xspec patches

A few new patches have been added recently. 12.4.0x fixes a number of problems that showed up when using the new g++-4.3.0 compiler and when using gcc on Solaris platforms. This patch level is released as part of HEAsoft 6.4.1. 12.4.0y improves the handling of cases when the L-M algorithm pegs a parameter due to a zero or negative pivot value in the Hessian. 12.4.0z keeps the fit valid when the fit method is changed. This allows the improve command to be run after an L-M fit. 12.4.0aa is a fix to allow filenames prompted for by the fakeit command to being with a / ie to be absolute paths. Patches available from the usual place.

xselect: Chandra response generation

Overhauled the response generation option on "save spec" for Chandra ACIS data. I added an extended? parameter to specify whether the source is a point source or not. For a point source the arf is created using mkarf, and for an extended source mkwarf is used. The rmf is generated using mkrmf or mkacisrmf depending on whether CTI correction has been performed. The script checks keywords in the spectrum to determine which is appropriate. Note that to find some of the auxiliary files required xselect should be being run in either main directory for the dataset or in a directory one level down (eg secondary).

Note that I also modified the FITS region extension written by extractor so that the EXTNAME is REGION and EXTVER is used to track multiple extensions. This is a change from an EXTNAME of REG001## and is made to conform with the CXC standard defined by McDowell & Rots. There may be an issue here is XMM-Newton SAS routines are expecting the REG001## names.

Science 2.0

SciAm has a brief survey of Web 2.0 applications in science. In particular, it argues that putting scientists working notes on-line will foster greater communication and collaboration.

reading old RMFs

Tahir points out that the BeppoSAX RMFs don't work in rbnrmf. The problem is that these files have only RMFVERSN='1992a' and do not have the HDUVERS keyword that the software now requires. The workaround is to add HDUVERS='1.0.0' to the MATRIX extension. I've updated the relevant subroutines to check for RMFVERSN if HDUVERS is not found.

setplot wave units in xspec

Changed the plotting units for the setplot wave option. Now showing f_nu and variants instead of f_lambda. The model and ufspec y-axis is photons/cm^2/s/Hz; emodel and eufspec are in Jansky (1e-23 erg/cm^2/s/Hz); eemodel and eeufspec in keV Jy. The latter is a bizarre mixed unit but it isn't clear to me what to use here. It might be an idea to have an option for the user to switch between f_lambda and f_nu.

Updated: decided that for the eemodel/eeufspec plots the units should be erg/cm^2/s corresponding to standard nu F_nu plots.

Independently, I fixed an error in setting the y-axis range for eufspec and eeufspec under setplot wave.

genrsp v2.02

Added an option for the resolution varying linearly with either the energy or the wavelength.

xspec updates

A number of updates and bug fixes, mainly minor.

12.4.0p This is the fix for suzpsf alluded to earlier.

12.4.0q CHANTYPE was not being set correctly after fakeit none leading to confusing warning messages.

12.4.0r A generalization of fakeit to work correctly with multiple models as requested by K.D. Kuntz.

12.4.0s A couple of patches for plot eff bizarreness that I ran across.

12.4.0t Fix for a seg fault in data when the user gives the filename as empty double quotes.

12.4.0u A fix for unusual compiler configurations.

12.4.0v Allow the unnamed model to be turned on and off.

12.4.0w Correction for a bug when using non-standard weighting and toggling between chi and cstat statistics.

Patches are available from the usual place.

likelihood-ratio test

Added a new Tcl script, lrt.tcl, to calibrate the likelihood ratio test. Currently only works with one data set. The user defines two models (by default model0 and model1) then the script generates a user-specified number of simulated data sets based on model0 and for each one evaluates the log(likelihood ratio) between model1 and model0. The script returns the fraction of simulations with a smaller log(LR) than the data.

I will then use lrt.tcl as the basis for a script to calculate F-test probabilities for
adding an additional model component.

fakeit with multiple sources

We now have the patch out for fakeit to simulate data from the case where there are multiple sources (with their own responses). This is xspec v12.4.0r and is available from the usual place.

suzaku xis response generation

Enabled the automated response generation in xselect for Suzaku XIS spectra. I'm still waiting for a decision from the XIS team on what default energy and channel binning to use in the response matrix. Ishisaki-san has suggested that the correct estepfile option for xissimarfgen is "medium" so the response generation uses this.

