IAEmu: Learning Galaxy Intrinsic Alignment Correlations

The introduction of IAEmu marks a significant advancement in the modeling of galaxy intrinsic alignments, which are critical for weak lensing analyses. Traditional methods often struggle with nonlinear scales or require extensive simulations, making them less efficient. IAEmu, leveraging a neural network approach, achieves remarkable accuracy with an average error of about 3% for position-position correlations and 5% for position-orientation correlations. This emulator not only enhances predictive capabilities but also captures the stochastic nature of orientation-orientation correlations without overfitting. Furthermore, it demonstrates versatility by generalizing to non-HOD alignment signals, as evidenced by its fitting to IllustrisTNG hydrodynamical simulation data. The ability to provide both aleatoric and epistemic uncertainties allows researchers to identify less reliable prediction regions, thereby improving the robustness of cosmological inferences. With a speed-up of approxima…