betaFeedback

WorksPhysics and Astronomy

Multi-reference variational states resolve stripe selection and local dd-wave pairing in the Hubbard model

Abstract

Whether stripes compete with dd-wave superconductivity or promote pairing through fluctuations remains unsettled in the two-dimensional Hubbard model. Using multi-reference neural variational auxiliary-field quantum Monte Carlo (MR-nVAFQMC) at U/t=8U/t=8, we superpose unpaired Slater determinants across stripe periods and domain-wall positions, reaching variational energies competitive with leading benchmarks on periodic lattices up to 32×3232\times32 sites. For cuprate next-nearest-neighbor hopping (t′/t=−0.2t'/t=-0.2), diagonal hopping preserves electron-doped antiferromagnetism while kinetically frustrating hole motion into Yamada stripes with a dome of short-range connected dd-wave pairing. At 1/81/8 hole doping, hole delocalization across antiphase spin boundaries outweighs on-site repulsion, splitting the filled stripes of the pure model into half-filled walls with strong local pairing. Coherently delocalizing shifted walls strongly suppresses static charge modulation yet leaves short-range pairing unchanged, showing that local pairing is intrinsic to half-filled walls rather than driven by stripe fluctuations. Across intervening antiferromagnetic domains, however, pairing correlations become unresolved at one stripe period and beyond. Both parameter transfer to larger lattices without reoptimization and deep benchmark checkpoints retain stripe spin order without resolving long-range pairing, within uncertainties allowing weak superconductivity. Half-filled antiphase walls thus generate local dd-wave pairs, whereas suppressing static stripe order alone yields no resolved coherence across stripes in the states studied.

Introduction

How unidirectional spin and charge stripes interact with dd-wave superconductivity has been a central question in the physics of high-TcT_c cuprates for more than three decades [1, 2, 3]. In La2−x_{2-x}Srx_xCuO4_4, neutron scattering reveals incommensurate magnetic peaks displaced from (π,π)(\pi,\pi) by ϵspin\epsilon_{\rm spin}, which grows approximately linearly with hole concentration (ϵspin≈δ\epsilon_{\rm spin} \approx\delta, the Yamada relation) up to δ≈1/8\delta\approx1/8 before saturating at higher doping [4]. The suppression of bulk superconductivity near 1/81/8 doping in La2−x_{2-x}Bax_xCuO4_4 is the longstanding 1/81/8 anomaly [5]. In La1.875_{1.875}Ba0.125_{0.125}CuO4_4, static period-four charge order (λc=4\lambda_c = 4) and period-eight antiphase spin order (λs=8\lambda_s = 8) coexist with a strongly depressed bulk three-dimensional (3D) transition at Tc≈4T_c \approx4 K, whereas transport and magnetization measurements indicate two-dimensional (2D) superconducting fluctuations below about 40 K and a Berezinskii–Kosterlitz–Thouless transition near 16 K [6]. A proposed explanation for the suppressed interlayer coherence is an intra-plane pair-density wave (PDW), whose pairing phase alternates across adjacent stripes and whose stripe direction rotates by 90° between layers, frustrating interlayer Josephson coupling [2, 7]. Because unidirectional antiphase stripes of charge period λc\lambda_c and linear hole filling f=δλcf = \delta\lambda_c carry magnetic incommensurability ϵspin=δ/(2f)\epsilon_{\rm spin} = \delta/(2f), the observed slope ϵspin=δ\epsilon_{\rm spin} = \delta points directly to half-filled domain walls (f=1/2f = 1/2) rather than full walls (f=1f = 1, ϵspin=δ/2\epsilon_{\rm spin} = \delta/2). Whether a single-layer 2D Hubbard model at δ=1/8\delta= 1/8 develops phase coherence across half-filled stripes—as uniform dd-wave or antiphase PDW order—or supports only short-range intra-wall pairing remains a central microscopic question [2, 8, 9, 10].

Whether the single-band 2D repulsive Hubbard model captures these intertwined phenomena has been investigated intensively with modern many-body methods [11, 12, 13, 14, 15, 16]. For the nearest-neighbor model (t′=0t' = 0) at δ=1/8\delta= 1/8 and U/t=8U/t = 8, combined density-matrix renormalization group (DMRG), auxiliary-field quantum Monte Carlo (AFQMC), density-matrix embedding, and tensor-network calculations established an insulating filled-stripe ground state (λc=8\lambda_c = 8, λs=16\lambda_s = 16) without long-range superconductivity [17, 18, 19]. Negative next-nearest-neighbor hopping t′/t<0t'/t < 0, relevant to cuprate Fermi surfaces and electron–hole asymmetry [20, 21, 22], favors shorter stripe wavelengths; at δ=1/8\delta= 1/8, infinite projected entangled-pair states find period-four stripes with suppressed superconductivity [23]. Pairing also depends sensitively on cylinder width: four-leg calculations find algebraically decaying pairing at finite t′t' [24], whereas eight-leg calculations find extremely short-ranged superconducting correlations for t′≤0t' \leq0 [25]. In a distinct setting, periodically modulating nearest-neighbor hopping enhances superconducting correlations on four- and six-leg cylinders [26]. Finite-temperature simulations reveal fluctuating stripes and intertwined spin, charge, and pair correlations [27, 28]. Complementary cold-atom experiments observe precursors of individual stripes in mixed-dimensional settings [29] and a pseudogap in a two-dimensional Fermi–Hubbard simulator [30].

Determining the 2D zero-field behavior at δ=1/8\delta= 1/8 with t′≠0t' \ne0 has nonetheless proved delicate because distinct uniform, superconducting-stripe, and non-superconducting-stripe states compete on the milli-tt (10−3t10^{-3}t) energy scale per site and respond sensitively to boundary conditions, applied fields, and variational bias. Combining DMRG with constrained-path AFQMC under twist-averaged boundary conditions and applied pairing fields—taking the thermodynamic limit at fixed pairing field before extrapolating the field to zero—Xu et al. reported dd-wave superconductivity coexisting with partially filled stripes (f≈0.6f \approx0.6–0.80.8) on the hole-doped side and Néel antiferromagnetism on the electron-doped side [31]. Working directly at zero external field on periodic 2D tori with neural quantum states [32, 33, 34], Rende et al. reported finite extrapolated dd-wave pairing up to 24×2424 \times24 using symmetry-preserving BCS backflow [35]. Extending Pfaffian backflow [36, 37] and neural-network fermionic projected entangled pair states (NN-fPEPS) up to 32×3232 \times32 tori, Chen et al. reached lower variational energies and revealed a near-degeneracy between a nearly uniform dd-wave state and lower-energy striped states carrying very weak or statistically unresolved long-range superconductivity [10]. On 8×88 \times8, Viteritti et al. found that near-degenerate determinant and Pfaffian backflow states can exhibit different pair correlations, but that symmetry restoration and variance reduction bring the three wavefunction families toward a common picture of coexisting stripe and pairing correlations on this finite lattice [38]. Related neural backflow calculations show how symmetry projection and multi-determinant improvements change both energies and correlation functions [39].

Two methodological considerations motivate probing this competing regime with multi-reference unpaired auxiliary-field wavefunctions. First, a single-reference variational state—even after space-group projection—inherits a single reference stripe period and rigid inter-wall spacing, preventing inequivalent stripe periods and shifted domain-wall arrangements from competing or fluctuating coherently within one quantum state. Second, variational families differ in how pairing enters their reference states. Canonical number-projected BCS and Pfaffian wavefunctions conserve total particle number just as Slater determinants do, but embed explicit pairing fields into their reference geminals; by contrast, each individual unprojected particle-conserving Slater determinant ∣Φk⟩\lvert\Phi_k\rangle carries no explicit pairing field and has vanishing connected two-body pairing (Cp=0C_p = 0), whereas space-group projection, multi-determinant superposition, and the shared imaginary-time auxiliary-field projector generate non-Wick four-fermion correlations without geminal pairing initialization. Although a variational upper bound on total energy implies neither upper nor lower bounds on order parameters, reaching competitive energies from particle-conserving Slater pools provides a complementary test of which features of stripe order govern local pair formation and which limit long-distance coherence.

