Speaker
Description
Cosmological hydrodynamical simulations are powerful tools for studying galaxy formation, yet their predictive precision is limited by stochastic variability and numerical uncertainty. We quantify this variability using four identical realizations of a zoom-in galaxy-cluster simulation evolved with OpenGadget3 under tightly controlled compiler, library, and hardware settings. Variability is measured through the properties of matched galaxies across repeated runs, including a mixed linear model that separates run-to-run variation from within-run noise.
Variations of the order of $\sim10$–$25\%$ are found in galaxy dark matter and stellar masses for the baseline simulations. The variability trending above the shot-noise floor reflects the combined effects of stochastic star formation and feedback regulation, and is further amplified when black hole physics is included.
Furthermore, our results indicate that feedback acts to regulate variability, reducing scatter in both stellar and black hole masses. Our inference from run-to-run variation indicates a noise-dominated regime that remains statistically reproducible despite clone-to-clone differences.
These results establish baseline, noise-dominated variability estimates at low resolution, demonstrate how feedback modulates predictability, and provide a statistical framework for future studies of reproducibility in cosmological hydrodynamical simulations.
| Affiliation of the submitter | INAF Trieste |
|---|---|
| Attendance | remote |
