Galaxy Decomposition Service

The aim of this project is to develop a simple web interface to the fitgal 2D image decomposition program.

* Yogesh Wadadekar, STScI

* Milan Bogosavljevic, Caltech

* Ashish Mahabal. Caltech

Science drivers

Quantitative galaxy morphology is widely used to address scientific questions such as:

* How do Hubble types evolve with redshift? At what redshift does the Hubble sequence form? Does galaxy formation occur heirarchically or through primodial collapse?

* How does the morphology in clusters change as a function of cluster mass and distance from cluster center?

Decomposition code

The galaxy photometric decomposition code fitgal essentially performs a chi^2 minimization between a user provided galaxy image and a multicomponent (bulge,disk,point etc.) model created using an analytical expression for the light distribution of each component. The weights for the minimization are computed using the signal to noise in each pixel of the image. Convolution with a user supplied point spread function is included. The multiparameter fitting is performed using the MINUIT minimization library. The specific algorithm used is Migrad, a variant of the Davidon-Powell-Fletcher method. The algorithm is nearly as fast as the more commonly used Levenberg-Marquardt method, and it is much less liable to get stuck in local minima.

The original version of the code was written in C in 1998. This summer, one of us (YW) rewrote the front end in Python. The codebase went from over 2000 lines in C to only 350 lines of Python code, supported by Python modules (matplotlib - Plotting, numarray - fast array access, pyfits pyfits). The interface between Minuit (written as a c++ library) and our code was developed using the Python.boost library as implemented in the Hippodraw application.

Why another web service?

There already have been attempts to use a grid based approach for studying galaxy morphology (Deelman, 2003). The GALCAS/GALMORPHS web service interfaces to the GIM2D package (using the Metropolis algorithm) running within IRAF on the server. Why then build another?

Obtaining the global minimum of a complicated multidimensional chi^2 surface is something of a black art. Factors such as the choice of the minimisation algorithm, correlated errors between fit parameters, signal to noise of the data can greatly influence the fit results. Having access to more than one service, each implementing a different method would be very useful to the community.

Building a web service: First Steps

We attempted to convert a stand alone program into a Web accessible tool. Due to the limited time available, we took the simplest approach. We wrote a CGI program on Milan's Mac laptop to upload a FITS galaxy image and a FITS PSF image from the user. The CGI program copies the uploaded files and parameter values to a Linux compute server (a modest 1800 MHz AMD Athlon XP machine at STScI). The compute server writes periodic fitting progress reports (as an HTML file) to a Solaris web server at STScI, to which we redirect the user's browser.

We will try to demonstrate this operation during our presentation.

Next steps

With a modest amount of work, we hope to extend this project as follows:

* Improve the web interface to provide users with fine control over parameter ranges and fixing of parameters to specific values. We will also offer users the option of fitting more components.

* Provide rest service (HTTP-GET) access. We are already testing a small client program to do this

* Provide a SOAP client to users after we set up a SOAP server

Bigger goals for the future

In the more distant future this work can be extended by:

* combine this service with GRID resources so that users can process large samples

* We will ourselves attempt to run this code on millions of Sloan survey galaxies using the Teragrid. Our extracted parameters can be added as extra columns to the Sloan catalog and made accessible to the community through Openskyquery.

-- YogeshWadadekar - 14 Sep 2005

• FITGAL.ppt: Slides from our project presentation, 15 Sep 2005

Attachment:	Action:	Size:	Date:	Who:	Comment:
FITGAL.ppt	action	800256	15 Sep 2005 - 15:56	YogeshWadadekar	Slides from our project presentation, 15 Sep 2005