Here we develop multi-reference neural variational auxiliary-field quantum Monte Carlo (MR-nVAFQMC) to study the 2D square-lattice tt–t′t'–UU Hubbard Hamiltonian (Figure 1A) on periodic Lx×LyL_x \times L_y tori (N=LxLyN = L_xL_y sites),

Figure 1. Square-lattice tt–t′t'–UU Hubbard model, Yamada relation, and real-space stripe correlations across doping. (A) Schematic of the 2D square tt–t′t'–UU Hubbard model (U/t=8U/t = 8; top) and competing filled (λc=8\lambda_c = 8, λs=16\lambda_s = 16) versus half-filled (λc=4\lambda_c = 4, λs=8\lambda_s = 8) antiphase stripe orders at δ=1/8\delta= 1/8 (bottom). (B) Spin incommensurability ϵspin=1/λs\epsilon_{\mathrm{spin}} = 1/\lambda_s (top) and nearest-neighbor connected dd-wave pairing vertex Cp(Δy=1)C_p(\Delta y = 1) (bottom) versus signed doping δ\delta (δ<0\delta< 0: t′/t=+0.2t'/t = +0.2; δ>0\delta> 0: t′/t=−0.2t'/t = -0.2) on 16×416 \times4 (circles), 12×1212 \times12 (squares), and 16×1616 \times16 at δ=1/8\delta= 1/8 (diamonds: blue, t′=0t' = 0; vermilion, t′/t=−0.2t'/t = -0.2, coinciding with ϵspin=1/8\epsilon_{\mathrm{spin}} = 1/8 in the top panel). Top: open symbols denote spin peaks below 3σ3\sigma; the grey band guides the Yamada trend (ϵspin=δ\epsilon_{\mathrm{spin}} = \delta up to δ=1/8\delta= 1/8 followed by an ϵspin=1/8\epsilon_{\mathrm{spin}} = 1/8 plateau across δ≤3/16\delta\le3/16), and the dashed blue line marks filled stripes (ϵspin=δ/2\epsilon_{\mathrm{spin}} = \delta/2). (C–E) Intrinsic 1D carrier-density profiles δ(x)\delta(x) [or ∣δ∣(x)|\delta|(x) in (C); top] and five-row 2D staggered spin-correlation maps (−1)x+yCs(r)(-1)^{x+y} C_s(\mathbf{r}) around a reference site (open circle; bottom) for (C) electron-doped 12×1212 \times12 at δ=−1/8\delta= -1/8 (t′/t=+0.2t'/t = +0.2, commensurate AFM), (D) 16×1616 \times16 at δ=1/8\delta= 1/8, t′=0t' = 0 (filled stripe, λc=8\lambda_c = 8, λs=16\lambda_s = 16), and (E) 16×1616 \times16 at δ=1/8\delta= 1/8, t′/t=−0.2t'/t = -0.2 (half-filled stripe, λc=4\lambda_c = 4, λs=8\lambda_s = 8). Error bars denote 1 s.e.

H^=−t∑⟨ij⟩,σ(c^iσ†c^jσ+H.c.)−t′∑⟨⟨ij⟩⟩,σ(c^iσ†c^jσ+H.c.)+U∑in^i↑n^i↓,(1)\hat{H} = -t \sum_{\langle ij\rangle,\sigma} \left(\hat{c}^{\dagger}_{i\sigma}\hat{c}_{j\sigma} + \mathrm{H.c.}\right) - t' \sum_{\langle\langle ij\rangle\rangle,\sigma} \left(\hat{c}^{\dagger}_{i\sigma}\hat{c}_{j\sigma} + \mathrm{H.c.}\right) + U \sum_i \hat{n}_{i\uparrow}\hat{n}_{i\downarrow}, \tag*{(1)}

where ⟨ij⟩\langle ij\rangle and ⟨⟨ij⟩⟩\langle\langle ij\rangle\rangle denote nearest- and next-nearest-neighbor pairs, n^iσ=c^iσ†c^iσ\hat{n}_{i\sigma} = \hat{c}^{\dagger}_{i\sigma}\hat{c}_{i\sigma}, and we set t=1t = 1 with U/t=8U/t = 8. Hole doping δ=1−Ne/N\delta= 1 - N_e/N is studied in the spin-balanced canonical sector N↑=N↓=Ne/2N_{\uparrow} = N_{\downarrow} = N_e/2 for both t′=0t' = 0 and cuprate hopping t′/t=−0.2t'/t = -0.2; physical electron doping (n=1+∣δ∣n = 1 + \lvert\delta\rvert at t′/t=−0.2t'/t = -0.2) maps via the bipartite particle–hole transformation c^iσ→(−1)xi+yic^iσ†\hat{c}_{i\sigma} \to(-1)^{x_i+y_i}\hat{c}^{\dagger}_{i\sigma} to filling 1−∣δ∣1 - \lvert\delta\rvert at reversed diagonal hopping t′/t=+0.2t'/t = +0.2, displayed at signed doping δ<0\delta< 0.

Results

Doping asymmetry, the Yamada relation, and kinetic stabilization of half-filled stripes

Figure 1B–E traces how magnetic order and local dd-wave pairing evolve across electron and hole doping in the absence of external pinning fields. On the electron-doped branch (δ<0\delta< 0, t′/t=+0.2t'/t = +0.2), doped carriers move coherently on the same magnetic sublattice without nucleating antiphase domain walls: the staggered spin correlation (−1)x+yCs(r)(-1)^{x+y}C_s(r) remains uniformly positive across the 12×1212 \times12 torus at δ=−1/8\delta= -1/8 (Figure 1C), keeping the magnetic Bragg peak locked at the commensurate Néel wavevector (π,π)(\pi,\pi) (ϵspin=0\epsilon_{\mathrm{spin}} = 0; Figure 1B, top), while no local connected dd-wave pairing dome develops (Figure 1B, bottom). On the hole-doped branch (δ>0\delta> 0, t′/t=−0.2t'/t = -0.2), by contrast, the magnetic peak splits into incommensurate pairs (π±2πϵspin,π)(\pi\pm2\pi\epsilon_{\mathrm{spin}},\pi) whose shift follows the half-filled Yamada relation ϵspin=δ\epsilon_{\mathrm{spin}} = \delta across commensurate lattices at δ=1/16\delta= 1/16 (λs=16\lambda_s = 16), δ=1/12\delta= 1/12 (λs=12\lambda_s = 12), and δ=1/8\delta= 1/8 (λs=8\lambda_s = 8). Above δ=1/8\delta= 1/8 on 16×416 \times4, the magnetic wavevector saturates at the ϵspin=1/8\epsilon_{\mathrm{spin}} = 1/8 plateau at δ=3/16\delta= 3/16—raising the linear wall filling from f=1/2f = 1/2 to f=3/4f = 3/4 while the stripe spin peak weakens nearly threefold—before departing from the ϵspin=1/8\epsilon_{\mathrm{spin}} = 1/8 plateau to a weaker

ϵspin=1/16\epsilon_{\mathrm{spin}} = 1/16 peak at heavy overdoping (δ=1/4\delta= 1/4), where static charge modulation falls below 3σ3\sigma. Across the entire hole-doped range δ∈[1/16,1/4]\delta\in[1/16,1/4], the nearest-neighbor connected dd-wave pairing vertex Cp(Δy=1)C_p(\Delta y = 1) (Section III) forms a broad positive dome, reaching its maximum around δ=3/16\delta= 3/16 where period-eight spin stripe order begins to soften (Figure 1B, bottom).

To identify how negative t′/tt'/t selects half-filled Yamada stripes over the filled stripes of the pure Hubbard model, we compare candidate charge periods λc∈{2,4,6,8,∞}\lambda_c \in\{2,4,6,8,\infty\} at δ=1/8\delta= 1/8 in Figure 2. On the four-leg 16×416 \times4 torus, whose length Lx=16L_x = 16 accommodates both λs=16\lambda_s = 16 and λs=8\lambda_s = 8 at deep projection (τ=1.6\tau= 1.6), the filled λc=8\lambda_c = 8 stripe (f=1f = 1) is favored at t′=0t' = 0 whereas introducing t′/t=−0.2t'/t = -0.2 reverses the energetic ordering (Figure 2A): the half-filled λc=4\lambda_c = 4 manifold (f=1/2f = 1/2) becomes the clear variational minimum among all single-period sectors, lying (41.0±1.1) mt(41.0 \pm1.1)\,mt per site below λc=8\lambda_c = 8 and well below λc=2\lambda_c = 2, λc=6\lambda_c = 6, and the uniform (λc=∞\lambda_c = \infty) state. When all candidate periods compete simultaneously inside unbiased K=8K = 8 and K=12K = 12 multi-period pools (Figure 2A, shaded column), the optimized wavefunctions spontaneously select the λs=8\lambda_s = 8 (λc=4\lambda_c = 4) sector and match or slightly improve upon the independently optimized λc=4\lambda_c = 4 energy.

