GENERAL GALFIT FAQ

What is GALFIT? In a nutshell:

• GALFIT is a data analysis algorithm that fits 2-D analytic functions to galaxies and point sources directly to digital images. A detailed description of the GALFIT technique can be found in Peng, Ho, Impey, & Rix (2002, Astronomical Journal, 124, 266), with new features found in Peng, Ho, Impey, & Rix (2010, Astronomical Journal, 139, 2097).

• The functions used are some of the most commonly found in astronomy literature, including: exponential, Sérsic/de Vaucouleurs, Nuker, Gaussian, King (v. 2.0+), Moffat, and PSF (v. 1.9a+). The PSF function is provided by the user.

• Allows for simultaneous fitting of arbitrary number of components and combination of the above functional forms. More functions may be added in the future based on user recommendations.

• The default models are ellipsoidal in shape. However, as of Version 3.0+ (Peng, Ho, Impey, & Rix 2010) much more detailed shapes are allowed, including Fourier modes, bending, truncation, etc..

• The model components can be completely independent, constrained or coupled to one another. Parameter coupling/constraints should be used with caution, however.

• GALFIT is pretty fast, compared to other similar 2-D galaxy fitting algorithms I know about. The minimization engine is based on the Levenberg-Marquardt (L-M) algorithm (downhill gradient).

• The models are convolved with the point spread function using Fast Fourier Transform to take into account the seeing.

• GALFIT is written in the standard, garden-variety, C. There are no plans on porting it to C++ or Fortran77.

• Most users have found it very easy to both install and use.

• GALFIT runs on Redhat, Ubuntu, and Suse Linux distributions, Unix, and MacOS X (versions more recent than v10.2, i.e. ``Jaguar'').

How is GALFIT different from GIM2D?

• Please see GIM2D Home Page about details and features of GIM2D not available in GALFIT, as I'm not an expert on GIM2D. However, I can say a few words about the difference between GALFIT and GIM2D, namely...

• The GALFIT minimization engine is based on the Levenberg-Marquardt algorithm (downhill gradient), while GIM2D is based on the Metropolis algorithm (simulated annealing).

• GALFIT can fit more than 2 functions simultaneously. In fact, GALFIT can in principle fit an arbitrary number of functions simultaneously.

• GIM2D allows users to fit stacks of images simultaneously. Work is in progress to implement capabilities to fit image data cubes in Version 4 of GALFIT.

• GALFIT allows parameters of the components to be explicitly coupled or be independent. But the user should use parameter coupling/constraints with caution.

• GIM2D was specifically designed to be fully automated on B/D decomposition in galaxy surveys. On the other hand, GALFIT was designed to be maximally flexible, fast, and easy to use. The difference means that whereas GIM2D does a lot of pre- and post-processing for users, GALFIT does one thing and one thing only, which is to fit galaxies. It does not extract all conceivable information out of an image, but it provides residual and model images for users to do so. Also, GALFIT was not originally intended to be fully automated. Having said this, a number of people (myself included) have adapted GALFIT (via wrapping scripts) to do distant galaxies in fully automated batch mode (see GALFIT in Action on distant galaxies). One example is a software called GALAPAGOS, developed by a colleague, Marco Barden.

• The design of GALFIT makes it possible to analyze nearby, highly resolved, galaxies, in an efficient and straightforward manner. One of the goals was to study the properties of nearby galaxies in great detail (see GALFIT in action on nearby galaxies), such as their nuclear profiles, detailed sub-structures (bars, embedded disks, AGNs, B/D decomposition etc.).

• GALFIT is dynamic and flexible for dealing with complicated situations. It can model detailed shapes like spiral structures, higher order shape deviations from an ellipse, rings, truncation, and bending modes, in ways previous not possible using new features in Version 3.

What are the limitations of GALFIT?

• Q: What can or can't GALFIT deal with?

  A: GALFIT can always fit something, but not all fits are physically meaningful, and what's meaningful or not is often open to interpretation. Many different kinds of non-axisymmetric galaxies can be fit well, especially using new features or using multiple components. The current real limitations are perturbations in spiral arms, or spiral winding that is not logarithmic or hyperbolic-tangent in radius, e.g. a spiral arm that wind back onto itself. Despite these limitations, most of the time GALFIT will allow one to do much more than the actual science requires.

• Unlike GIM2D, GALFIT was not initially conceived to do automated batch fitting of galaxies. However, there are ways for GALFIT to do so by writing a ``wrapping script.'' There are now several wrapping scripts available created by individual groups, e.g. GEMS, that do this efficiently and seamlessly. One of them is available in the GALFIT package. (See GALFIT In Action.)

• When it comes to the difference between a downhill-gradient method and a metropolis technique (e.g. GIM2D), the conventional wisdom is that a gradient algorithm is less robust. However, because no algorithm is fool-proof, and because galaxy fitting has a number of subtleties that can only be learned by experience, the decision about which algorithm to use should be based on broader consideration about the application, rather than conventional wisdom. In realistic simulation studies created to test limitations of GALFIT vs. GIM2D, GALFIT performed just as well, if not better, as discussed in this published study.

How fast is GALFIT?

• In absolute terms, back in the good ol' days of year 2000 AD, GALFIT can fit a single galaxy (200×200 pixels) with Sérsic function, in ~30 - 60 seconds on a Pentium 833 MHz (over half this time is spent doing convolution).

• In relative terms, it's at least 20 times faster than GIM2D, and sometimes over 100 times.

• For other benchmarks, please see Peng, Ho, Impey, & Rix 2002, AJ, 124, 266.

Are 1-D algorithms better than 2-D at fitting peculiar galaxies or low surface brightness regions?

This falls under the technical FAQ section, which is located here.

Are 1-D algorithms more robust against neighboring contamination than 2-D fitting techniques?

This again falls under the technical FAQ section, which is located here.

Are you going to automate GALFIT for batch fitting of galaxies?

• That is not in the plans and, in fact, it's the only course I am careful to avoid. I feel fairly strongly about the need to provide user support for something that's potentially complex once it is released out to the public. So my main worry is that I won't have sufficient time, on top of GALFIT, to provide such support for the automated algorithm as well. There will no doubt be difficult issues that require fixing and debugging, and these are not easy to anticipate or address for everyone in a general manner. Therefore, users are encouraged to devise their own algorithms (via, for example, wrapper scripts), so they may tailor them to their own needs. If help is needed to diagnose problems during this process, feel free to contact me.

• I encourage people who wish to automate GALFIT do so by writing their own ``wrapping scripts'' instead of modifying GALFIT itself. Wrapper scripts are much simpler and more elegant, which allows users to more easily customize and optimize the fitting outcome themselves. Wrapper scripts often also have useful I/O subroutines written for them that facilitate many tedious tasks that will be involved in this process. A really good algorithm was developed for this purpose by Marco Barden called GALAPAGOS.

• I also encourage those who write their own automated wrappers to contact me if they do not understand certain GALFIT behaviors found during that fitting process.

• Having wiped my hands clean from touching this issue, I do provide a crude example of automated batch fitting in the GALFIT directory (see the galfit.pl and galfit.cl tasks in the GALFIT directory) for those who wish to learn the basics of how to do so. This example may not work right away without some tweaking to make sure the correct GALFIT input file is generated.

Should I fit spiral arms or higher order modes?

• The only way to answer that question is to clearly identify the science goal and judge whether the science results will be affected by doing something simple.

• For most science goals, especially to measure average, global, galaxy morphology (e.g. luminosity, average size, concentration), the answer is ``no." Classical ellipsoid models often will do just fine, even if they do not look particularly realistic. That is a good thing, because detailed analysis using spiral structures can be really tough to do.

• If you are interested in accurate decomposition of something that's embedded inside a galaxy, e.g. bulge/disk/bar, then you may want to consider going into the gory analysis. Doing so can be an infinite time sink, so be sure that the science goal requires it rather than that it is a nice thing to do for all objects.

• If you're now wondering how to fit spiral structures, go here.

Do you provide user support/updates?

• Absolutely -- without conditions, especially if there is a suspicion of a problem (e.g. systematic errors, segmentation fault during run-time, etc.) please contact me at to report it first. I answer all email queries regarding GALFIT. So if you don't hear from me, it's very likely I didn't get your message. I generally respond quite quickly, so if you don't receive a message after a day, then your message didn't get to me. Try again and make sure you have the correct email address and remove things in the message which may be flagged as spam.

• Users are very much welcomed to contact me to also be placed on my user list. I will use it only to inform users when there are important updates or bug fixes. The list won't be redistributed in any way or for any other purpose.

• One reason why GALFIT is regularly updated is that I try to keep up-to-date on numerical techniques to make it run faster and better, and to facilitate GALFIT for other uses. I also look forward to hearing suggestions from users on how to improve GALFIT, on all issues, including usability and techniques.

• There is also now a Facebook page for updates, questions, and discussions about GALFIT in case you want to get help from a larger user group.

• I always welcome knowing about bugs that users find, no matter how trivial it might seem. Any user who suspects there might be a bug in GALFIT is urged to contact me. I know I might not have tested everything myself. I will be very grateful for your keen eye.

• Data and the science which GALFIT users share with me in the course of asking for help are, of course, kept confidential. The proprietary status of the data and science will be fully respected.

DISCLAIMERS, USER AGREEMENT and COPYRIGHTS

• The reason for this formality is that I have (as most of us have) legal bindings to our institutions (for me it's the University of Arizona) more than anything else. So here goes....

• There may be other software by the same name (in the past or the future). The agreements below obviously apply only to the software specifically published in Peng, Ho, Impey, & Rix 2002, AJ, 124, 266 (note, it's AJ, not ApJ), and described above, which is provided to the public at large free of charge, and without further obligation.

• Users of this code do so at own risk and should understand this is work in progress. There will likely be future updates. I very much welcome suggestions on how to improve GALFIT and fix potential bugs. If there's a bug or something important which you don't understand feel free to contact me (). Bug reports are always greatly appreciated! (But please read the Technical FAQ first to see if others have asked the same questions.)

• Users are encouraged, but not required, to e-mail me to be put on the user list for updates and upgrades.

• Users are encouraged to check back regularly to download the newest version of this code.

• The source code is not distributed because portions of it are modified from copyrighted codes. Work is currently being done to replace those algorithms by better or equivalent codes. Once that is done, GALFIT will be open source. But at the moment it is a fairly low priority.

Go back to GALFIT Home Page.
Go back to Chien Peng's Home Page.

---

Chien Peng ()