Useful tools for working with CosmoGridV1

This page contains non-exhaustive list of tools that can be useful at different stages of simulation-based inference with CosmoGridV1.

UFalcon - creating probe maps from lightcone shells

To create a new set of probe maps for a given set of redshift distribution n(z), one needs to calculate the corresponding probe integration kernels. The publicly available UFalcon (Ultra Fast Lightcone) code allows for just that. UFalcon is described in these papers: Sgier et al. 2019 and Sgier et al. 2020. The code is available from the ETH Zurich public Gitlab repository cosmo-gitlab.phys.ethz.ch/cosmo/UFalcon. More information about UFalcon can be found on the UFalcon Homepage.

DeepSphere - convolutional neural network for Healpix maps

While CNNs were developed mainly for natural images, we created a CNN that work on the sphere, called DeepSphere. It is described in this paper: arxiv.org/abs/1810.12186. DeepSphere is one of the fastest available spherical convolutions, although has its limitations. The public code for Tensorflow 1/2, and PyTorch, is available from this repository: github.com/deepsphere.

PyCosmo - cosmological calculator and Boltzman solver

Cosmological background parameters, distances, and perturbations are computed analytically. The package PyCosmo uses a fast Boltzman solver to achieve this, including new cosmological models. It is available at cosmo-docs.phys.ethz.ch/PyCosmo. A live cosmological calculator in Jupyter notebooks is avalilable at PyCosmoHub. The public release is described in this paper Tartsitano et al. 2020.

PkdGrav3 - fast N-body simulations

The CosmoGridV1 simulations were run using the PkdGrav3 code developed at the University of Zurich. PkdGrav3 is one of the fastest N-body simulation codes available, efficiently capitalizing on the GPU acceleration. It is used to create large simulations Potter et al. 2016. The code is available at the PkdGrav3 repository.