FitSKIRT: genetic algorithms to automatically fit dusty galaxies with a Monte Carlo radiative transfer code

TL;DR

FitSKIRT employs genetic algorithms to automatically fit complex radiative transfer models to galaxy images, effectively handling high-dimensional parameter spaces and noise, enabling efficient and unbiased analysis of dusty galaxies.

Contribution

This paper introduces FitSKIRT, a novel automated fitting method combining genetic algorithms with Monte Carlo radiative transfer to analyze galaxy images.

Findings

Successfully recovered all model parameters in simulated tests.

Achieved reasonable agreement with previous models of NGC4013.

Demonstrated high automation suitability for large datasets.

Abstract

We present FitSKIRT, a method to efficiently fit radiative transfer models to UV/optical images of dusty galaxies. These images have the advantage that they have better spatial resolution compared to FIR/submm data. FitSKIRT uses the GAlib genetic algorithm library to optimize the output of the SKIRT Monte Carlo radiative transfer code. Genetic algorithms prove to be a valuable tool in handling the multi- dimensional search space as well as the noise induced by the random nature of the Monte Carlo radiative transfer code. FitSKIRT is tested on artificial images of a simulated edge-on spiral galaxy, where we gradually increase the number of fitted parameters. We find that we can recover all model parameters, even if all 11 model parameters are left unconstrained. Finally, we apply the FitSKIRT code to a V-band image of the edge-on spiral galaxy NGC4013. This galaxy has been modeled previously by other authors using different combinations of radiative transfer codes and optimization methods. Given the different models and techniques and the complexity and degeneracies in the parameter space, we find reasonable agreement between the different models. We conclude that the FitSKIRT method allows comparison between different models and geometries in a quantitative manner and minimizes the need of human intervention and biasing. The high level of automation makes it an ideal tool to use on larger sets of observed data.