Thursday, October 30, 2008

latest xspec bug fixes

Three patches have been put out in the last couple of weeks. Two of these deal with the case when parameters differ by very large amounts. This is not recommended - xspec will work best if all the parameters are scaled to O(1).

12.4.0ap In a Levenberg-Marquardt fit, the zero threshold for the diagonals of the second-derivatives matrix needs to be reduced. This is relevant for cases where fit parameters may differ by enormously large (~10^20) magnitudes. Our thanks to Roderick Johnstone for pointing this out. Report added on Oct 21, 2008.

12.4.0aq The tclout model command no longer appends a newline character to the model name. This allows its output value to be used more easily as an argument to another XSPEC command. Our thanks to Jeremy Sanders for pointing this out. Report added on Oct 23, 2008.

12.4.0ar This adds automatic normalization to the covariance calculation performed at the end of a Levenberg-Marquardt fit, which makes it more robust when dealing with large differences in magnitude among the fit parameters. Report added on Oct 29, 2008.

No comments: