GalFish

Welcome to GalFish


The spectral energy distribution (SED) of a galaxy contains information on the galaxy's physical properties, and multi-wavelength observations are needed in order to measure these properties via SED fitting. GalFish calculates the uncertainties on the parameters of SED fitting, such as age, dust content, and stellar mass, using the Fisher Matrix approximation. This method is extremely quick and can be used to try out different experimental settings to optimize the planned observations according to the desired results.

The uncertainties predicted by GalFish are an approximation of the results one would obtain by performing the observations and using a parameter estimation technique (for example, a Markov Chain Monte Carlo code like GalMC) to analyze the data. This is true for two reasons: one, the Fisher Matrix formalism assumes that the posterior probability distribution is a Gaussian function of the parameters, which leads to optimistic results; two, one always assumes a fiducial model when planning the observations, while in reality it is possible that the models that we use are not a good representation of the data, or that photometric scatter takes a spectrum far away from the fiducial model. The latter factor is a concern for any method that attempts to predict uncertainties before taking the data, and is not specific to the Fisher Matrix.

In Acquaviva et al 2012, we tried to understand how well the Fisher Matrix approximation works, by comparing the uncertainties predicted by GalFish to the ones found using GalMC, first using simulated data (which tests how much we should be concerned with reason #1 above) and then using real data (which tests both reasons). We found that for a large variety of target galaxies, differing in redshift, mass, age, star formation history, dust content, and wavelength coverage, the uncertainties on SED fitting parameters predicted by the Fisher Matrix are comparable (within a factor of two in about the 90% of cases) to the ones obtained by analyzing the data. If you've read this far, you deserve a...

Summary: You can use GalFish to plan and optimize observations, keeping in mind that the errors computed by the Fisher Matrix might be optimistic (at the 20-30% level) and that the fiducial spectra used in planning observations are always an approximation of what the data will look like. Trust the numbers to within a factor of two.

Download journal article: Acquaviva et al, ApJ 749 (2012): Acquaviva_et_al_2012.pdf