Figure 2. Stripe wavelength switch and kinetic stabilization at δ=1/8\delta= 1/8. (A) Ground-state energy per site E/NE/N on 16×416 \times4 across charge periods λc∈{2,4,6,8,∞}\lambda_c \in\{2,4,6,8,\infty\} for t′/t=0t'/t = 0 (blue circles) and t′/t=−0.2t'/t = -0.2 (vermilion squares, τ=1.6\tau= 1.6), alongside unbiased K=8K = 8 RR-CV and K=12K = 12 multi-period pools (shaded column). (B) Longitudinal spin structure factor Sz(qx,π)S_z(q_x,\pi) on 16×1616 \times16 at δ=1/8\delta= 1/8, shifting from qx∗/2π=7/16q_x^*/2\pi= 7/16 (λs=16\lambda_s = 16, t′=0t' = 0) to 3/83/8 (λs=8\lambda_s = 8, t′/t=−0.2t'/t = -0.2). (C) Kinetic (ΔEkin/N\Delta E_{\mathrm{kin}}/N), interaction (UΔdU\Delta d), and total (ΔE/N\Delta E/N) energy changes on 16×416 \times4 (t′/t=−0.2t'/t = -0.2, τ=1.6\tau= 1.6) from λc=8\lambda_c = 8 to λc=4\lambda_c = 4 single-sector control (left), λc=8\lambda_c = 8 to λc=4\lambda_c = 4 multi-sector K=8K = 8 (middle), and pinned K=1K = 1 to multi-sector K=8K = 8 at fixed λc=4\lambda_c = 4 (right). Error bars denote 1 s.e.

The same period-halving transition occurs on the isotropic 256-site 16×1616 \times16 torus at δ=1/8\delta= 1/8: the Bragg peak of Sz(qx,π)S_z(q_x,\pi) shifts from qx∗/2π=7/16q_x^*/2\pi= 7/16 (λs=16\lambda_s = 16, λc=8\lambda_c = 8) at t′=0t' = 0 to 3/83/8 (λs=8\lambda_s = 8, λc=4\lambda_c = 4) at t′/t=−0.2t'/t = -0.2 (Figure 2B), replacing two broad filled walls with four half-filled antiphase walls in real space (Figure 1D,E) and increasing the intra-wall nearest-neighbor pairing vertex more than fivefold. Decomposing the total energy into hopping and on-site repulsion, E/N=Ekin/N+UdE/N = E_{\mathrm{kin}}/N + Ud with double occupancy d=⟨n^↑n^↓⟩d = \langle\hat{n}_{\uparrow}\hat{n}_{\downarrow}\rangle (Figure 2C), reveals why half-filled stripes win: splitting filled λc=8\lambda_c = 8 walls into twice as many half-filled λc=4\lambda_c = 4 walls increases local double occupancy and raises the Hubbard interaction energy (UΔd>0U\Delta d > 0), yet this repulsion penalty is outweighed more than twofold by the gain in transverse kinetic energy (ΔEkin/N<0\Delta E_{\mathrm{kin}}/N < 0) as holes hop across closer antiphase domain boundaries without leaving extended strings of overturned spins. Sublattice kinetic frustration thus unifies both sides of the phase diagram: positive effective t′/tt'/t on the electron-doped side assists intrasublattice motion within a uniform Néel background, whereas negative t′/tt'/t frustrates intrasublattice hopping on the hole-doped side and stabilizes half-filled antiphase walls.

Controlled stripe superpositions: pinned and delocalized domain walls pair alike

Testing whether transverse stripe fluctuations amplify dd-wave pairing requires separating three situations that are easily conflated: (i) a pinned stripe, whose domain walls sit at fixed lattice columns with static hole-density modulation A4>0A_4 > 0; (ii) an incoherent translation mixture of shifted copies of that pinned stripe, which sets the one-body modulation to A4=0A_4 = 0 and transfers its Bragg weight N4A42\frac{N}{4}A_4^2 into the connected density structure factor Nc(π/2,0)N_c(\pi/2,0), while leaving every linear translation-averaged observable—total energy E/NE/N, full Scc(q)=N4A42δq,(±π/2,0)+Nc(q)S_{\mathrm{cc}}(\mathbf{q}) = \frac{N}{4}A_4^2\delta_{\mathbf{q},(\pm\pi/2,0)} + N_c(\mathbf{q}), bare Pd(r)P_d(\mathbf{r}), and bare dd-wave Bloch eigenvalues—identically unchanged [whereas Wick-subtracted Cp(r)=Pd(r)−N0[Gˉ](r)C_p(\mathbf{r}) = P_d(\mathbf{r}) - N_0[\bar{G}](\mathbf{r}) shifts slightly only because Gˉ\bar{G} is translation-averaged before squaring]; and (iii) a coherent quantum superposition of shifted stripe sectors, which differs from the classical mixture solely through off-diagonal quantum interference between distinct domain-wall positions.

Figure 3 carries out this controlled comparison on 16×416 \times4 at fixed Hamiltonian and projection depth (δ=1/8\delta= 1/8, t′/t=−0.2t'/t = -0.2, τ=1.60\tau= 1.60), contrasting a pinned single-reference stripe (K=1K = 1), a single-sector K=8K = 8 control localized in one wall sector, and delocalized multi-sector K=8K = 8 and K=16K = 16 superpositions distributed across translated wall positions. Relative to the matched single-sector K=8K = 8 control, coherent multi-sector mixing suppresses the intrinsic-frame period-four modulation amplitude A4A_4 roughly threefold while shifting weight into the connected charge peak Nc(π/2,0)N_c(\pi/2,0) at conserved total Scc(π/2,0)S_{\mathrm{cc}}(\pi/2,0) (Figure 3A), and re-optimizing the projector under coherent shifted-stripe superposition lowers the variational energy by ∼3t\sim3t per site below the pinned state.

Figure 3. Multi-reference stripe unpinning and local dd-wave pairing at δ=1/8\delta= 1/8, t′/t=−0.2t'/t = -0.2. (A) Intrinsic 2D hole-density maps comparing the K=8K = 8 single-sector control and fluctuating multi-sector K=8K = 8 state on 16×416 \times4 (τ=1.60\tau= 1.60; top), together with 1D hole profiles δ(x)\delta(x) and period-four amplitudes A4×103A_4 \times10^3 across K=1K = 1 pinned, K=8K = 8 control, K=8K = 8 fluctuating, and K=16K = 16 fluctuating states (bottom). (B) Nearest-neighbor connected dd-wave vertex along the walls [Cp(0,1)C_p(0,1)], across the walls [Cp(1,0)C_p(1,0)], and scaled cumulative short-range sum 14χSCvtx(R≤4)\frac{1}{4}\chi_{\mathrm{SC}}^{\mathrm{vtx}}(R \le4) across the K=1→16K = 1 \to16 unpinning series, showing surviving local dd-wave pairing without finite-range enhancement. Error bars denote 1 s.e.

Across the entire K=1→16K = 1 \to16 unpinning series, short-range connected dd-wave pairing remains robust and essentially unchanged (Figure 3B). Subtracting the normal Wick contraction N0(r)N_0(r) from Pd(r)P_d(r) yields a positive nearest-neighbor connected vertex along the walls in all four states; by contrast, matched-depth uniform paramagnetic and filled λc=8\lambda_c = 8 reference controls at identical Hamiltonian and projector architecture collapse both variational energy and local Cp(0,1)C_p(0,1) toward zero, confirming that the half-filled antiphase wall geometry itself generates the local pairing vertex. While expanding from K=1K = 1 to K≥8K \ge8 restores local bond isotropy between Cp(1,0)C_p(1,0) and Cp(0,1)C_p(0,1), this restoration is already present in the pinned K=8K = 8 single-sector control; comparing the delocalized K=8K = 8 superposition directly against its matched single-sector control reveals no statistically resolved change in either the cumulative short-range vertex χSCvtx(R≤4)\chi_{\mathrm{SC}}^{\mathrm{vtx}}(R \le4) or its strictly non-overlapping bond subset where three-site contact terms vanish identically. At inter-stripe separations (Δx=λc=4\Delta x = \lambda_c = 4) and across distant bond pairs, connected pair correlations remain unresolved within uncertainty, and split-block cross-validated Rayleigh quotients of the full 2N×2N2N \times2N bond-singlet pair matrix Γbb′=⟨b^b†b^b′⟩\Gamma_{bb'} = \langle\hat{b}^{\dagger}_b \hat{b}_{b'} \rangle on 64-site (16×416 \times4 and 8×88 \times8) tori remain non-extensive (λ1CV/N≈8\lambda_1^{\mathrm{CV}}/N \approx8–9×10−39 \times10^{-3}) with comparable connected weight in uniform q=(0,0)\mathbf{q} = (0,0) and period-eight (π/4,0)(\pi/4,0) modes. Local dd-wave pairing is therefore already an intrinsic property of an individual half-filled domain wall, whereas unpinning the charge profile within our multi-reference subspace neither amplifies local pairing nor induces resolved inter-stripe coherence.

