robust with only a small increase in the run time. The allowed limits of the parameter value changes per iteration are as specified in the fifth and sixth columns of the function information file. The fifth column specifies the relative parameter change limit-the parameter can change by no more than this fraction of its value per iteration. The sixth column specifies the absolute parameter change limit. The parameter change per iteration is limited to the larger of the two limits. Either or both of the limits can be specified as zero, in which case the limit is not applied. Note that if only a relative limit is specified, the parameter can never change sign.

A `P' in the the seventh column indicates that the parameter must be positive. This option only checks the user's input, it does not require the fitted value to be positive. Requiring the fitted value to be positive can be done by specifying a non-zero relative change limit and a zero absolute change limit in addition to the `P' specification.

The remaining specifications (eighth through tenth columns) in the function information file allow the user to control the fixing of ill-determined parameters in the batch fit mode. The eighth column specifies the priorities for fixing ill-determined parameters. A value of zero indicates that a parameter is never to be fixed. After a batch fit, MFIT loops through the varying parameters in increasing order of the priority value specified in the eighth column, starting with priority 1. The first parameter found to be ill-determined is fixed, the fit is repeated, and the results are again examined for ill-determined parameters. The process stops when there are no varying, ill-determined parameters. The ninth and tenth columns specify the meaning of ``ill-determined'': the ninth column specifies the relative requirement, the tenth the absolute requirement. If the estimated parameter uncertainty exceeds either the absolute specification OR the product of the relative specification and the parameter value, that parameter is considered ill-determined. Either requirement can be omitted by specifying zero.

BATSE Home Page Responsible Manager: Jerry Fishman
Web Curator: Valerie Connaughton
Last Updated: November 8, 1996.