Perfectly parallel cosmological simulations using spatial comoving Lagrangian acceleration

Abstract

Context. Existing cosmological simulation methods lack a high degree of parallelism due to the long-range nature of the gravitational force, which limits the size of simulations that can be run at high resolution. Aims. To solve this problem, we propose a new, perfectly parallel approach to simulate cosmic structure formation, which is based on the spatial COmoving Lagrangian Acceleration (sCOLA) framework. Methods. Building upon a hybrid analytical and numerical description of particles’ trajectories, our algorithm allows for an efficient tiling of a cosmological volume, where the dynamics within each tile is computed independently. As a consequence, the degree of parallelism is equal to the number of tiles. We optimised the accuracy of sCOLA through the use of a buffer region around tiles and of appropriate Dirichlet boundary conditions around sCOLA boxes. Results. As a result, we show that cosmological simulations at the degree of accuracy required for the analysis of the next generation of surveys can be run in drastically reduced wall-clock times and with very low memory requirements. Conclusions. The perfect scalability of our algorithm unlocks profoundly new possibilities for computing larger cosmological simulations at high resolution, taking advantage of a variety of hardware architectures.

Nonlinear 3D cosmic web simulation with heavy-tailed generative adversarial networks

“…(4) The discriminator predictions are compared to the ground truth labels, from which a loss LðD; GÞ is computed. (5) The loss is used to update the weights of G and D through backpropagation, flowing through the discriminator and then through the generator. (6) Steps 1-5 are repeated, looping through the training set over several epochs.…”

“…The discriminator network outputs a set of scores representing how much it believes the samples come from the data distribution, and these scores are compared with ground truth labels to calculate a loss (4). This loss is then used to update the parameters of both the generator and discriminator network through backpropagation (5).…”

“…3D visualizations of GADGET subvolumes evolved to redshift z ¼ 0 (top row) and samples generated with a DCGAN (bottom row). Each subvolume is visualized by plotting a number of points within the volume with size R ¼ ðcðxÞ þ 1Þ 5 , where cðxÞ is the scaled density from (5) for a given location. This is done to highlight the relevant features of large-scale structure.…”

“…The need for fast and accurate simulations at a cheaper computational cost has led to the development of many "approximate" methods of cosmic structure simulation [3][4][5][6][7][8][9][10][11], which implement various physical approximations of gravity to decrease the simulation run-time and/or memory requirement while attempting to maintain a high degree of accuracy. These methods have shown to be very useful, but they still require a non-negligible amount of computation time, and the accuracy of their results has room to be improved upon.…”

Bayesian forward modelling of cosmic shear data

“…Initial and evolved density fields are linked by deterministic gravitational evolution mediated by various physics models of structure growth. Specifically, BORG incorporates several physical models based on Lagrangian Perturbation Theory (LPT), fully non-linear particle-mesh models , a model based on spatial COmoving Lagrangian Acceleration framework (Leclercq et al 2020), and a semiclassical analogue to LPT (Porqueres et al 2020). While in this work, we used LPT to approximately describe gravitational clustering, any of these dynamical models can be straightforwardly employed within the flexible block sampling illustrated in Fig.…”