Currently using a "medium" binning scheme which produces 2048 channels. Spectra extracted from xselect are binned using this scheme whether or not the automated response generation is used.

Note that I also modified the MDB code to allow trailing wild cards eg XIS* will match to any of XIS0, XIS1, XIS2, or XIS3.

Also, updated addascaspec so it runs correctly if there is no ARF used.

suzaku xis 5x5 and 3x3 files

Checked in a change to xselect so that when there is a mix of 5x5 and 3x3 suzaku xis files read in then, when doing an extract, xis5x5to3x3 is used to create a temporary 3x3 file. This was requested by the xis team and ensures that doing an extract event creates a correct output 3x3 file. At the moment we aren't doing anything with the extra information in the 5x5 file.

local models in xspec v11 on the Mac OS X

Bryan points out that the documentation for local models in v11 doesn't include the special instructions required for the Mac OS X ie because we can't make a shared library the xspec11 executable must be rebuilt with the new local model library. I've added this to the relevant manual page although hopefully most people are using v12 now anyway.

grade strings in xselect/extractor

The specifications rules for grade strings in the xselect "filter grade" command do seem lead to a little confusion so made them a but more lenient. Now allow spaces in between individual elements. eg "0:0 2:4 6:6" is now allowed as well as "0:0,2:4,6:6". This change is in extractor v4.83.

twittering

There is now an xspector twitter. I'm not sure whether I will keep this up or where it will go - it's an experiment.

suzpsf model in xspec

As noted yesterday the current version of suzpsf assumes the WMAP is in sky coordinates, which is no longer true. I've now fixed this. There are two changes required. Firstly, extractor has been modified (v4.83) to write the REF* keywords which XMM files use to record the original sky WCS keywords. Secondly, the suzpsf model has been modified to use these keywords (as the xmmpsf model already did).

To fix spectra that were created using older versions of extractor add the following keywords to the primary header with the values of the corresponding keywords from the EVENTS extension of the input event file.
REFXCRVL <- TRCVL10 from events files
REFXCRPX <- TCRPX10 from events files
REFXCDLT <- TCDLT10 from events files
REFYCRVL <- TRCVL11 from events files
REFYCRPX <- TCRPX11 from events files
REFYCDLT <- TCDLT11 from events files

more minor xspec patches

12.4.0m The start-up statistic, method, and weight settings in the user's Xspec.init file are effectively ignored. Instead these always start with Xspec's hardcoded default settings (chisq, leven, and standard respectively). Our thanks to Maurice Leutenegger for pointing this out. Report added on Mar 05, 2008.

12.4.0n If the precise best-fit values are input as starting parameter values to a Levenberg-Marquardt fit, for certain data sets and models the fit may run through a large number of redundant loop iterations and produce "SVDCMP" warning messages. Our thanks to Jeremy Sanders for pointing this out. Report added on Mar 06, 2008.

12.4.0o For xmmpsf and suzpsf mixing components, Xspec is unable to parse RA and DEC strings with colon delimiters (entered with the xset command). Report added on Mar 06, 2008.

Note also the there is a problem with suzpsf for all options except that of assuming the surface brightness distribution is centered on the middle of the WMAP. This issue arises because the model was written when the Suzaku XIS spectra used sky coordinates in their WMAP. Since late 2006 we have used detector coordinates but due to an oversight the suzpsf model was not updated.

minor patch for error and named models.

Once the error command is run for a parameter belonging to a named model, it does not allow the user to specify unnamed-model parameters during future error runs. Our thanks to Kip Kuntz for pointing this out. Bug fix 12.4.0l.

background in xspec

I've started a wiki page on how to deal with background in xspec. At the moment it gives an example on how to deal with the case when the background requires a different response matrix to the source.

photos from COSPAR workshop in Alexandria

A selection of my photos from the COSPAR Capacity Building Workshop on X-ray Astronomy held at the New Library of Alexandria are now up on Flickr.

A couple of fakeit notes

Kip has pointed out that fakeit doesn't handle multiple sources correctly. If you set up with two sources with their own responses modelling a dataset then do a fakeit only the first source and response contribute to the simulated spectrum. This will take a bit of work to fix.

Jennifer Donley at Arizona e-mailed to worry about the warning message about incompatible CHANTYPE after simulating Chandra ACIS spectra for CY10. This is a minor bug in xspec which fails to pick up the CHANTYPE value from the response matrix when doing a "fakeit none".