Astrophysicists are interested in recovering the 3D gas emissivity of a galaxy cluster from a 2D telescope image. Blurring and point sources make this inverse problem harder to solve. The conventional approach requires in a first step to identify and mask the point sources. Instead we model all astrophysical components in a single Poisson generalized linear model. To enforce sparsity on the parameters, maximum likelihood estimation is regularized with two ${\ell }_1$ penalties with weights λ₁ for the radial emissivity and λ₂ for the point sources. The method has the advantage of not employing cross validation to select λ₁ and λ₂. To judge the significance of interesting features, we quantify uncertainty with the bootstrap. We apply our method to two X-ray telescopes (XMM-Newton and Chandra) data to estimate gas emissivity. The results are more stable and seems less biased than the conventional method, in particular in the outskirt of galaxy clusters.

天体物理学家对从2D望远镜图像恢复星系团的3D气体发射率感兴趣。模糊和点源使这个反问题更难解决。传统方法在第一步中需要识别并掩盖点源。取而代之的是，我们在单个Poisson广义线性模型中对所有天体物理分量进行建模。为了对参数实施稀疏性，将最大似然估计调整为两个${\ell }_{\mathrm{1个}}$与重量处罚λ ₁用于径向发射率和λ ₂为点源。该方法具有不采用交叉验证来选择的优点λ ₁和λ ₂。为了判断有趣功能的重要性，我们使用自举量化了不确定性。我们将我们的方法应用于两个X射线望远镜（XMM-Newton和Chandra）数据以估计气体发射率。结果比常规方法更稳定，而且似乎偏差更少，特别是在星系团的郊区。