TL;DR
This report outlines a strategic roadmap for the "Tri-Agency" (DOE, NASA, NSF) to coordinate the massive cosmological simulations required for the next generation of sky surveys: the Vera C. Rubin Observatory, Euclid, and the Nancy Grace Roman Space Telescope (formerly WFIRST). It argues that simulations must transition from fragmented R&D projects into shared, essential infrastructure to overcome the "systematics" bottleneck and maximize the scientific return of these multi-billion dollar missions.
What Problem Does This Paper Address?
Cosmology is entering the "Stage 4" era, where the limiting factor for discovering new physics (like the nature of dark energy) is no longer the amount of data, but our ability to model it. To interpret observations from upcoming wide-field surveys, scientists need synthetic universes—simulations that replicate the distribution of billions of galaxies.
Historically, these simulations have been developed by isolated groups with short-term R&D funding. However, the scale required today is staggering: we need volumes of several Gigaparsecs with enough resolution to "see" small galaxies. This creates three main challenges:
- Redundancy: Different agencies are currently "reinventing the wheel" for similar simulation needs.
- Systematic Errors: Complex astrophysical processes (like gas cooling and supernova feedback) can mimic or mask cosmological signals.
- Human Capital: The field is losing specialized talent to data science industries because simulation work is often viewed as "infrastructure" rather than "pure research," leading to a lack of stable career paths.
Methods
The Task Force categorizes the necessary simulation efforts into two primary tiers:
- Extreme-scale Simulations: "Grand challenge" runs (like the Euclid Flagship or Outer Rim simulations) that use millions of GPU/CPU hours to create a single, high-resolution realization of the universe. These serve as the "base" for synthetic sky maps.
- Large Simulation Campaigns: Ensembles of hundreds or thousands of medium-resolution simulations used to build Emulators. These emulators act as computationally cheap surrogate models, allowing researchers to predict the universe's appearance across a wide range of theoretical parameters without running a new full-scale simulation every time.
The report also evaluates the transition from Gravity-only (N-body) simulations to Hydrodynamic simulations, which include the complex "baryonic" physics of gas, stars, and black holes.
Datasets
The report references several landmark simulations that currently define the state-of-the-art:
- Euclid Flagship Simulation: A 3.78 $h^{-1}$Gpc box with 2 trillion particles.
- Outer Rim Simulation: A 3 $h^{-1}$Gpc box used by the Rubin Observatory’s Dark Energy Science Collaboration (DESC).
- Dark Sky Simulation: An 8 $h^{-1}$Gpc box (large volume but lower mass resolution).
- Hydrodynamic Benchmarks: Projects like EAGLE, IllustrisTNG, and BAHAMAS are used to calibrate how gas and stars affect the underlying dark matter distribution.
Key Insights and Ideas
- Simulations as Infrastructure: The authors emphasize that simulations are now as critical to a mission's success as the telescope's mirrors or cameras. They should be funded and managed as core infrastructure.
- The "Baryon" Problem: On small scales, the way gas moves and stars explode significantly alters the matter power spectrum. If not modeled correctly, this "baryonic feedback" will lead to biased conclusions about dark energy.
- Cross-Survey Synergy: Since Rubin (ground-based), Euclid, and Roman (space-based) will often look at the same patches of sky, they should use shared synthetic catalogs to ensure cross-correlation results are consistent.
- Joint Pixel Analysis: There is a massive potential gain in "joint pixel" efforts, where data from different telescopes are analyzed together at the image level, but this requires shared, high-fidelity simulations to validate the pipelines.
Key Results
The Task Force delivered a phased implementation plan for the agencies:
- Phased Development: Each simulation effort (Extreme-scale, Campaigns, and Systematics) is broken into three phases: Definition (~6–12 months), Tooling/Validation (~12–24 months), and Implementation/Production (2021 and beyond).
- Resource Requirements: To reach the necessary accuracy (e.g., 1% accuracy in the matter power spectrum), the community needs access to exascale computing facilities and petabyte-scale storage.
- Infrastructure Models: The report highlights existing successful models for data sharing, such as PIC (Port d’Informació Científica) in Spain and the Data-Scope at Johns Hopkins University, as templates for a common Tri-Agency infrastructure.
Limitations and Future Work
- Funding Gaps: The primary limitation is the lack of a dedicated funding stream for cross-agency simulation work. Current grants are often too small or too specific to support the broad infrastructure needed.
- Hydrodynamic Scaling: While N-body simulations are relatively mature, hydrodynamic simulations are still too computationally expensive to run at the volumes required for Stage 4 surveys.
- Sub-grid Modeling: Our understanding of "sub-grid" physics (processes smaller than the simulation's resolution, like individual star formation) remains a major source of uncertainty that requires more iterative comparison with real-world data.
Summary updated