Ground-state energy benchmarks on large two-dimensional square tori

Figure 4 and Table I benchmark the ground-state energies of our multi-reference MR-nVAFQMC wavefunctions at δ=1/8\delta= 1/8 and U/t=8U/t = 8 across periodic 16×1616 \times16 (256-site) and 32×3232 \times32 (1,024-site) square tori. Throughout, we separate fixed-checkpoint variational upper bounds from empirical Sorella projection-depth extrapolations E(τ)/N=E∞/N+aτ−3E(\tau)/N = E_{\infty}/N + a\tau^{-3} [19] (marked “not variational” in Table I and shown as open stars in Figure 4), alongside both variational and extrapolated literature benchmarks. For the periodic 16×1616 \times16 VAFQMC benchmark, E/N=−0.75865(3)E/N = -0.75865(3) is the Sorella–Tocchio result reported in the V-score dataset of Wu et al. [34] and reproduced in Table S4 of Chen et al. [10]; Ref. [19] describes the underlying method and its twist-averaged TDL extrapolation.

Figure 4. Ground-state energy benchmarks across large periodic 2D square tori at δ=1/8\delta= 1/8 (U/t=8U/t = 8). Filled or solid-outlined symbols denote strictly variational energies; open symbols denote empirical τ→∞\tau\to\infty or D→∞D \to\infty extrapolations. Because Monte Carlo standard errors span two orders of magnitude across sign-free VMC and finite-sign VAFQMC, error bars are omitted from the plot and listed in Table I. (A) Energy per site E/NE/N on 16 × 16 and 32 × 32 tori for the nearest-neighbor Hubbard model (t′/t=0t'/t = 0), comparing HFPS [36], Transformer-BF [40], SCALE and ACE-BF [41], 16 × 16 PBC variational VAFQMC [10, 19, 34] and twist-averaged thermodynamic-limit (TDL) τ→∞\tau\to\infty Sorella VAFQMC [19], variational (D=16D = 16) and D→∞D \to\infty extrapolated fPEPS, NN-fPEPS (D=4D = 4), BCS backflow, and Pfaffian backflow [10] with our multi-reference (K=4K = 4) variational curve and same-manifold τ−3→0\tau^{-3} \to0 extrapolations (K=4K = 4 on 16 × 16, K=1K = 1 on 32 × 32; open vermilion stars). (B) Energy per site E/NE/N for the cuprate Hubbard model (t′/t=−0.2t'/t = -0.2) across 16 × 16, 24 × 24, and 32 × 32, comparing HFPS [36], SCALE and ACE-BF [41], symmetry-preserving BCS backflow [35], and NN-fPEPS (D=4D = 4) and Pfaffian backflow [10] with our multi-reference variational benchmarks (K=8K = 8 on 16 × 16, K=4K = 4 on 32 × 32) and same-manifold τ−3→0\tau^{-3} \to0 extrapolations (K=1K = 1 on 16 × 16, K=4K = 4 on 32 × 32; open vermilion stars).

Across both 16×1616 \times16 and 32×3232 \times32 tori, MR-nVAFQMC reaches variational energies competitive with the most accurate neural- and tensor-network simulations reported to date (Figure 4 and Table I). In the nearest-neighbor model (t′/t=0t'/t = 0; Figure 4A), our K=4K = 4 states (τ=1.40\tau= 1.40 on 16×1616 \times16, τ=1.31\tau= 1.31 on 32×3232 \times32) match or lie slightly below BCS backflow, Sorella VAFQMC, Pfaffian backflow, ACE backflow, and D=4D = 4 NN-FPEPS within statistical uncertainty while lying well below unprojected SCALE and variational D=16D = 16 fPEPS [10, 19, 34, 41]; extrapolating E(τ)/NE(\tau)/N in τ−3\tau^{-3} on 16×1616 \times16 (−0.7604(32)-0.7604(32)) further agrees closely with both 16×1616 \times16 D→∞D \to\infty extrapolated fPEPS (−0.7602(14)-0.7602(14)) [10] and Sorella’s twist-averaged 2D thermodynamic-limit τ→∞\tau\to\infty estimate (−0.7605(1)-0.7605(1)) [19]. In the frustrated cuprate model (t′/t=−0.2t'/t = -0.2; Figure 4B), our K=8K = 8 (τ=1.45\tau= 1.45) checkpoint on 16 × 16 and K=4K = 4 (τ=1.26\tau= 1.26) checkpoint on 32 × 32 likewise match or lie slightly below Pfaffian backflow, symmetry-preserving BCS backflow, and D=4D = 4 NN-fPEPS within statistical error [10, 35], demonstrating that particle-conserving multi-determinant neural auxiliary-field states access the same milli-tt low-energy regime without geminal pairing initialization.

Directional pairing decay and exploratory transferred-parameter scaling

We next examine how connected dd-wave pairing correlations decay spatially on 16 × 16 and how spin and pairing observables behave in an exploratory fixed-depth series up to 32 × 32 (Figure 5). Along the stripe direction on 16 × 16 (t′/t=−0.2t'/t = -0.2; Figure 5A, top), the bare two-bond correlator Pd(0,Δy)P_d(0,\Delta y) at short distance is dominated by the disconnected one-body Wick contribution N0(0,Δy)N_0(0,\Delta y), underscoring the need to subtract N0N_0. The resulting connected intra-stripe vertex Cp(0,Δy)=Pd−N0C_p(0,\Delta y) = P_d - N_0 is clearly positive at nearest-neighbor separation (Δy=1\Delta y = 1) across K=1,4,8K = 1, 4, 8, but drops sharply already by Δy=2\Delta y = 2—where K=1K = 1, 4 lie near zero within 1σ1\sigma and K=8K = 8 fluctuates slightly negative within uncertainty—and remains statistically indistinguishable from zero over Δy=6\Delta y = 6–8 (Figure 5A, inset). Perpendicular to the stripes (Figure 5A, bottom), Cp(Δx,0)C_p(\Delta x, 0) decays even more rapidly, becoming indistinguishable from zero within a single charge-stripe period (Δx=λc=4\Delta x = \lambda_c = 4) and beyond. Deepening projection on 16 × 16 up to τ=1.45\tau= 1.45–1.52 closes the energy offset relative to Pfaffian backflow while leaving both stripe spin order and distant-window pairing unchanged within 1σ1\sigma, and doping scans into δ=1/6\delta= 1/6 and 2/92/9 on 12 × 12 similarly enhance quasi-1D along-stripe Cp(0,1)C_p(0,1) while transverse and distant-window pairing remain unresolved.

Figure 5. Directional decay of connected dd-wave pairing on 16×1616 \times16 and exploratory transferred-parameter scaling up to 32×3232 \times32 at δ=1/8\delta= 1/8. (A) Directional 1D two-bond cuts on 16×1616 \times16 (t′/t=−0.2t'/t = -0.2). Top: along-stripe bare Pd(0,Δy)P_d(0,\Delta y) (open diamonds), Wick background N0(0,Δy)N_0(0,\Delta y) (grey squares) for K=1K = 1 (τ=1.52\tau= 1.52), and connected vertex Cp(0,Δy)=Pd−N0C_p(0,\Delta y) = P_d - N_0 for K=1,4,8K = 1, 4, 8 (inset: Δy=6\Delta y = 6–88). Bottom: across-stripe Cp(Δx,0)C_p(\Delta x,0) (Δx=λc=4\Delta x = \lambda_c = 4 shaded). (B,C) Exploratory fixed-depth transferred-parameter scaling (1/L→01/L \to0) at (B) t′/t=0t'/t = 0 (K=1K = 1, τ=1.16\tau= 1.16; L=16,32L = 16, 32) and (C) t′/t=−0.2t'/t = -0.2 (K=4K = 4, τ=1.15\tau= 1.15; L=16,24,32L = 16, 24, 32), comparing full-SU(2) stripe spin order Mstr,4p2,fullM_{\mathrm{str},4p}^{2,\mathrm{full}} (top) and the exact four-bond site-centered singlet distant-window pairing ΔSC,4b2×103\Delta_{\mathrm{SC},4b}^{2} \times10^3 Eq. (5), matching Ref. [10]; bottom of This work (transf.) (vermilion circles and dashed guides; L=24,32L = 24, 32 parameters transferred from 16×1616 \times16 without per-size re-optimization) together with unified same-checkpoint Table 1 SOTA evaluations on 16×1616 \times16 (K=8K = 8, τ=1.45\tau= 1.45) and 32×3232 \times32 (K=4K = 4, τ=1.26\tau= 1.26) (purple diamonds in C) against NN-fPEPS, Pfaffian backflow, and BCS backflow [10, 35] as well as pinned-field CP-AFQMC [31] (open star). Error bars denote 1 s.e. [omitted for Ref. [10] points whose uncertainties lie below the quoted 3×10−43 \times10^{-4} plot-unit threshold; the Ref. [31] star shows the asymmetric 1σ1\sigma interval propagated from Δd=0.017(3)\Delta_d = 0.017(3)].

