KiDS-1000 cosmology: machine learning - accelerated constraints on interacting dark energy with CosmoPower
KiDS-1000 cosmology: machine learning - accelerated constraints on interacting dark energy with CosmoPower
复制标题
KiDS-1000 宇宙学：机器学习 - 加速限制暗能量与 CosmoPower 的相互作用
DOI：
10.1093/mnrasl/slac019
复制
发表时间：
2022
期刊：
影响因子：
--
通讯作者：
Spurio Mancini A
中科院分区：
文献类型：
--
作者：
Spurio Mancini A
我们从Kilo-Degree Survey和thePlanck 2018宇宙微波背景数据的公共1000 deg 2宇宙剪切测量中获得了对具有纯动量交换的耦合精粹模型的约束。我们将这个模型与Lambda冷暗物质进行比较，发现相似的χ 2和对数证据值。我们通过从神经网络仿真器CosmoPower中获取宇宙学功率谱来加速参数估计。我们强调这种基于仿真器的方法，以减少未来类似的分析，特别是从第四阶段调查的计算运行时间的必要性。作为一个例子，我们提出了马尔可夫链蒙特卡罗预测相同的耦合的精粹模型aEuclide-like调查，揭示耦合的精粹参数和重子反馈和内在的对齐参数之间的简并，但也突出了约束功率的大幅增加阶段IV调查将实现。与玻尔兹曼代码所需的几个月相比，CosmoPower在几个小时内就可以获得轮廓。
We derive constraints on a coupled quintessence model with pure momentum exchange from the public ∼1000 deg2cosmic shear measurements from the Kilo-Degree Survey and thePlanck2018 cosmic microwave background data. We compare this model with Lambda cold dark matter and find similar χ2and log-evidence values. We accelerate parameter estimation by sourcing cosmological power spectra from the neural network emulatorCosmoPower. We highlight the necessity of such emulator-based approaches to reduce the computational runtime of future similar analyses, particularly from Stage IV surveys. As an example, we present Markov Chain Monte Carlo forecasts on the same coupled quintessence model for aEuclid-like survey, revealing degeneracies between the coupled quintessence parameters and the baryonic feedback and intrinsic alignment parameters, but also highlighting the large increase in constraining power Stage IV surveys will achieve. The contours are obtained in a few hours withCosmoPower, as opposed to the few months required with a Boltzmann code.