The nature of dark matter remains one of the greatest unanswered questions in science. The largest concentrations of dark matter appear to lie in galaxy clusters. By modifying the properties of dark matter, the distribution of mass in clusters is altered in an observable way. However, uncertain astrophysical mechanisms also alter the mass distribution, often mimicking the effect of different dark matter properties. Here I present a machine learning method that ‘learns’, from simulations, how the impact of dark matter self-interactions differs from that of astrophysical feedback. In the idealized case, my algorithm is 80% accurate at identifying whether a galaxy cluster harbours collisionless dark matter, dark matter with a self interaction cross-section, σ_DM/m = 0.1 cm² g⁻¹ or dark matter with σ_DM/m = 1 cm² g⁻¹. It is found that weak-lensing information primarily differentiates self-interacting dark matter, whereas X-ray information disentangles different models of astrophysical feedback. The data are forward modelled to imitate observations from Euclid and Chandra, and it is found that the model has a statistical error of σ_DM/m < 0.01 cm² g⁻¹ and is insensitive to shape-measurement bias and photometric-redshift errors. This method represents a way to analyse data from upcoming telescopes that are an order of magnitude more precise and many orders faster than current methods, enabling us to explore the properties of dark matter like never before.

暗物质的本质仍然是科学中尚未解答的最大问题之一。暗物质的最大浓度似乎位于星系团中。通过改变暗物质的性质，团簇中的质量分布以可观察的方式改变。然而，不确定的天体物理机制也会改变质量分布，通常模仿不同暗物质特性的影响。在这里，我提出了一种机器学习方法，可以通过模拟“学习”暗物质自相互作用的影响与天体物理反馈的影响有何不同。在理想情况下，我的算法在识别星系团是否包含无碰撞暗物质、具有自相互作用截面的暗物质σ _DM / m = 0.1 cm ² g ⁻¹或具有σ _DM / 的暗物质时准确率为 80% m = 1 cm ² g ⁻¹ 。研究发现，弱透镜信息主要区分自相互作用暗物质，而 X 射线信息则解开不同的天体物理反馈模型。对数据进行正向建模以模仿Euclid和Chandra的观测结果，发现该模型的统计误差为σ _DM / m < 0.01 cm ² g ⁻¹并且对形状测量偏差和光度红移误差不敏感。这种方法代表了一种分析来自即将到来的望远镜的数据的方法，这些望远镜比当前的方法更精确一个数量级，速度更快许多数量级，使我们能够以前所未有的方式探索暗物质的特性。