To compare distant-window pairing directly with Chen et al. [10] across commensurate square tori up to 32 × 32, Figure 5B,C evaluates the full-SU(2) four-peak stripe spin order Mstr,4p2,fullM_{\mathrm{str,4p}}^{2,\mathrm{full}} together with the exact four-bond site-centered singlet pairing order parameter ΔSC,4b2(L)\Delta_{\mathrm{SC,4b}}^2(L) matching (2) of Ref. [10] (Section III). Two caveats govern the interpretation of the common-depth multi-size series and motivate labeling Figure 5B,C as exploratory. First, the fixed-depth series (τ=1.16\tau= 1.16 at t′=0t' = 0; τ=1.15\tau= 1.15 at t′/t=−0.2t'/t = -0.2) uses shallower wavefunctions at higher variational energies than the deeper τ=1.26\tau= 1.26–1.451.45 energy-benchmark checkpoints of Table I [for instance, on 32×3232 \times32 at t′/t=−0.2t^{\prime}/t = -0.2, the τ=1.15\tau= 1.15 evaluation gives E/N=−0.736(6)E/N = -0.736(6) compared with −0.7476(37) for the τ=1.26\tau= 1.26 checkpoint in Table I]. Second, the 24×2424 \times24 and 32×3232 \times32 wavefunctions in the common-depth series are constructed by tiling and transferring parameters optimized on 16×1616 \times16 without independent variational re-optimization on the larger tori (steps = 0 on 32×3232 \times32), and under G=D2×Z2spinG = D_2 \times Z^{\mathrm{spin}}_2 (∣G∣=8|G| = 8) these large-torus states remain frame-pinned (A4>0A_4 > 0). To remove checkpoint mismatch directly at L=16L = 16 and L=32L = 32, we additionally performed unified same-checkpoint joint evaluations of energy, Mstr,4p2,fullM_{\mathrm{str},4p}^{2,\mathrm{full}}, and four-bond ΔSC,4b2\Delta_{\mathrm{SC},4b}^{2} on the exact Table I SOTA wavefunctions (16×1616 \times16, K=8K = 8, τ=1.45\tau= 1.45; 32×3232 \times32, K=4K = 4, τ=1.26\tau= 1.26; purple diamonds in Figure 5C).

Within both the common-depth transferred series and the unified Table I SOTA checkpoints, Mstr,4p2,full(L)M_{\mathrm{str},4p}^{2,\mathrm{full}}(L) retains pronounced period-16 (t′=0t^{\prime} = 0; Figure 5B, top) and period-8 (t′/t=−0.2t^{\prime}/t = -0.2; Figure 5C, top) stripe spin order up to 32×3232 \times32, comparable in magnitude to D=4D = 4 NN-fPEPS and Pfaffian backflow [10], whereas C4vC_{4v}-symmetric BCS backflow suppresses stripe magnetism toward a nearly uniform state [10, 35]. In the four-bond pairing channel (Figure 5B, bottom), ΔSC,4b2(L)\Delta_{\mathrm{SC},4b}^{2}(L) is compatible with zero at every finite size in the transferred series [giving a three-point linear 1/L1/L intercept of (−0.01±0.06)×10−3(-0.01 \pm0.06) \times10^{-3} at t′/t=−0.2t^{\prime}/t = -0.2] and remains statistically indistinguishable from zero on the exact Table I SOTA checkpoints on 16×1616 \times16 and 32×3232 \times32 (purple diamonds); likewise, the maximum site-centered dx2−y2d_{x^2-y^2} Bloch eigenvalue density max⁡QΓ~d(Q)/N\max_Q \widetilde{\Gamma}_d(Q)/N across all momenta QQ—including uniform (0,0)(0, 0) and PDW (π/4,0)(\pi/4, 0), (π/2,0)(\pi/2, 0)—scales downward as O(1/N)O(1/N) from 8×88 \times8 to 32×3232 \times32. By comparison, while C4vC_{4v}-symmetric BCS backflow extrapolates to a sizable uniform order parameter [10, 35], the lower-energy striped wavefunctions of Chen et al. [10] retain positive finite-size pairing on 16×1616 \times16–32×3232 \times32 that extrapolates to very weak or fit-window-sensitive superconductivity in the thermodynamic limit [ΔSC=0.007(2)\Delta_{\mathrm{SC}} = 0.007(2) for D=4D = 4 NN-fPEPS, and 0.010(3)0.010(3) from L=24,32L = 24, 32 versus a 1σ1\sigma bound ΔSC<0.006\Delta_{\mathrm{SC}} < 0.006 from L=16,24,32L = 16, 24, 32 for Pfaffian backflow]. Because ΔSC≈0.007\Delta_{\mathrm{SC}} \approx0.007 corresponds to ΔSC2≈0.05×10−3\Delta_{\mathrm{SC}}^{2} \approx0.05 \times10^{-3}, comparable to our statistical uncertainty, our results neither resolve long-range pairing statistically nor exclude weak superconductivity at the NN-fPEPS scale; establishing fully converged ground-state scaling across all sizes will still require independent per-size variational re-optimization at matched deep τ\tau across L=16,24,32L = 16, 24, 32.

Method

Multi-reference neural VAFQMC (MR-nVAFQMC) wavefunction

Building on the general strategy of superposing nonorthogonal, symmetry-projected determinants [42], we construct the multi-reference, symmetry-restored neural VAFQMC (MR-nVAFQMC) wavefunction

∣Ψ(θ,c)⟩=∑k=1KckP^GP^θ(τ)∣Φk⟩,(2)\lvert\Psi(\theta,c)\rangle= \sum_{k=1}^{K} c_k \hat{P}_G \hat{P}_{\theta}(\tau)\lvert\Phi_k\rangle, \tag*{(2)}

where {∣Φk⟩}k=1K\{\lvert\Phi_k\rangle\}_{k=1}^{K} is a pool of inequivalent canonical particle-conserving Slater determinants spanning candidate stripe periods and shifted domain-wall configurations, P^G=∣G∣−1∑g∈GU^g\hat{P}_G = \lvert G\rvert^{-1}\sum_{g\in G}\hat{U}_g projects onto a discrete lattice and spin-flip symmetry group GG of order ∣G∣\lvert G\rvert, and P^θ(τ)\hat{P}_{\theta}(\tau) is a shared parameterized auxiliary-field correlation projector of imaginary-time depth τ\tau, extending single-reference VAFQMC [19] to coherent multi-determinant superpositions. Applying symmetric Suzuki–Trotter factorization and discrete Hubbard–Stratonovich decoupling across LτL_{\tau} slices writes P^θ(τ)=∑xpθ(x)B^θ(x)\hat{P}_{\theta}(\tau)=\sum_{\mathbf{x}}p_{\theta}(\mathbf{x})\hat{B}_{\theta}(\mathbf{x}), where x∈{±1}LτN\mathbf{x}\in\{\pm1\}^{L_{\tau}N} denotes the auxiliary Ising fields and B^θ(x)\hat{B}_{\theta}(\mathbf{x}) is a product of spin-factored one-body exponentials. Beyond tunable Trotter slice weights, we incorporate a neural network in the auxiliary-field sector that takes the auxiliary-field configuration x\mathbf{x} as input and modulates the auxiliary-field path weights and effective one-body slice couplings shared across all KK reference determinants, increasing expressive flexibility while keeping every field configuration an exact one-body Slater-determinant propagation. On 64-site (16×416\times4, 8×88\times8) and 16×1616\times16 tori we also employ stripe or full space-group projections (∣G∣=64,128,256,512\lvert G\rvert=64,128,256,512), whereas for the 16×1616\times16 (t′/t=−0.2t'/t=-0.2, K=8K=8) and 32×3232\times32 multi-reference benchmarks in Table 1 we use the eight-element group G=D2×Z2spinG=D_2\times\mathbb{Z}_2^{\mathrm{spin}} (∣G∣=8\lvert G\rvert=8) generated by lattice reflections Rx,RyR_x,R_y and global spin flip ↑↔↓\uparrow\leftrightarrow\downarrow, because summing many off-diagonal translations at large τ\tau increases computational cost and sign fluctuations while the sub-extensive projection gain ΔEsymm/N∼O(1/N)\Delta E_{\mathrm{symm}}/N\sim O(1/N) becomes small on 256–1,024-site tori where shifted stripe sectors are represented through the multi-reference pool and translation-averaged observables. Whereas each individual unprojected Slater determinant ∣Φk⟩\lvert\Phi_k\rangle carries no explicit BCS or Pfaffian pairing field and has zero connected two-body pairing (Cp=0C_p=0), symmetry restoration P^G\hat{P}_G, multi-determinant interference (K>1K>1), and the neural auxiliary-field projector P^θ(τ)\hat{P}_{\theta}(\tau) together generate non-Gaussian spin, charge, and pairing correlations without geminal pairing initialization. Superposition weights c\mathbf{c} (c1≡1c_1\equiv1), reference orbitals, and projector parameters θ\theta are optimized jointly by damped stochastic reconfiguration, followed by frozen-checkpoint sign-reweighted Markov-chain evaluation [19] with statistical uncertainties estimated by coarse-grained block jackknife over independent Markov chains after thermalization burn-in.

