highz-accretion-atlas

A standardized, assumption-tracked catalogue of JWST-identified high-redshift ($z \ge 4$) accreting massive-black-hole systems and candidates, and their possible formation and growth scenarios.

Manuscript: read Early Black-Hole Growth Constraints from JWST.

Identity audit status: all admitted redshifts and available coordinates pass independent source checks, but one conflicting duplicate group and two close-neighbour groups need reconciliation. The stored 338-object count is provisional; see the identity audit.

Background

The James Webb Space Telescope pushes observational cosmology into the first few billion years of cosmic history, and it has revealed massive accreting objects that are hard to explain due to the limited time for them to grow. It follows that one of the biggest questions cosmologists are asking right now is: “How did these objects get so big, so fast?” My research project aims to contribute to this question by creating a standardized cross-paper catalogue of these objects and then testing what scenarios could have theoretically formed each one. Either they started from a sufficiently massive seed, accreted continuously for a long enough time, formed very early, or some combination of these conditions. This is explored and visualized with parameter-space maps and growth tracks based on the analytic black hole growth equation used in Dayal (2024).

Across the literature, key quantities are often inferred with different methods and assumptions, even though small shifts in inferred quantities like M_BH, M_*, or L_bol can imply very different growth histories. A standardized, assumption-tracked catalogue would clarify which objects are genuinely challenging for standard formation models rather than artifacts of inference choices.

Overall, this project aims to help determine what objects are the best candidates for follow-up and deeper theory work.

The strongest reference-model tensions occur in the early subset, not across the whole catalogue. Results constrain combinations of seed, efficiency and accretion assumptions; they do not reconstruct unique histories or prove that super-Eddington episodes were necessary. See the scientific limits and interpretation.

Theory

The growth model follows Eq. 1 of Dayal (2024):

$M_{BH}(t) = M_{seed}(t_{seed})e^{\frac{4\pi Gm_p f_{Edd}}{c\sigma_T}\frac{1-\epsilon}{\epsilon}(t-t_{seed})}$

The optional merger-assisted case multiplies this smooth-accretion result by $B_{\rm merge}$. Thus $B_{\rm merge}=2$ adds a fixed $\log_{10}(2)=0.301$ dex to the predicted mass; it does not double $f_{\rm Edd}$ or the exponential growth rate. For ideal Kerr thin disks, the spin cases $a=-1,0,+1$ correspond to $\epsilon=0.038,0.057,0.423$ (rounded), respectively.

The spin-separated $f_{Edd}$-mass maps additionally show an illustrative slim-disk coupling above Eddington. They retain the spin-dependent thin-disk efficiency for $f_{Edd}\leq1$ and use $\epsilon_{\rm eff}=\epsilon_{\rm spin}f_{Edd}/e^{f_{Edd}-1}$ for $f_{Edd}>1$, representing the decline in effective radiative efficiency from photon trapping. The growth-track figure continues to show constant-efficiency reference curves.

Cosmic time in Gyr, derived from the FLRW form with flat $\Lambda$ CDM:

$t(z) = \frac{2}{3H_0\sqrt{\Omega_{\Lambda}}}{\sinh^{-1}}(\frac{\sqrt{\Omega_{\Lambda}/{\Omega_m}}}{(1+z)^{3/2}})$

All catalogue ages, growth intervals, tables, and figures use a flat Planck 2018-style cosmology with $H_0=67.3,{\rm km,s^{-1},Mpc^{-1}}$, $\Omega_m=0.315$, and $\Omega_\Lambda=0.685$.

For each object, the pipeline computes the cosmic time available between $z_{seed}$ and the observed redshift. It then explores a grid of:

• ${M_{seed}}$
• $f_{Edd}$
• $z_{seed}$
• radiative efficiency $\epsilon$

Manuscript and Paper Products

The complete first manuscript draft is available as paper/highz_accretion_atlas_v3.pdf, with its editable LaTeX source beside it. It uses the frozen 338-object v3 catalogue with conservative 220/234-object inference samples. The revised draft adds source-aware target assessments and explicitly separate literature/direct-mass comparisons; see paper/analysis/.

