Loving ISIS - Confessions of a Former XSPEC User
ISIS vs. XSPEC 12:
XSPEC has recently undergone a major upgrade to version 12. This has given it a number of abilities not present in XSPEC v. 11. Quite a large portion of this list, however, has been available in ISIS for several years now. For example: support for OGIP Type II files; support of linear combinations of RMFs comprising a response (e.g., for LETG spectra); linking model parameters via polynomial functions (in fact, ISIS model parameters can be arbitrary functions of other parameters - you want to use a hypergeometric function? Download the GSL module and go for it!); extensible fitting methods and fit statistics; easy incorporation of user models, etc.
Except that ISIS doesn't stop there. Thanks to its programmability, it
can do a lot more. For example, I use ISIS for
timing analysis
(obviating any need to use a separate timing package, such as
XRONOS,
in addition to a spectral fitting package). The power spectrum shown
below was performed entirely in ISIS. The data was read, the
FFTs were performed and averaged, the Power Spectrum was
logarithmically binned over Fourier frequency, the constant + two
Lorentzian model was fit, and the results plotted, all in ISIS without
ever leaving the program.
Have people done similar things in XSPEC? Sure. But the procedure there usually involves something like reading the data, and calculating and averaging the power spectrum, in a separate package such as IDL. The intermediate results are then written to a file. A fake, diagonal response matrix is created, and then the data is read in and fit in XSPEC. ISIS removes a lot of those intermediate steps, and it doesn't require the creation of a response matrix (a diagonal one is automatically assumed if no response is assigned).
With the use of the S-Lang Modules Packages, lots of other cool things can be done. For example, the XPA module allows 2D arrays to be passed back and forth between ISIS and the DS9 imaging package. (I have plans to incorporate this into a 'dynamical power spectra' viewer.)
The Parallel Virtual Machine (PVM) module allows extremely time consuming calculations (which are becoming more and more common in astrophysics) to be done quickly and efficiently. Here is an example of the PVM module being used to produce a temperature map of an X-ray cluster. Additionally, the PVM module has recently been used to do a pixel-by-pixel spectral fit to the Chandra megasecond observation of the Cas A supernova remnant.
The PVM module also opens up the exciting possibility of parallelized fit functions and parallelized error bar searches. There are lots of examples of complicated astrophysical fit functions that can be broken up into independent pieces which can then be farmed out to networks of machines (which most physics and astronomy departments have these days - with lots of CPU cycles sitting idle). Work is going on here to provide a step-by-step outline to show how this can be done. For some test cases run here, we've seen factors of 30 improvements in speed; i.e., fits that are normally done in minutes occurring in seconds. Links, when available, will be placed in the Power Tools directory.
XSPEC 12 introduces some extensibility into that spectral fitting
package. The S-lang scripting language, plus the
SLIRP code generator, on
the other hand, allows a much greater breadth, depth, and ease of
extensibility to ISIS. From my own personal perspective, everything
that I normally would want to do in XSPEC 12, I can do in ISIS. There
are things that I can do in ISIS that I can't easily do in XSPEC 12.
And looking a year or two down the line to things I want to do, I see
a much better path forward with ISIS.
For a further comparison between ISIS and XSPEC, see this ISIS web page.