Lattice, t′/tt^{\prime}/tMethodE/NE/N (tt)
16 × 16, 0MR-nVAFQMC (this work) (K=4K = 4, ∣G∣=8|G| = 8, τ→∞\tau\to\infty extrap., not variational)−0.7604(32)
fPEPS, D→∞D \to\infty extrap. (Chen 2026, not variational)−0.7602(14)
MR-nVAFQMC (this work) (K=4K = 4, τ=1.40\tau= 1.40, ∣G∣=128|G| = 128)−0.7589(19)
BCS backflow (Chen 2026)−0.75878(2)
VAFQMC (Sorella 2023; via Wu 2024, Chen 2026)−0.75865(3)
NN-fPEPS, D=4D = 4 (Chen 2026)−0.75851(2)
16 × 16, −0.2nVAFQMC (this work) (K=1K = 1, ∣G∣=64|G| = 64, τ→∞\tau\to\infty extrap., not variational)−0.7482(19)
MR-nVAFQMC (this work) (K=8K = 8, τ=1.45\tau= 1.45, ∣G∣=8|G| = 8)−0.7466(38)
Pfaffian backflow (Chen 2026)−0.74533(1)
NN-fPEPS, D=4D = 4 (Chen 2026)−0.74467(2)
SBP [35]−0.74411(1)
32 × 32, 0nVAFQMC (this work) (K=1K = 1, ∣G∣=8|G| = 8, τ→∞\tau\to\infty extrap., not variational)−0.7581(17)
MR-nVAFQMC (this work) (K=4K = 4, τ=1.31\tau= 1.31, ∣G∣=8|G| = 8)−0.7577(26)
NN-fPEPS, D=4D = 4 (Chen 2026)−0.75677(1)
BCS backflow (Chen 2026)−0.75617(1)
SCALE det. backflow [40] [41]−0.7501
32 × 32, −0.2MR-nVAFQMC (this work) (K=4K = 4, ∣G∣=8|G| = 8, τ→∞\tau\to\infty extrap., not variational)−0.7494(35)
MR-nVAFQMC (this work) (K=4K = 4, τ=1.26\tau= 1.26, ∣G∣=8|G| = 8)−0.7476(37)
Pfaffian backflow (Chen 2026)−0.74569(1)
NN-fPEPS, D=4D = 4 (Chen 2026)−0.74257(2)

Table 1. TABLE I. Best MR-nVAFQMC ground-state energies per site (this work) on large 2D square tori at δ=1/8\delta= 1/8 (U/t=8U/t = 8) compared with leading published benchmarks. For each periodic L×LL \times L torus and t′/tt^{\prime}/t, our best multi-reference strictly variational state and same-manifold τ−3→0\tau^{-3} \to0 extrapolation (marked “not variational”) plotted in Figure 4 are listed alongside the lowest published periodic-torus benchmarks [10, 19, 34, 35, 41] [for reference, Sorella’s twist-averaged 2D thermodynamic-limit τ→∞\tau\to\infty extrapolation at t′=0t^{\prime} = 0 is −0.7605(1) [19]]. Numbers in parentheses denote 1 s.e. statistical uncertainty in the last displayed digit(s) [e.g., −0.7589(19) denotes −0.7589 ± 0.0019]; ∣G∣=8|G| = 8 denotes projection onto lattice reflections (RxR_x, RyR_y) and global spin flip (D2×Z2spinD_2 \times Z^{\mathrm{spin}}_2), while ∣G∣=64,128|G| = 64, 128 additionally include commensurate stripe translations (Section III).

Spin, charge, and connected pairing observables

From the equal-time spin and density correlators ⟨S^i⋅S^j⟩=⟨S^izS^jz⟩+12⟨S^i+S^j−+S^i−S^j+⟩\langle\hat{\mathbf{S}}_i\cdot\hat{\mathbf{S}}_j\rangle=\langle\hat{S}^z_i\hat{S}^z_j\rangle+\frac{1}{2}\langle\hat{S}^+_i\hat{S}^-_j+\hat{S}^-_i\hat{S}^+_j\rangle and ⟨n^in^j⟩\langle\hat{n}_i\hat{n}_j\rangle (n^i=n^i↑+n^i↓\hat{n}_i=\hat{n}_{i\uparrow}+\hat{n}_{i\downarrow}), longitudinal spin structure factors and the full-SU(2) four-peak stripe magnetic order parameter (matching Ref. [10]) are obtained from

Sz(q)=1N∑i,jeiq⋅(ri−rj)⟨S^izS^jz⟩,Mstr,4p2,full=14N∑α=14S(qα),(3)S_z(\mathbf{q})=\frac{1}{N}\sum_{i,j}e^{i\mathbf{q}\cdot(\mathbf{r}_i-\mathbf{r}_j)}\langle\hat{S}^z_i\hat{S}^z_j\rangle,\qquad M_{\mathrm{str,4p}}^{2,\mathrm{full}}=\frac{1}{4N}\sum_{\alpha=1}^{4}S(\mathbf{q}_{\alpha}), \tag*{(3)}

where S(q)=N−1∑i,jeiq⋅(ri−rj)⟨S^i⋅S^j⟩S(\mathbf{q})=N^{-1}\sum_{i,j}e^{i\mathbf{q}\cdot(\mathbf{r}_i-\mathbf{r}_j)}\langle\hat{\mathbf{S}}_i\cdot\hat{\mathbf{S}}_j\rangle and qα∈{((π±2πϵspin,π),(π,π±2πϵspin))}\mathbf{q}_{\alpha}\in\{((\pi\pm2\pi\epsilon_{\mathrm{spin}},\pi),(\pi,\pi\pm2\pi\epsilon_{\mathrm{spin}}))\} with spin incommensurability ϵspin=1/λs\epsilon_{\mathrm{spin}}=1/\lambda_s. Static 1D hole-density modulation δ(x)=1−Ly−1∑y⟨n^x,y⟩\delta(x)=1-L_y^{-1}\sum_y\langle\hat{n}_{x,y}\rangle at commensurate stripe period λc\lambda_c (wavevector qx=2π/λcq_x=2\pi/\lambda_c, e.g., A4A_4 at qx=π/2q_x=\pi/2) and translation-invariant connected charge fluctuations are quantified by

Aλc=2Lx∣∑x=0Lx−1δ(x)e−2πix/λc∣,Nc(q)=1N∑i,jeiq⋅(ri−rj)(⟨n^in^j⟩−⟨n^i⟩⟨n^j⟩),(4)A_{\lambda_c}=\frac{2}{L_x}\left|\sum_{x=0}^{L_x-1}\delta(x)e^{-2\pi ix/\lambda_c}\right|,\qquad N_c(\mathbf{q})=\frac{1}{N}\sum_{i,j}e^{i\mathbf{q}\cdot(\mathbf{r}_i-\mathbf{r}_j)}\left(\langle\hat{n}_i\hat{n}_j\rangle-\langle\hat{n}_i\rangle\langle\hat{n}_j\rangle\right), \tag*{(4)}

satisfying Nc(0)=0N_c(0)=0 identically in the canonical ensemble.