Main-text figures use the conservative manuscript samples:

• sample overview — paper/figures/landscape.png
• growth tracks across four fixed efficiencies — paper/figures/growth_tracks.png
• reported-error uncertainty — paper/figures/uncertainty.png
• seed/efficiency growth boundaries — paper/figures/growth_boundaries.png
• alternate-measurement sensitivity — paper/figures/measurement_sensitivity.png
• scenario compatibility summary — paper/figures/compatibility.png

The detailed efficiency prescription remains in the appendix. Generated tables, source-aware target assessments, and sensitivity comparisons are in paper/analysis/; their LaTeX fragments are included directly in the draft. See manuscript analysis products for their scope and reproduction commands. Full-catalogue plots and navigation rankings remain available under results/v3/ as supporting atlas products.

Supporting products:

• full catalogue schema — docs/reference/admission-schema.md
• full source registry — data/source_family_registry.csv, data/source_provenance_registry.csv, and data/mass_method_registry.csv
• full processed catalogue tables — data/processed/v3/ with identity products in data/crossmatch/v3/
• full result tables — results/v3/tables/
• full per-object $f_{Edd}$-mass map gallery — results/v3/parameter_maps/fedd_mass_maps/
• full seed-redshift-mass map gallery — results/v3/parameter_maps/seedredshift_mass_maps/
• comprehensive v3 growth-track grid preserving all historical v1 line combinations — results/v3/figures/v3_all_object_growth_tracks_full_assumptions.png
• validation checks — results/v3/tables/v3_exclusion_audit.csv, results/v3/tables/v3_all_object_visual_coverage.csv, and results/v3/tables/v3_science_policy.csv; executable gate in scripts/04_verify.ipynb
• sensitivity tests — results/v3/tables/v3_alternate_measurement_sensitivity.csv and results/v3/figures/v3_measurement_sensitivity.png
• source-by-source caveats — results/v3/tables/v3_source_caveat_summary.csv
• source-level selection/completeness audit — data/selection_function_registry.csv and results/v3/tables/v3_selection_completeness_summary.csv
• immutable extraction audit — data/manual_extraction_audit.csv

The canonical v3 gallery covers all 338 objects with one $f_{Edd}$-mass map in results/v3/parameter_maps/fedd_mass_maps/ and one seed-redshift-mass map in results/v3/parameter_maps/seedredshift_mass_maps/. The 237 growth-eligible objects receive numerical panels; the other 101 receive explicit no-inference status panels. Growth tracks are retained only as combined catalogue-wide figures under results/v3/figures/; the full-assumption v3 companion contains 72 curves spanning three seed masses, three $f_{Edd}$ values, four efficiencies, and two merger boosts. Seed mass is encoded by color, $f_{Edd}$ by line style, efficiency by line width, and merger boost by opacity. Catalogue-wide compatibility and Monte Carlo atlases retain every object label, while class-specific summaries avoid reporting a heterogeneous pooled demographic fraction. Start with results/README.md or the version manifests under releases/.

The main text should showcase the atlas logic and strongest rankings, while the appendix preserves the comprehensive technical and visual record. This keeps the project centered on observational triage rather than on claiming that any single seed or accretion channel is proven.

Workflow

Dataset versions describe nested scientific datasets, not software releases or chronological development checkpoints. Every version uses the same latest applicable corrections, identity rules, cosmology, growth model, uncertainty propagation, comparison policy, and visual grammar.

Version	Dataset	Measurements	Objects	Hosts
v1	Original Juodzbalis et al. JADES BLAGN catalogue	23	23	23
v2	v1 plus comparable JWST BLAGN sources with canonical masses	218	211	210
v3	v2 plus heterogeneous JWST-identified candidates	350	338	337

For each version, canonical catalogues are under data/processed/<version>/, identity products are under data/crossmatch/<version>/, and science tables, figures, and per-object galleries are under results/<version>/. Source-specific raw files retain descriptive publication names because they are immutable extractions.

