arXiv:2610.30321
Extreme competition between stripes and superconductivity in the hole-doped Hubbard model
Ao Chen, Johnnie Gray, Si-Jing Du, Garnet Kin-Lic Chan
Abstract
The two-dimensional Hubbard model was originally proposed as the minimal model of uniform $d$-wave superconductivity in the cuprate high-temperature superconductors, but recent calculations with relevant parameters in a realistic doping regime have often found other ground states. Further, uncertainty remains over such results due to the intrinsic bias of different computational methods. Here we address this uncertainty through state-of-the-art neural tensor network and neural backflow state simulations, enabled by large-scale GPU computation, that together reach an unprecedented variational accuracy on lattices with up to 1024 sites. At $1/8$ doping in the pure Hubbard model, as well as with next-nearest-neighbor hopping relevant to the hole-doped cuprates, we reveal a near degeneracy among a low-energy uniform $d$-wave state, superconducting stripes, and non-superconducting stripe states on the milli-$t$ energy scale. Our work rebalances the narrative that the Hubbard model supports a single dominant order in relevant parameter regimes towards a picture of extreme competition between stripes and superconductivity.
AI Summary
The paper asks whether stripes or uniform $d$-wave superconductivity dominate the square-lattice hole-doped Hubbard model, and whether conflicting numerical results reflect variational biases. It studies $U/t=8$, hole doping $1/8$, and $t'/t=0,-0.2$ using BCS and Pfaffian neural-backflow states and neural-network fermionic PEPS, with periodic lattices up to $32\times32$ sites. Energies and spin, charge, and pairing correlations are compared across ansatzes; order parameters are extrapolated to the thermodynamic limit. At $t'=0$, neural-fPEPS yields non-superconducting filled stripes, while pairing-based backflow states yield a competing uniform superconductor with extrapolated pairing amplitude $\Delta_{\mathrm{SC}}=0.0382(8)$. On $32\times32$, the stripe has lower variational energy by $0.0006t$ per site. On $16\times16$, all three energies lie within $0.002t$ per site of an independently extrapolated fPEPS estimate, $E/N=-0.7602(14)t$. At $t'=-0.2t$, the lowest-energy Pfaffian-backflow states exhibit half-filled stripes. Their extrapolated pairing is fit-dependent: consistent with zero using three sizes, or $\Delta_{\mathrm{SC}}=0.010(3)$ using the two largest. A competing nearly uniform BCS-backflow state retains $\Delta_{\mathrm{SC}}=0.019(4)$. These results place distinct stripe and superconducting variational states within milli-$t$ energy scales per site. Computational ingredients include configuration-dependent fermionic tensors, GPU-optimized contraction, symmetry projection, and adaptive stochastic reconfiguration. The paper emphasizes persistent ansatz biases and the need for cross-validation. Stated uncertainties include strong finite-size effects in the nearly uniform state's stripe extrapolation at negative $t'$, fit-sensitive weak pairing, and condensation-energy scales below $10^{-3}t$; the competing orders are not definitively resolved.
These kernels were prepared with the assistance of Claude Code, as acknowledged in the main text.
Central claim may not hold: The t′=0 results demonstrate two competing variational orders, not three distinct thermodynamic phases including a superconducting stripe. At t′=−0.2, superconductivity in the lowest-energy stripe remains extrapolation-dependent, and no energy comparison establishes that the robust superconducting stripe from prior work is a third milli-t competitor. The sub-milli-t differences between optimized variational energies are not controlled physical phase splittings, so they do not establish the asserted thermodynamic ground-state ordering. The claimed few-milli-t absolute accuracy on the largest systems is not independently established by the uncertain 16×16 reference or by comparable benchmarks at the other sizes and hopping parameter.
Written by gpt-6.1-sol from the paper's full text. Automated: it can be wrong.
- Comments
- 33 pages, 25 figures