Guarded Low-Rank Woodbury Covariance Updates for Reproducible Cosmological Likelihood Evaluation
This manuscript audits Gaussian likelihood evaluation using dense and guarded low-rank Woodbury covariance updates. It reports numerical agreement, timing and process peak memory for specified CPU workloads on Apple M1 Pro. Independent CPU reproduction remains pending. The Planck TT appendix reports aggregate results for 19 archived fixed inputs. The standalone MIT-licensed NumPy reference is available at https://github.com/fujikix1102/nova-woodbury/tree/v0.0.2.dev0. CC BY 4.0 applies to original material for which the author holds licensing rights. The prescribed DESI acknowledgment is excluded from that grant. DESI DR1 inputs used in Figures 4–9 retain their attribution and terms.
Authors
- JUNICHI FUJIKI (ORCID: https://orcid.org/0009-0004-6729-678X)
Publication Details
- Journal
- Zenodo (CERN European Organization for Nuclear Research)
- Published
- 2026-11-13
- DOI
- https://doi.org/10.5281/zenodo.24181678
- Primary Topic
- Cosmology and Gravitation Theories
- Type
- preprint