With normalized nearest-neighbor singlet creation operator b^ij†=(c^i↑†c^j↓†−c^i↓†c^j↑†)/2\hat{b}^{\dagger}_{ij}=(\hat{c}^{\dagger}_{i\uparrow}\hat{c}^{\dagger}_{j\downarrow}-\hat{c}^{\dagger}_{i\downarrow}\hat{c}^{\dagger}_{j\uparrow})/\sqrt{2} and dx2−y2d_{x^2-y^2} bond signs h±x^=+1h_{\pm\hat{x}}=+1, h±y^=−1h_{\pm\hat{y}}=-1, we distinguish two standard dd-wave pair operators: the two-bond link-origin operator Δ^2b,r†\hat{\Delta}^{\dagger}_{2b,\mathbf{r}} used for short-range vertices and 1D directional cuts (Figure 1B, Figure 3B, and Figure 5A), and the four-bond site-centered operator Δ^4b,r†\hat{\Delta}^{\dagger}_{4b,\mathbf{r}} matching Eq. (2) of Ref. [10] used for distant-window comparisons in Figure 5B,C,

Δ^2b,r†=12∑η∈{+x^,+y^}hηb^r,r+η†,Δ^4b,r†=14∑η∈{±x^,±y^}hηb^r,r+η†.(5)\hat{\Delta}^{\dagger}_{2b,\mathbf{r}}=\frac{1}{2}\sum_{\boldsymbol{\eta}\in\{+\hat{x},+\hat{y}\}}h_{\eta}\hat{b}^{\dagger}_{\mathbf{r},\mathbf{r}+\boldsymbol{\eta}},\qquad\hat{\Delta}^{\dagger}_{4b,\mathbf{r}}=\frac{1}{4}\sum_{\boldsymbol{\eta}\in\{\pm\hat{x},\pm\hat{y}\}}h_{\eta}\hat{b}^{\dagger}_{\mathbf{r},\mathbf{r}+\boldsymbol{\eta}}. \tag*{(5)}

From the bare pair correlator Pd(r)=N−1∑r0⟨Δ^r0†Δ^r0+r⟩P_d(\mathbf{r})=N^{-1}\sum_{\mathbf{r}_0}\langle\hat{\Delta}^{\dagger}_{\mathbf{r}_0}\hat{\Delta}_{\mathbf{r}_0+\mathbf{r}}\rangle built from bond-singlet amplitudes ⟨b^b†b^b′⟩\langle\hat{b}^{\dagger}_{b}\hat{b}_{b'}\rangle [b=(i,j)b=(i,j), b′=(k,l)b'=(k,l)], subtracting the disconnected normal Wick contraction built from the full state’s symmetrized equal-time one-body Green’s function Gˉσ;ij=⟨c^iσ†c^jσ⟩\bar{G}_{\sigma;ij}=\langle\hat{c}^{\dagger}_{i\sigma}\hat{c}_{j\sigma}\rangle,

Cp(r)=Pd(r)−N0(r),N0,bb′=12(Gˉ↑;ikGˉ↓;jl+Gˉ↑;ilGˉ↓;jk+Gˉ↓;ikGˉ↑;jl+Gˉ↓;ilGˉ↑;jk),(6)C_p(\mathbf{r})=P_d(\mathbf{r})-N_0(\mathbf{r}),\qquad N_{0,bb'}=\frac{1}{2}\left(\bar{G}_{\uparrow;ik}\bar{G}_{\downarrow;jl}+\bar{G}_{\uparrow;il}\bar{G}_{\downarrow;jk}+\bar{G}_{\downarrow;ik}\bar{G}_{\uparrow;jl}+\bar{G}_{\downarrow;il}\bar{G}_{\uparrow;jk}\right), \tag*{(6)}

isolates genuine connected two-body pairing in either convention [10, 43] [used throughout Figure 3 and Figure 5; the broad doping scan in Figure 1B uses a uniform archive transition-N0N_0 subtraction across all points that agrees within statistical uncertainty with Eq. (6), e.g., Cp(0,1)=(+4.14±0.47)×10−3C_p(0,1)=(+4.14\pm0.47)\times10^{-3} versus (+4.35±0.61)×10−3(+4.35\pm0.61)\times10^{-3} on 16×416\times4 at δ=1/8\delta=1/8]. From the two-bond vertex we form the cumulative short-range sum χSC(2b)(R≤4)=∑0<∣r∣≤4Cp(2b)(r)\chi_{\mathrm{SC}}^{(2b)}(R\le4)=\sum_{0<|\mathbf{r}|\le4}C_p^{(2b)}(\mathbf{r}), while from the four-bond vertex we evaluate the 2D distant-window average matching Ref. [10],

ΔSC,4b2(L)=1ML∑∣r∣≥L/(22)Cp(4b)(r),(7)\Delta_{\mathrm{SC,4b}}^{2}(L)=\frac{1}{M_L}\sum_{|\mathbf{r}|\ge L/(2\sqrt{2})}C_p^{(4b)}(\mathbf{r}), \tag*{(7)}

over the ML=159M_L=159, 355, and 623 periodic minimum-image displacements outside L/(22)L/(2\sqrt{2}) on L=16L=16, 24, and 32. On 64-site tori, we also construct the full 2N×2N2N\times2N bond-singlet matrix Γbb′=⟨b^b†b^b′⟩\Gamma_{bb'}=\langle\hat{b}^{\dagger}_{b}\hat{b}_{b'}\rangle [44] and suppress positive noisy-matrix diagonalization bias via the 50/50 split-block cross-validated Rayleigh quotient λ1CV=12(v1,ATΓBv1,A+v1,BTΓAv1,B)\lambda^{\mathrm{CV}}_1=\frac{1}{2}(\mathbf{v}^{\mathsf{T}}_{1,A}\Gamma_B\mathbf{v}_{1,A}+\mathbf{v}^{\mathsf{T}}_{1,B}\Gamma_A\mathbf{v}_{1,B}), which provides a conservative (downward-biased in expectation relative to λ1(Γ)\lambda_1(\Gamma), up to finite-sample ratio bias) estimate of the leading pair-matrix weight.

Discussion and Conclusion

Using MR-nVAFQMC wavefunctions built from particle-conserving Slater determinants without explicit pairing fields—at variational energies matching or improving upon leading neural- and tensor-network benchmarks up to 32×3232 \times32 sites—our single-layer calculations establish a clear separation between intra-stripe local pair formation and inter-stripe phase coherence in the 2D tt–t′t'–UU Hubbard model. Negative next-nearest-neighbor hopping t′/t=−0.2t'/t=-0.2 switches the δ=1/8\delta=1/8 state from filled (λc=8\lambda_c=8) to half-filled (λc=4\lambda_c=4, λs=8\lambda_s=8) antiphase stripes obeying the Yamada relation ϵspin=δ\epsilon_{\mathrm{spin}}=\delta, stabilized by transverse kinetic delocalization against an opposing Hubbard repulsion penalty. Crucially, each half-filled domain wall already hosts robust short-range dd-wave singlet pairing: the connected intra-wall vertex is enhanced more than fivefold over the filled stripes at t′=0t'=0, absent on the electron-doped Néel branch, peaking on the overdoped side near δ=3/16\delta=3/16 where period-eight spin stripe order softens, and unchanged within uncertainty at δ=1/8\delta=1/8 whether the walls are pinned or coherently delocalized across translated stripe sectors so that static charge modulation drops threefold.

Comparing these strictly single-layer results with the experimental 1/81/8 anomaly in La1.875_{1.875}Ba0.125_{0.125}CuO4_4 highlights both an encouraging microscopic agreement and an instructive open tension [2, 5, 6, 7]. Experimentally, bulk 3D superconductivity at x=1/8x=1/8 occurs at only Tc≈4T_c \approx4 K [6], whereas each individual CuO2_2 layer exhibits 2D superconducting fluctuations setting in near the spin-stripe ordering scale (∼40\sim40 K) and reaches a 2D zero-resistance state near 16 K; the suppression of 3D TcT_c arises from interlayer Josephson decoupling, widely attributed to an intra-plane period-eight PDW that alternates sign between adjacent stripes and rotates by 90∘90^\circ from layer to layer [2, 6, 7]. On the local scale, our finding that half-filled λc=4\lambda_c=4 domain walls intrinsically form short-range dd-wave pairs—with uniform q=(0,0)\mathbf{q}=(0,0) and period-eight PDW q=(π/4,0)\mathbf{q}=(\pi/4,0) modes nearly degenerate in 64-site pair-matrix spectra because inter-wall Josephson coupling across the antiphase spin domain is very weak—captures the microscopic origin of intra-wall pairing and near-degenerate 0- versus π\pi-phase inter-stripe channels. On the macroscopic 2D scale, however, our single-layer states show transverse pair correlations decaying within one stripe spacing (Δx=4\Delta x=4) and resolve no long-range 2D superconducting order up to 32×3232 \times32, in contrast to the 2D planar superconductivity observed below 16 K in La1.875_{1.875}Ba0.125_{0.125}CuO4_4 [6] and consistent with the weak or statistically unresolved 2D pairing in lower-energy striped Pfaffian and NN-fPEPS wavefunctions [10].

