Kunal Bhatia — Independent researcher, Meerut, India · ORCID 0009-0007-4447-6325
A reproducible, pre-registered gravitational-wave methodology project. Analysis code, locked predictions, per-event results, figures, and the manuscript for a single result, with every number in the paper generated from a committed artifact rather than typed by hand.
License: MIT (code and docs). Gravitational-wave data are not redistributed here; they are public GWOSC releases under CC-BY. The manuscript is provided for transparency and remains the author's work. See
LICENSE.
The component-mass posterior of a compact-binary coalescence is elongated along a contour of near-constant chirp mass — a shape usually treated as a nuisance degeneracy. We show that its orientation is quantitatively reconstructible from a single one-dimensional marginal of the same posterior.
Evaluating the constant-chirp-mass curve over an event's own mass-ratio marginal predicts the principal-axis position angle to a median 1.26° on GWTC-4.0/O4a and 1.22° on GWTC-5.0/O4b for elongated posteriors (axis ratio ≥ 3), with no coefficient calibrated on either catalog and with the prediction fixed on GWTC-3 before the later data were examined. The local tangent approximation used by rapid parameter-estimation tools achieves 3.9–6.7° on the same events.
The reconstruction requires the event's own marginal (substituting another event's degrades it several- fold; the achieved error lies below the minimum of 300 catalog-stratified permutations), and it transfers between separately sampled waveform families, so it is not an artifact of shared Monte Carlo noise. The predicted orientation is exactly invariant to chirp mass — rescaling the chirp mass is a dilation, which leaves covariance eigenvectors unchanged — so the reconstruction has one input, the mass-ratio marginal, not two.
It is presented as a measurement of posterior geometry and a systematics diagnostic, not as a test of general relativity: the posteriors are generated with general-relativistic waveform models, so their internal geometry cannot bound departures from the theory that produced them.
- Manuscript:
paper/manuscript.pdf— 9 pp, two-column (Physical Review D format,revtex4-2). - Reader's guide:
docs/EXTERNAL_READER_PACKET.md— the honest summary, strongest and weakest claims, and a reproduction checklist.
Raw parameter-estimation data are not included in this repository.
data/is gitignored (~68 GB of LVK PE releases). Every source is pinned with record numbers and DOIs indocs/DATA_AVAILABILITY.md, with download helpers inscripts/.results/e94_posterior_cache.npz(~572 MB, gitignored) is a one-time extract of the released posteriors. It stores every usable sample (no subsampling), so a cache-backed number is a full- sample number. Rebuild it withpython3 src/e94_build_posterior_cache.py(~104 s on the author's machine; one HDF5 pass, I/O-bound and machine-dependent). It is the single provenance source for the downstream batteries, which perform no HDF5 access of their own.
# regenerate every number and caption in the paper from the committed artifacts
python3 src/build_paper_numbers.py # -> paper/numbers.tex
python3 src/build_manuscript_figures.py # -> paper/fig_captions.tex
python3 src/build_doc_numbers.py # -> generated blocks in the docs
# build the manuscript
cd paper && pdflatex manuscript.tex && pdflatex manuscript.tex
# contract tests are data-free and run anywhere (184 tests)
python3 -m pytest tests/ -qAn empty git diff after the three build scripts means the paper matches its committed artifacts exactly.
Every result number in the manuscript is a LaTeX macro emitted from a committed JSON; none is typed by
hand, and a test fails if the paper and the artifacts disagree.
This repository contains an internal submission-gate audit of its own headline result. What is and is not established:
| item | status |
|---|---|
| curved-law reconstruction | reproducible from the committed cache (src/e94, src/e95) |
| residual is a real systematic | ~1°, about 17× the Monte Carlo resolution — not sampling noise (src/e92) |
| finite posterior thickness | supported out-of-sample; arc-variation NOT established (src/e96) |
| self-consistency correction | in-sample only; does not clear the out-of-sample bar (src/e97) |
| precision law | NOT PASSED — exploratory; its mass-band split is post-hoc (src/e93) |
| geometric GR-exponent diagnostic | a naive 3.1σ offset, demoted to 1.5σ by two pre-committed checks (Appendix A) |
| E85 Bayesian ringdown | RETRACTED — its posterior was prior-dominated |
Two provenance caveats, stated plainly. The O4b preregistration is timestamped in this repository's public history before the data were opened; the O4a preregistration is not independently timestamped — its prereg and results entered in the same commit, so its out-of-sample status rests on a private history. And O4a/O4b are disjoint event catalogs, not independent experiments: they share detectors, calibration, waveform families and priors.
GW/
├── paper/ manuscript (revtex4-2), generated numbers.tex and fig_captions.tex
├── src/ analysis code and the build_*.py generators
├── results/ numbers of record (JSON); the cache manifest (cache itself gitignored)
├── figures/ figures and their machine-readable sidecar JSONs
├── preregs/ locked pre-registrations
├── reports/ per-battery lab-notebook reports
├── tests/ data-free contract tests
├── scripts/ data fetchers
├── docs/ reference and working documentation — see docs/README.md
└── data/ parameter-estimation files (gitignored, not redistributed)
docs/README.md indexes the reference documentation — data availability,
referee-readiness summary, citation verification, literature, workflow, testing. The project's internal
working record (planning notes, review rounds, dated lab notes) is kept out of the public tree; the
manuscript and the committed artifacts under results/ are the authoritative record.
If you use this software or the analysis it reproduces, please cite both the manuscript and the software
record; see CITATION.cff. Archival metadata for Zenodo is in .zenodo.json.
This research has made use of data or software obtained from the Gravitational Wave Open Science Center (gwosc.org), a service of the LIGO Scientific Collaboration, the Virgo Collaboration, and KAGRA. This is independent work; the author is not a member of the LIGO–Virgo–KAGRA Collaboration, and the Collaboration has not reviewed this analysis and bears no responsibility for its conclusions.