next up [*] [*]
Next: XSPEC v11.1 issues fixed Up: Revision History for Version Previous: Version 10 Changes

XSPEC v11.2 issues fixed in 11.3

There are two bugs in fakeit. If there is no background file in use then the BACKFILE command in the faked file is set incorrectly. The value of the fit statistic written to the screen by the fakeit command is wrong.

The normalization (Mdot) of the cooling flow models does not depend on the cosmology set by the cosmo command.

The model is not recalculated when the abundance ratios are changed by the abund command.

The error command can go into an infinite loop in (hopefully) unusual circumstances.

The help on apec and vapec erroneously refers to the XSPEC_APEC environment variable instead of the new xset APECROOT method for changing the files used.

The error option on flux and lumin does not work correctly under Solaris.

The xspec library (libxspec.a) gives a link error when used with a program other than XSPEC.

There is an error in the (v)npshock and (v)sedov models for neispec versions 1.0 and v1.1.

The BACKFILE keyword is written incorrectly by fakeit when a background file is in use.

The cooling flow models produce no flux. The reason is that the cosmological parameters are not initialized to non-zero values.

The ignore command is really slow with grating data.

The identify command (and seplot id option) miss off the final line in the list. Also, identify using the apec table does not change the file in use if xset apecroot is used.

The (v)pshock models produce garbage if only the abundance is changed between successive model evaluations.

The ? command fails to give a list of the available commands if local models are in use.

Integral IBIS files do not work in XSPEC due to a certain laxity in applying the standard. This fix allows the nth spectrum in the file to be accessed by filename{n} as though the file was in the standard type II format. A similar fix for the ARF is included.

XSPEC will not execute a script in the parent directory to the current one.

Within a script, a blank line following the newpar command will cause the command to be ignored. Additionally, any newpar command in a script that attempts to modify multiple parameters will not change the first parameter requested.

Under Linux or MacOS X garbage is written to the screen if the user gives the command cpd / or if the extend command is given when no datasets have been read.

The newpar command sets the first parameter entered twice: this problem was introduced by patch 11.2.0q which is therefore superseded by this fix.

The reflect model produces incorrect results if the response matrix does not have constant width energy bins.

There are three minor problems with saved command files : if the model string is too long it will be split into separate lines which cannot then be reread by xspec; any string parameters specified by xset are not written out; cornorm commands are written out even if no correction file is in use.

Under very unusual (and obscure) circumstances fakeit can overwrite the existing file instead of creating a new one with different name.

Giving show an incorrect argument can crash xspec under Linux and MacOS X.

The mdef command does not work if any bug patch has been applied.

The plot sum command doesn't work with gcc/g77 compilers.

The model is not completely reset if an error is made when using the model command. This can lead to problems if a fit command is given before redefining the model.

The help files for xspec and plt cannot both be accessed in the same xspec session. If the help command is invoked from the xspec prompt then subsequent invocations from the plt prompt (after iplot) will give help for xspec and not for plt. If the plt help is accessed first then that is what appears when the xspec help command is invoked.

Removing a dataset (through data none) can cause the final bin in a remaining dataset to have incorrect energy bounds. This can lead to all the channels being removed when ignoring on energy.

Local models cannot include C routines.

The name for a save file cannot contain upper case characters.

The tclout flux command actually writes out the luminosity (in 1044 erg/s) if a lumin command has been given more recently than a flux command.

The NEI models can sporadically die with seg fault, memory error, or floating exception (depending on the system).

plot icounts doesn't work if setplot add has been set. Also, 2-panel plots with the top being icounts don't work.

In an attempt to reduce compute time XSPEC doesn't recalculate models for any datasets with identical energy ranges to an earlier dataset. This is fine except for models that actually use the IFL argument and treat each dataset individually. There are no standard models which do this but users may run into this issue when writing local models.

The net errors can be (obviously) incorrect if changing to a non-standard weighting (eg gehrels) for a dataset without a background file.

In the data command attempting to skip over a non-existent dataset (eg data ,,file2 when only one file has been read in) causes a crash.

The lumin command writes q0 and lambda0 with only one decimal place of accuracy leading to confusion if lambda0 is set to eg 0.65.

The tclout plot contour command doesn't write out all the values in the steppar run.

The xion model can give zero flux after a parameter is changed (ie when the function is called for a second or subsequent time) for systems compiling using g77.

The tclout flux and eqwidth commands take the dataset number as an argument not the datagroup number. There was also a bug that the valid argument range was based on datagroups not datasets.

