We introduce a new adaptive and fully Bayesian grid-based method to model
strong gravitational lenses with extended images. The primary goal of this
method is to quantify the level of luminous and dark-mass substructure in
massive galaxies, through their effect on highly-magnified arcs and Einstein
rings. The method is adaptive on the source plane, where a Delaunay
tessellation is defined according to the lens mapping of a regular grid onto
the source plane. The Bayesian penalty function allows us to recover the best
non-linear potential-model parameters and/or a grid-based potential correction
and to objectively quantify the level of regularization for both the source and
the potential. In addition, we implement a Nested-Sampling technique to
quantify the errors on all non-linear mass model parameters -- ... -- and allow
an objective ranking of different potential models in terms of the marginalized
evidence. In particular, we are interested in comparing very smooth lens mass
models with ones that contain mass-substructures. The algorithm has been tested
on a range of simulated data sets, created from a model of a realistic lens
system. One of the lens systems is characterized by a smooth potential with a
power-law density profile, twelve include a NFW dark-matter substructure of
different masses and at different positions and one contains two NFW dark
substructures with the same mass but with different positions. Reconstruction
of the source and of the lens potential for all of these systems shows the
method is able, in a realistic scenario, to identify perturbations with masses
>=10^7 solar mass when located on the Einstein ring. For positions both inside
and outside of the ring, masses of at least 10^9 solar mass are required (...).

介绍一种新的自适应全贝叶斯网格方法，用于建模具有扩展图像的强引力透镜，通过高倍率放大的弧和爱因斯坦环的效应来量化大质量星系中的明亮和暗质量亚结构水平，并使用拟合广义线性模型的贝叶斯惩罚函数去修正潜在的网格潜在的亚结构，实现源和潜在的正则化多样性的客观评估；基于 Nested-Sampling 技术，实现了对所有的非线性质量模型参数的误差量化，能将不同的势能模型根据边缘化证据进行客观排名，其精度能检测 Einstein 环上质量大于等于 10 ^ 7 太阳质量的扰动。