Fast and Differentiable Emulator for the Effective Field Theory of the Universe (arxiv.org)

The authors released Effort (EFfective Field theORy surrogaTe), a fast, differentiable emulator for the Effective Field Theory of Large-Scale Structure (EFTofLSS). Effort combines modern numerical algorithms and preprocessing strategies to produce rapid model evaluations while retaining the fidelity required for cosmological inference. Crucially, the emulator is differentiable, enabling gradient-based sampling methods (e.g., Hamiltonian Monte Carlo) and tighter integration into end-to-end pipelines for parameter estimation and model comparison. Validation shows Effort reproduces Bayesian posteriors obtained with the established pybird code: HMC samples using Effort match Metropolis–Hastings results from pybird on large-volume simulations and the BOSS dataset, with deviations consistent with Monte Carlo noise. That agreement demonstrates the emulator’s accuracy alongside its computational advantages. For the AI/ML and cosmology communities, Effort lowers the cost of sampling high-dimensional posterior spaces, supports gradient-based optimization and uncertainty quantification, and is positioned to speed analyses of next-generation surveys and joint analyses with complementary tools.