Mega-patch of bugs and some small new features that have been requested. Includes: a) fixed setplot dev none; b) fixed projct model so it works correctly for multiple observations ie more than one dataset per datagroup; c) added fstat and query options to tclout to return information on results of last ftest command and current setting of the query; d) if steppar has been performed on only one parameter then plot contour produces a 1-D plot; e) the model description written out can now be cut and paste to use in the model or editmod commands.

The command fakeit filename does not work correctly. If no previous dataset has been read it creates fake source and background files with zero counts and if a previous dataset has been read then it uses the backscal and areascal from this dataset instead of from the filename specified in the fakeit command.

There is a subtle bug in the code that interpolates mtables. This can cause incorrect models where an input table energy range is split across response energy ranges.

Parameters are written with an insufficient number of significant figures for some analysis of grating data.

Long model string parameters can be truncated when using the xset command with no arguments to get a list of parameters.

The mdef command doesn't work correctly for convolution models.

The bug fix to reflct in 11.2.0t introduced a new bug that causes the calculated amount of reflection above 10 keV to be smaller than it ought to be. The effect of this can be seen by comparing reflct(cutoffpl) with pexrav.

XSPEC under Linux can die due to undefined symbols when running local models which are included at run-time. This has been extensively reported for the acisabs and compps models.

There are several points to watch out for when using the error option on the flux or lumin commands. The model cannot be changed (using model, editmod, addc, or delc) between doing a fit and flux/lumin with errors. If a parameter is changed, eg the absorption column set to zero, the realizations used to estimate the error will vary the parameter around its new value but with a variance from the last fit.

There is an error in the cooling flow models (cflow, mkcfl, vmkcf) causing incorrect weightings to the emission measure at the extremes of the temperature range. The main effect seems to be an approximately 10 normalization. The old model is available using the XSPEC command xset cflow_version 1. The new version is the default and can be set by xset cflow_version 2. The new version also allows the number of temperature steps in the integration to be varied. The default is 10 and can be changed by eg xset cflow_ntemps 20. Note that changing the number of integration steps will have little practical effect unless switch=0 in the mkcfl or vmkcf models.

The compbb model crashes under Linux if the electron temperature is a frozen parameter.

If MAXVPR in is set to more than 100 then XSPEC core dumps.

Fakeit produces an incorrect background file if the input source and background files have different values of BACKSCAL.

Gain fitting parameters that are saved to a script are not read back correctly.

There is an error in the XSPEC implementation of the APEC model which causes the continuum to be calculated a factor of (1+z)2 too low. Note that this will change fitted abundance values for high redshift clusters.

The diskbb model crashes under some (apparently rare) circumstances.

The documentation for the srcut model contains an error.

The save command (and autosave) do not save the abundance table or cross-section option.

The NEI models v1.x do not work correctly with response matrices with varying energy binning - spurious line features can arise at energies where the binning changes.

The projct model produces spurious warning messages if all the channels are ignored for one or more datasets.

Models with gain fit, linked parameters, and multiple datagroups are not saved correctly. The linked parameters end up linked to the wrong parameters.

Command files created by the save command cannot contain upper case letters.

Multiplicative models set up using mdefine are (incorrectly) multiplied by the response bin width.

The model command can seg fault if the model erroneously contains delimiters (eg (,),*,+) that are not part of a valid component. An example is ending the model string with (.

Reading the xautosav.xcm file back into XSPEC does not work correctly under some circumstances involving complicated models and many datasets.

Giving the error or uncertainty command between doing a fit and a flux err causes the latter to give an incorrect result.

The etable model does not work correctly for response energies which lie outside those tabulated in the table model file. The model is set to exp(-1.0) and not the correct value of 1.0.

The eqwidth command does not reset the internal total model array correctly. The result is that following an eqwidth command with a flux or lumin command will give the flux or luminosity of the model without the line component.

When using parameters whose maximum is not equal to their top or minimum is not equal to their bottom it is possible for apparent parameter values to swing between maximum and minimum but for the model itself not to change. One diagnostic of this can be that saving a model then later reading it back in will give very different values of the fit statistic and a clearly incorrect fit.

The cutoffpl model can generate floating exceptions or NaNs when using with responses that run up to high energies (>100 keV). It also becomes less accurate at these energies especially if the individual energy bins are wide.

Fitting with C-stat and a background file does not work correctly if the background file has lots of channels with zero counts.

There is a bug in the projct model when analyzing multiple observations. The calculated models are applied to the wrong observations.

The thleqw command doesn't work.

The component plots produced after ``setplot add'' are slightly wrong for models of the form A1 + M1(....). The plot of the A1 component is actually of M1*A1. This does not effect the fit or plot of total model.

next up [*] [*]
Next: XSPEC v11.1 issues fixed Up: Revision History for Version Previous: Version 10 Changes
Ben Dorman