Run the numbered notebooks in scripts/ from top to bottom. They call tested Python modules under src/internal/; scientific implementation does not live only in notebook state. To execute the complete workflow non-interactively:

mkdir -p /tmp/highz-atlas-notebooks
for notebook in scripts/0[0-4]_*.ipynb; do
  .venv/bin/python -m nbconvert --to notebook --execute \
    --ExecutePreprocessor.timeout=1800 \
    --output-dir=/tmp/highz-atlas-notebooks "$notebook" || exit 1
done

Historical source-admission builders and the shared ranking/uncertainty core remain under src/internal/compatibility/; their names do not define public dataset versions or write legacy output trees.

Getting Started

The project requires Python 3.12. Create a repository-local virtual environment and install the pinned project requirements (including the explicit build backend). The notebook lock covers the full dependency closure on Linux and macOS:

python3.12 -m venv .venv
.venv/bin/python -m pip install --requirement requirements-notebook-lock.txt --requirement requirements-build-lock.txt
.venv/bin/python -m pip check

Run the complete regression and verification suite:

PYTHONDONTWRITEBYTECODE=1 .venv/bin/python -m unittest discover -s tests
.venv/bin/python -m src.internal.verify_manual_extractions
.venv/bin/python -m src.internal.verify_primary_source_values
.venv/bin/python -m src.internal.verify_source_provenance
.venv/bin/python -m src.internal.verify_versions

Full dataset generation commands are listed in the workflow above. The source review cutoff and explicit admission boundary are documented in docs/reference/literature-scope.md; versioning details are in docs/guides/versioning.md.

Repository map

Folder-level guides keep the data, results, documentation, releases, and code easy to navigate:

• data/: raw sources, processed catalogues, and identity products
• results/: science tables, figures, galleries, and inventory
• docs/: contracts, methods, guides, and source notes
• releases/: exact dataset manifests and hashes
• src/, scripts/, and tests/: implementation, commands, and validation

References

Catalogue data sources are documented authoritatively in data/sources.md. The following is background and prospective reading, not a list of sources currently represented by catalogue rows:

1. Dayal, P. 2024, A&A, 690, A182
2. Ji, X., Maiolino, R., Übler, H., et al. 2025, MNRAS, 544, 3900
3. Maiolino, R., Übler, H., D’Eugenio, F., et al. 2025, arXiv:2505.22567
4. Dayal, P. & Maiolino, R. 2025, arXiv:2506.08116
5. Prole, L. R., Regan, J. A., Mehta, D., et al. 2025, arXiv:2506.11233
6. Adamo, A., Atek, Hakim., Bagley, M., et al. 2025, arXiv:2405.21054
7. Dayal, P. & Ferrara, A. 2018, arXiv:1809.09136
8. Stark, D., Topping, M., Endsley, R., et al. 2025, arXiv:2501.17078

Reproduction compares regenerated CSV values and PNG pixels with an independent baseline before refreshing hashes; see reproduction and intentional updates. Independent source fixtures cover all 32 families with 2,041 field checks; all 244 numerical masses and both error bounds are independently checked. The separate redshift/identity fixture checks all central redshifts and available coordinates but reports three unresolved identity groups. Other observable fields retain representative coverage. See the validation scope and extension requirements and scientific limits; passing CI does not imply a complete source audit or permit population-demographic claims.

The manuscript additionally excludes all six records linked to the three open identity groups and retains four tentative JADES detections only in the exploratory sample, giving 220 primary and 234 exploratory numerical objects. The catalogue retains every source measurement and its provisional identity. Notebook 01 reproduces the manuscript selection and sensitivity tables; notebook 04 verifies them with src.internal.publication_selection. This conservative analysis check does not replace the strict identity-resolution gate. See manuscript scope and reproduction.