Code name:

Galaxy Fitting (gfit)

Code author:

Stephane Paulin-Henriksson (CEA) and Marc Gentile (EPFL)

Method summary:

Shear estimate by fitting a PSF-convolved Sersic galaxy model

More information

The algorithm of the method is the following:

- For each galaxy

     - Extraction of galaxy postage stamps of a specific size

     - FFT convolution of the PSF with a 6-parameter Sersic model of the galaxy.

     - Least squares fitting of the convolved galaxy, over the 6 parameters of the Sersic model.

     - The fitted parameters of the Sersic model are then used as an estimate of the shape and profile of the unconvolved (but sheared) source galaxy.

     - The shear is derived from the ellipticity of the fitted source galaxy

- The same process is repeated for all galaxies in each set. The final shear estimate for a set is the average ellipticity of fitted galaxies on that set.

Additional technical details:
The PSF was modeled as a Moffat whose parameters have been given by the Great08 team.

A galaxy model is assumed where the intensity profile is Sersic and the galaxy shape elliptical. The parameter of the model are then:

    n: Sersic index, re: Sersic radius, q: ellipse axis ratio, theta: position angle between horizontal and semi-major axes, (xc, yc): galaxy centroid
A galaxy is fitted by minimizing the 6-parameter space S (n, re, q, theta, xc, yc) using a Levenberg–Marquardt algorithm (in real space). Initial guesses for (q, theta) are found from quadrupole moment estimate.
Because analytical models for the PSF and galaxies ared used, it is possible to fit at arbitrary higher resolution. On a preliminary version of the pipeline, coded in IDL, we found that increasing the resolution does indeed significantly increase the Q factor. However, this is very CPU consuming. For instance, on the preliminary IDL pipeline, we found that the CPU goes with the resolution at the power k, 3<k<4. This is why this feature was not used in the Great08 submission (i.e. we used a resolution of 1).
No special treatment of the noise is performed.
Other parameters that influence the results are: stamp size, fitting resolution

Parameters used for the submission: stamp size: 25 pixels, resolution: 1

Programming language:

Python 2.5

Required libraries:

Scipy, Numpy, Pyfits, AstroAsciiData

url for specific code:

Runtime per galaxy: 

0.2 sec on average. The program is designed to take advantage of multiprocessor machines for reducing the overall processing.time.

Q from leaderboard:

135.5  (an improved version now reaches Q ~1500 in Low Noise Blind, stamp size: 31, resolution: 1)

Date of leaderboard submission:

Sat 25 Apr 2009

url for specific code:

Runtime per galaxy: 

Q from leaderboard:

32.0

Date of leaderboard submission:

Tue 28 Apr 2009