A Bayesian Approach to Gravitational Lens Model Selection, SF2A proceeding

TL;DR

This paper applies Bayesian model selection to gravitational lens modeling, demonstrating that simpler models may be preferable given current data accuracy, thus aiding in cosmological parameter inference.

Contribution

It introduces a Bayesian approach to distinguish between competing lens models, emphasizing the importance of model complexity in gravitational lens analysis.

Findings

Simpler lens models can be sufficient with current data quality.

Bayesian model selection helps avoid overfitting in lens modeling.

More complex models are not always justified by data accuracy.

Abstract

Over the past decade advancements in the understanding of several astrophysical phenomena have allowed us to infer a concordance cosmological model that successfully accounts for most of the observations of our universe. This has opened up the way to studies that aim to better determine the constants of the model and confront its predictions with those of competing scenarios. Here, we use strong gravitational lenses as cosmological probes. Strong lensing, as opposed to weak lensing, produces multiple images of a single source. Extracting cosmologically relevant information requires accurate modeling of the lens mass distribution, the latter being a galaxy or a cluster. To this purpose a variety of models are available, but it is hard to distinguish between them, as the choice is mostly guided by the quality of the fit to the data without accounting for the number of additional parameters introduced. However, this is a model selection problem rather than one of parameter fitting that we address in the Bayesian framework. Using simple test cases, we show that the assumption of more complicate lens models may not be justified given the level of accuracy of the data.