Come analyze HEASARC, IRSA, and MAST data in the cloud! The Fornax Initiative is now welcoming all interested beta users.
v12.8.0 Dec 2012
New features
- Parallel-processing capability for specific tasks has been added by way of the new parallel command. This allows the user to set a maximum number of processes to spawn when running certain XSPEC commands. Currently the options for parallel-processing are limited to use with the fit and error and commands, though we plan to implement more in the near future. This first implementation is considered to be a BETA release, and we look forward to hearing your comments and suggestions.
- XSPEC now distinguishes between the fit statistic and the test statistic. The fit statistic is used to find the best-fit parameter values and errors while the test statistic is used to provide a goodness-of-fit. Consequently, the goodness command now uses the test statistic. Separating these two classes of statistic has allowed us to add a number of new test statistics. These include Kolmogorov-Smirnov and the related Anderson-Darling and Cramer-von Mises. Also new are the runs statistic, based on the number of runs of consecutive positive (or negative) residuals, and the Pearson Chi-Square (pchi) statistic, Pearson's original test statistic.
- The basic Levenberg-Marquardt fit algorithm has undergone a number of changes. The most visible is an additional column in the output during the fit. |beta|/N is the norm of the vector of derivatives of the statistic with respect to the parameters divided by the number of parameters. At the best fit this should be zero so provides another measure of how well the fit is converging. |beta|/N can also be used as the criterion to stop the fit instead of the statistic delta although this is still considered experimental.
- Other internal changes to the fit algorithm are to treat the first iteration as a special case where only normalizations are allowed to change and to add the option of using delayed gratification, which can speed up convergence.
- New models: compmag, comptb, rgsxsrc. The latter is reinstated from it use in XSPEC v11.
- New plotting command plot goodness, for plotting a histogram of the most recent goodness simulation.
- New tclout option <goodness sims>.
- Added the option of using the Goodman-Weare algorithm instead of Metropolis-Hastings when using the chains command to run MCMC.
- Added the Whittle statistic for fitting models to power density spectra.
- If a data file is read which has RESPFILE=”[]” then the response extensions (MATRIX, EBOUNDS) are read from the same file.
- Added support for a new type of model component: a mixing component which operates on the model pre-multiplied by the effective area. This is similar to the pile-up model type but allows for a models which require mixing between multiple spectra.
- Table model files can now have interpolated and additional parameters intermixed provided that additional parameters are indicated by METHOD=-1 in the PARAMETERS extension.
Enhancements previously released as patches to 12.7.1
- The "plot chain" has a new "thin" option for thinning the display of chain points.
- Added a "tclout ignore" option for easy retrieval of currently ignored channels. Similar capability added to PyXspec's Spectrum class.
Bug fixes
All bug fixes to v12.7.1 released as patches are included in v12.8.0. In addition the following problems have been corrected:
- The eqpair model used wrongly-sized arrays when multiple spectra were used with RMFs having varying numbers of photon energy bins. Eqpair (plus eqtherm and compth) also needed a new default value for its optical depth parameter, which fixes the problem if parameter 4 is zero.
- A fit parameter which started close to the minimum/maximum could under some circumstances be incorrectly pegged at the maximum/minimum.
- When the error command was run on a model with soft limits
(ie. soft limits for some parameters set to a narrower range than
hard limits), and the best fit value fell within the soft limit
region, the displayed differences (in parentheses) were not
consistent with the reported
limits.
- Problem occurred in apec model when zero width lines were in the final energy bin.
- Fix to a potential normalization problem occurring in lorentz model with low-energy large-width lines.
- When running initpackage and lmod from PyXspec on certain Linux builds, local models calling functions in XSPEC's xslib library producing unresolved symbol link errors.