Lensed: a code for the forward reconstruction of lenses and sources from strong lensing observations

TL;DR

Lensed is a GPU-accelerated code for fast, accurate, and simultaneous Bayesian modeling of strong lensing systems, enabling detailed reconstruction of lens and source properties from observational data.

Contribution

The paper introduces Lensed, a novel GPU-based forward modeling code that efficiently performs Bayesian inference on strong lensing data, handling multiple parameters simultaneously.

Findings

Successfully recovers unbiased lens parameters from mock data.

Demonstrates consistent results with existing literature on real SLACS lenses.

Offers a flexible and robust approach for strong lens modeling.

Abstract

Robust modelling of strong lensing systems is fundamental to exploit the information they contain about the distribution of matter in galaxies and clusters. In this work, we present Lensed, a new code which performs forward parametric modelling of strong lenses. Lensed takes advantage of a massively parallel ray-tracing kernel to perform the necessary calculations on a modern graphics processing unit (GPU). This makes the precise rendering of the background lensed sources much faster, and allows the simultaneous optimisation of tens of parameters for the selected model. With a single run, the code is able to obtain the full posterior probability distribution for the lens light, the mass distribution and the background source at the same time. Lensed is first tested on mock images which reproduce realistic space-based observations of lensing systems. In this way, we show that it is able to recover unbiased estimates of the lens parameters, even when the sources do not follow exactly the assumed model. Then, we apply it to a subsample of the SLACS lenses, in order to demonstrate its use on real data. The results generally agree with the literature, and highlight the flexibility and robustness of the algorithm.