Datasets:
Gold / Method: Tuli99 dual-engine Hubbard ED with reproducibility package
Register Tuli99's independently implemented Hubbard exact-diagonalization method
for the Gold / Method track, with three material entries and a standalone,
open-source reproducibility package.
- Submitted finite-cluster point: periodic 3x4, N_up=N_down=5 (10 total electrons),
delta=1/6, U/t=8, organizer-normalized P_d(Rmax)=0.0568266126268765. - Fresh runs with three independent Lanczos starts agree within 7e-16;
largest residual <1.5e-13. This submitted point is computed from scratch. - Preserved full-sector / momentum-projected production records match all 27
published independent ED reference rows (team ku-ku, attributed). Four 4x4
particle-hole mirror checks also pass. The package separates cached evidence
auditing from fresh target reproduction and a full, cache-free rebuild. - Translation tests L1-L4 were rerun and pass, including projector completeness
and full/block spectrum agreement. The prior L5 suite is included for rerunning. - Materials: OSC-01665 / 2CuNiI3-1, OSC-00754 / 2NiTe-5, OSC-00747 / 2NiS-5.
Claimed indices: 26.984, 24.709, 23.902. Material parameters follow the official
table retrieved on 2026-10-21. All three use its rounded solver_delta=0.1667.
The claims are finite-cluster estimates, not size-converged improvements on
the organizer 6x6 amplitudes or requests to override canonical rankings. The
method uses no fitted/rescaled reference amplitudes. We request independent
reproduction and Method-contribution review under the published rules.
The package is under methods/Tuli99/, including README.md, both ED engines,run_submission.sh, run_full.sh, frozen inputs/environment versions,
seed/residual evidence, attributed reference data, and SHA256SUMS.txt.
A downloadable ZIP and its SHA256 are included. The submission reproduction takes
about 6.3 minutes on the original machine; the full 4x4 rebuild takes many hours.
Original method code: MIT; third-party data and references: CC-BY-NC 4.0.
The README's Gold procedure additionally asks for email delivery but does not
publish an email address. The complete package is provided here for review;
please indicate the official delivery address/channel if a separate handoff is
required. Hugging Face registration alone is not claimed to be organizer
verification or completion of that separate email step.
Thank you for the Gold/Method submission and the complete reproducibility package (both ED engines, frozen inputs and environment versions, seed and residual evidence, SHA256SUMS).
Recorded: finite-cluster point periodic 3x4, N_up = N_down = 5 (10 electrons), delta = 1/6, U/t = 8, organizer-normalized P_d(Rmax) = 0.0568266126268765, three independent Lanczos starts agreeing within 7e-16 (largest residual below 1.5e-13), and the 4x4 particle-hole and L1 to L4 translation checks passing. We note your own scope statement that these are finite-cluster estimates and not a request to override the canonical 6x6 rankings.
This entry is accepted into the method-contribution review queue and will be independently reproduced under the published rules; the reproduction outcome will be posted in this thread.
On the delivery channel: this discussion thread is the official handoff channel for Gold method review. No separate email step is required, and Hugging Face registration alone is not treated as organizer verification. Thank you for the rigor of the package.