Gold / Method: rsturg exact 6x6 two-carrier ground-state reference

#62
by rsturg - opened

Entrant and contact: Riley Sturgill (rsturg).

Hello OSC organizers,

I am registering a reproducible finite-cluster ground-state reference for Method-contribution review. The primary material example is 1FeH2O2-1 / OSC-00931, using the frozen published t = 0.18777 eV, U = 1.50216 eV, N(E_F) = 0.60844 and effective solver_delta = 0.9444. This submission changes no DFT band selection or downfolding input and does not request an automatic canonical-score override.

For the uniform, flux-free, periodic 6×6 square Hubbard model at U/t = 8 with one up and one down electron, the exact ground-state nearest-neighbor d-wave pair correlation is zero at every separation, including R = (3,3). Perron–Frobenius gives a unique positive ground state, and rotation about each anchor site makes the four directional singlet-annihilation amplitudes equal. Their d-wave signs cancel exactly. The included proof states all assumptions.

Independent full-sector (1,296 states) and relative-coordinate (36 states) diagonalizations give E₀/t = −7.906146146983354. The full-vector residual is 1.13×10⁻¹⁴. An independent exact-rational secular-equation calculation brackets E₀/t between −7.906146146984 and −7.906146146982. Explicit occupation-bit fermion operators, U = 0, scale, s-wave and broken-C4 controls are included. Ground-state diagonalization uses no stochastic optimization and no fitted or rescaled reference amplitudes.

The public table's A_d = 0.00283 gives this material a proxy index of 0.51656556. Our zero is a true-ground-state reference under the stated model, not a claim that a positive expectation in the d-wave-biased VMC trial state is mathematically impossible. Please consider it as a calibration/validation contribution, rather than a claim to a higher raw score.

The package also proves the maximum-distance zero for two holes on the even bipartite lattice when the pair supports are disjoint, and checks the actual 70-electron sector with independent Fock signs. It does not claim zero at all two-hole separations: the onsite control is nonzero. The frozen table audit finds 86 rows sharing the effective dilute-sector amplitude, with the 67 high-density rows interpreted conditionally through the public particle-hole convention. The primary material above avoids relying on that extension.

Please confirm that the private 6×6 target has periodic, flux-free nearest-neighbor hopping and that its reported A_d is the corresponding raw equal-time d-wave correlator at R = (3,3), up to the documented normalization. If the canonical target instead fixes a particular trial-state expectation or a different observable, please keep this as a separately identified ground-state reference. Please also indicate whether this limited same-size reference qualifies for Method review under the current rules.

The instrument-schema submission JSON records the OSC-00931 method name and zero ground-state reference. The frozen public release is commit d86ad39. The reproducibility package contains original code, proof, frozen inputs, recorded results, reproduction instructions and SHA256 evidence. Public package: https://huggingface.co/datasets/rsturg/osc-two-carrier-reference . Download osc-method-submission.zip. Archive SHA256: d888f4c1d34982f6bb6678ad855f1960923224baa7233fd7a33983981ca8e9b6. Earlier work by Ku-ku and Tuli99 is credited; I do not claim global novelty or first-ever Hubbard ED. No organizer verification, acceptance or award is claimed.

AI assistance disclosure: OpenAI Codex (AI) assisted with prize research, the mathematical derivation, implementation and local computational verification. The named human entrant is the submitter and contact person.

Thank you,

Riley Sturgill

Hugging Face: rsturg

FINAL_Bench org

Accepted as a Gold/Method calibration reference, @rsturg , with no canonical-score override (none requested). Your exact result is the key point: the d-wave pair correlation of the true 6x6 two-carrier ground state is zero, so the table's A_d is a d-wave-biased trial-state (VMC) estimator, not a ground-state value. We log E0/t = -7.906146146983 and the disjoint-support two-hole zero as a validation contribution. This also informs the delta=0 calibration raised in #56.

Sign up or log in to comment