Building the CosmicGrain Zoom Suite¶

This page records the current production workflow for selecting target halos from the 50 \(h^{-1}\,\mathrm{Mpc}\) dark-matter-only parent run and generating CosmicGrain-ready MUSIC2 initial conditions. The older Halo 569 workflow was useful during development, but the current science suite consists of 12 halos at four resolution levels.

Current suite¶

The selected parent-catalog halo IDs are:

295  308  441  859  1481  1534
3352  3879  3886  5834  7723  9235

Each halo has ICs at nominal effective resolutions \(512^3\), \(1024^3\), \(2048^3\), and \(4096^3\), for 48 files in total. The canonical filename is:

ICs/halo<HALO>/IC_halo<HALO>_zoom_<RES>.hdf5

The suite spans dwarf through super-Milky-Way halo masses. Candidate selection also considers isolation, distance from the periodic box boundary, particle count, and suitability of the traced Lagrangian region.

1. Run and catalog the parent volume¶

Run the \(512^3\) dark-matter-only parent calculation through \(z=0\), with FOF and SUBFIND enabled at the output used for selection. The parent and zoom calculations use:

Quantity	Value
Box size	\(50\,h^{-1}\,\mathrm{Mpc}\) comoving
\(\Omega_{\rm m}\)	0.3158
\(\Omega_\Lambda\)	0.6842
\(\Omega_{\rm b}\) for baryonic zoom ICs	0.04936
\(h\)	0.6732
\(n_s\)	0.965
\(\sigma_8\)	0.811
Initial scale factor	0.01 (\(z=99\))

The parent run itself has OmegaBaryon 0.0 because it contains only dark matter. MUSIC2 must use the physical baryon density when splitting the high-resolution matter into gas and dark matter.

2. Build and rank the halo census¶

From scripts/:

python3 parent_halo_census.py
python3 select_zoom_halo_candidates_v2.py

The census and ranking products are:

parent_50Mpc_halo_census.csv
zoom_halo_candidates_all_ranked.csv
zoom_halo_candidates_selected.csv
zoom_halo_candidates_selected.txt

Selection should not be based on mass alone. Inspect the isolation metrics, edge distance, particle count, and neighboring massive halos before accepting a target.

3. Prepare and trace the Lagrangian regions¶

Prepare the selected parent-halo particle sets:

python3 prepare_lagrangian_particle_sets_v2.py

Then trace those particle IDs to the parent initial conditions:

python3 trace_lagrangian_regions_to_initial.py

The corresponding utilities and outputs live under:

scripts/lagrangian_particle_sets/
scripts/lagrangian_regions_initial/
scripts/lagrangian_regions_minimal/

Before generating expensive ICs, inspect the traced extents:

python3 diagnose_lagrangian_region_extents.py

4. Generate the MUSIC2 configuration files¶

Generate all halo/resolution configurations with:

python3 make_music2_zoom_config_v2.py

Review at least one configuration from every resolution level. In particular, confirm the output filename, region file, cosmology, transfer-function input, box length, random seed/noise settings, refinement levels, and baryon split.

5. Generate and post-process all ICs¶

From the IC directory, run the canonical suite driver:

cd ~/gadget4/ICs
bash run_music2_suite.sh

For every configuration, the driver:

runs MUSIC2;
writes the canonical HDF5 IC;
extends the GADGET header arrays from six to seven particle types;
creates the intentionally empty PartType6 datasets required by CosmicGrain; and
performs the per-file readiness check.

The empty dust group is deliberate: dust particles are created later by stellar feedback, not placed in the initial conditions.

6. Validate the complete suite¶

Run the independent suite validator before deleting MUSIC2 intermediates or starting production simulations:

python3 ~/gadget4/scripts/validate_music2_ic_suite.py \
    --ic-root ~/gadget4/ICs

For a long unattended run:

mkdir -p ~/gadget4/ICs/MUSIC2_logs
nohup python3 ~/gadget4/scripts/validate_music2_ic_suite.py \
    --ic-root ~/gadget4/ICs \
    > ~/gadget4/ICs/MUSIC2_logs/validate_ic_suite.log 2>&1 &

The detailed checklist, result interpretation, and September 2026 suite result are recorded in IC-suite validation.

Do not remove wnoise_*.bin files until the validator reports:

Files: 48 PASS, 0 WARN, 0 FAIL (48 total)
FINAL STATUS: PASS

After a clean result, first review the exact deletion set:

find ~/gadget4/ICs -type f -name 'wnoise_*.bin' -print

Then remove only those MUSIC2 intermediates:

find ~/gadget4/ICs -type f -name 'wnoise_*.bin' -delete

Keep the MUSIC2 configurations, selected-halo tables, Lagrangian-region inputs, logs, validation CSV, and final HDF5 ICs as the reproducibility record.

7. Run a GADGET startup test¶

IC validation establishes structural and numerical consistency; it does not prove that a production configuration has enough memory or that the final zoom remains contamination-free at \(z=0\). Before committing to a long run:

start each intended IC with the actual CosmicGrain executable and parameter file;
confirm that all seven particle types are accepted;
confirm domain decomposition and the first force/hydrodynamic steps;
inspect memory use and particle balance; and
after evolution, measure low-resolution contamination within the target halo and its analysis aperture.

The \(4096^3\) IC particle counts vary greatly with Lagrangian-region geometry. Halo 1534 contains 517,537,024 initial particles, so validity alone does not make every highest-resolution run equally practical.

Historical note: Halo 569¶

Halo 569 was the original single-halo development target. Its earlier three-script workflow established the direct particle-ID trace-back approach, but its hardcoded filenames and one-halo selection cuts are not the maintained interface for the new suite. Use the versioned scripts listed above for new targets.