Numerical evidence for even-odd width effects in 2D repulsive Hubbard model (cylinder) at half-filling: Spin correlations and Haldane conjecture connection?

I’m investigating the spin-spin correlation behavior of the repulsive 2D Hubbard model at half-filling under a narrow cylinder geometry (periodic boundary in circumferential direction, open in axial direction). Theoretical studies (e.g.cond-mat/0103159) suggest a striking even-odd width dependence:

  • Even widths (4-leg, 6-leg): Exponential decay of spin correlation while the correlation length is expected to diverge as width increases. (I guess this can be a spin-1 chain analogy as the spin-1/2 fermion paired into integer spin along the circumferential direction and there is Haldane gap?).
  • Odd widths: Algebraic decay from gapless spinon modes (I guess this is a half-integer spin chain analogy?).

Key questions:

  1. Validation: Have DMRG studies explicitly observed this?
  2. Haldane connection: Do numerical results support the interpretation via Haldane’s conjecture? (e.g., extracted spin gaps vs. width parity)
  3. Width scaling: For wider cylinders (e.g., 6-leg+), does the correlation length diverge (algebraic recovery) as 2D fluctuations dominate?

My results:
cylinder geometry, strong U=10t, Lx=48:

  • Width=4 (even): Clear exponential decay.
  • Width=3 (odd):
    • Ambiguous Power-law: Attempted r^{-\alpha} fits fail at long distances (e.g., beyond r>15).
    • Technical details: Bond dimension=20000, cutoff=1e-9, reference points averaged over ix=Nx/4-2:Nx/4+2 and circumferential positions (y). I think the result is free from bond dimension and boundary effect.

It seems there’s no DMRG study on the odd-even effect for cylinder geometry. As for ladder, check this review.
Personally speaking, the cylinder geometry introduces 1 additional frustration along the circumference and also lead to a “spin-liquid” (strictly speaking, spin-liquid is valid in thermodynamic limit). Besides, the MPS with finite bond will pick only part of the configuration so that the spin structure factor may show several peaks with different magnitude.
While, all these should be finite-size gap which will approach 0 with increasing width.