Three concrete possibilities can reconcile this tension and define a clear roadmap for future work. First, a weak 2D intra-plane condensate at the scale corresponding to TBKT≈16 K∼4×10−3tT_{\mathrm{BKT}}\approx16\ \mathrm{K}\sim4\times10^{-3}t—such as the small amplitude ΔSC=0.007(2)\Delta_{\mathrm{SC}}=0.007(2) (ΔSC2≈5×10−5\Delta_{\mathrm{SC}}^2\approx5\times10^{-5}) suggested by D=4D=4 NN-fPEPS [10]—lies at the statistical noise floor of our present distant-window estimator, and the spatially summed ΔSC,4b2(L)\Delta_{\mathrm{SC},4b}^2(L) cancels any sign-alternating q=(π/4,0)\mathbf{q}=(\pi/4,0) PDW component. Second, although unified same-checkpoint evaluations on the lowest-energy Table I wavefunctions at 16×1616 \times16 and 32×3232 \times32 confirm strong stripe spin order with unresolved four-bond ΔSC,4b2\Delta_{\mathrm{SC},4b}^2, the common-depth multi-size series in Figure 5B,C still relies on transferred 16×1616 \times16 parameters on 24×2424 \times24 and 32×3232 \times32 under frame-pinned D2×Z2spinD_2 \times\mathbb{Z}_2^{\mathrm{spin}} symmetry (A4>0A_4>0). Independent per-size variational re-optimization at matched deep τ\tau, combined with direct superfluid-stiffness and translation-restored multi-sector scaling [45, 46], will test whether full larger-torus relaxation resolves weak 2D uniform or PDW coherence. Third, if converged ground states of the pure single-band 2D tt–t′t'–UU model at δ=1/8\delta=1/8 remain devoid of long-range 2D order, then stabilizing planar coherence at commensurate 1/81/8 lock-in requires ingredients beyond the minimal single-band Hubbard Hamiltonian—such as multi-orbital CuO2_2 physics, longer-range interactions, or low-temperature-tetragonal lattice couplings [2, 3, 7]—or doping away from 1/81/8 lock-in into the δ≳3/16\delta\gtrsim3/16 regime where spin stripes soften.

References

  1. [1]J. M. Tranquada, B. J. Sternlieb, J. D. Axe, Y. Nakamura, S. Uchida, Evidence for stripe correlations of spins and holes in copper oxide superconductors. Nature 375, 561–563 (1995).DOI
  2. [2]E. Fradkin, S. A. Kivelson, J. M. Tranquada, Colloquium: Theory of intertwined orders in high temperature superconductors. Rev. Mod. Phys. 87, 457–482 (2015).DOI
  3. [3]B. Keimer, S. A. Kivelson, M. R. Norman, S. Uchida, J. Zaanen, From quantum matter to high-temperature superconductivity in copper oxides. Nature 518, 179–186 (2015).DOI
  4. [4]K. Yamada et al., Doping dependence of the spatially modulated dynamical spin correlations and the superconducting-transition temperature in La₂−xSrxCuO₄. Phys. Rev. B 57, 6165–6172 (1998).DOI
  5. [5]A. R. Moodenbaugh, Y. Xu, M. Suenaga, T. J. Folkerts, R. N. Shelton, Superconducting properties of La₂−xBaxCuO₄. Phys. Rev. B 38, 4596–4600 (1988).
  6. [6]Q. Li, M. Hücker, G. D. Gu, A. M. Tsvelik, J. M. Tranquada, Two-dimensional superconducting fluctuations in stripe-ordered La₁.₈₇₅Ba₀.₁₂₅CuO₄. Phys. Rev. Lett. 99, 067001 (2007).pubmed.ncbi.nlm.nih.gov/17930855
  7. [7]E. Berg, E. Fradkin, E.-A. Kim, S. A. Kivelson, V. Oganesyan, J. M. Tranquada, S. C. Zhang, Dynamical layer decoupling in a stripe-ordered high-Tc superconductor. Phys. Rev. Lett. 99, 127003 (2007).
  8. [8]V. J. Emery, S. A. Kivelson, O. Zachar, Spin-gap proximity effect mechanism of high-temperature superconductivity. Phys. Rev. B 56, 6120–6147 (1997).DOI
  9. [9]S. A. Kivelson, E. Fradkin, V. J. Emery, Electronic liquid-crystal phases of a doped Mott insulator. Nature 393, 550–553 (1998).DOI
  10. [10]A. Chen, J. Gray, S.-J. Du, G. K.-L. Chan, Extreme competition between stripes and superconductivity in the hole-doped Hubbard model. arXiv:2610.30321v2 (2026).arxiv.org/abs/2610.30321
  11. [11]P. W. Anderson, The resonating valence bond state in La₂CuO₄ and superconductivity. Science 235, 1196–1198 (1987).pubmed.ncbi.nlm.nih.gov/17818979
  12. [12]F. C. Zhang, T. M. Rice, Effective Hamiltonian for the superconducting Cu oxides. Phys. Rev. B 37, 3759–3761 (1988).pubmed.ncbi.nlm.nih.gov/9944993
  13. [13]D. J. Scalapino, A common thread: The pairing interaction for unconventional superconductors. Rev. Mod. Phys. 84, 1383–1417 (2012).DOI
  14. [14]D. P. Arovas, E. Berg, S. A. Kivelson, S. Raghu, The Hubbard model. Annu. Rev. Condens. Matter Phys. 13, 239–274 (2022).
  15. [15]M. Qin, T. Schäfer, S. Andergassen, P. Corboz, E. Gull, The Hubbard model: A computational perspective. Annu. Rev. Condens. Matter Phys. 13, 275–302 (2022).DOI
  16. [16]J. P. F. LeBlanc et al. (Simons Collaboration on the Many-Electron Problem), Solutions of the two-dimensional Hubbard model: Benchmarks and results from a wide range of numerical algorithms. Phys. Rev. X 5, 041041 (2015).
  17. [17]B.-X. Zheng et al., Stripe order in the underdoped region of the two-dimensional Hubbard model. Science 358, 1155–1160 (2017).DOI
  18. [18]M. Qin, C.-M. Chung, H. Shi, E. Vitali, C. Hubig, U. Schollwöck, S. R. White, S. Zhang, Absence of superconductivity in the pure two-dimensional Hubbard model. Phys. Rev. X 10, 031016 (2020).DOI
  19. [19]S. Sorella, Systematically improvable mean-field variational ansatz for strongly correlated systems: Application to the Hubbard model. Phys. Rev. B 107, 115133 (2023).DOI
  20. [20]O. K. Andersen, A. I. Liechtenstein, O. Jepsen, F. Paulsen, LDA energy bands, low-energy Hamiltonians, t′, t″, t⊥(k), and J⊥. J. Phys. Chem. Solids 56, 1573–1591 (1995).DOI
  21. [21]E. Pavarini, I. Dasgupta, T. Saha-Dasgupta, O. Jepsen, O. K. Andersen, Band-structure trend in hole-doped cuprates and correlation with T_c^max. Phys. Rev. Lett. 87, 047003 (2001).pubmed.ncbi.nlm.nih.gov/11461638
  22. [22]M. Hirayama, Y. Yamaji, T. Misawa, M. Imada, Ab initio effective Hamiltonians for cuprate superconductors. Phys. Rev. B 98, 134501 (2018).DOI
  23. [23]B. Ponsioen, S. S. Chung, P. Corboz, Period 4 stripe in the extended two-dimensional Hubbard model. Phys. Rev. B 100, 195141 (2019).DOI
  24. [24]H.-C. Jiang, T. P. Devereaux, Superconductivity in the doped Hubbard model and its interplay with next-nearest hopping t′. Science 365, 1424–1428 (2019).DOI
  25. [25]H.-C. Jiang, T. P. Devereaux, S. A. Kivelson, Competition between charge-density-wave and superconducting orders on eight-leg square Hubbard cylinders. npj Quantum Mater. 11, 80 (2026).DOI
  26. [26]H.-C. Jiang, S. A. Kivelson, Stripe order enhanced superconductivity in the Hubbard model. Proc. Natl. Acad. Sci. U.S.A. 119, e2109406119 (2022).DOI
  27. [27]E. W. Huang, C. B. Mendl, S. Liu, S. Johnston, H.-C. Jiang, B. Moritz, T. P. Devereaux, Numerical evidence of fluctuating stripes in the normal state of high-Tc cuprate superconductors. Science 358, 1161–1164 (2017).DOI
  28. [28]P. Mai, S. Karakuzu, G. Balduzzi, S. Johnston, T. A. Maier, Intertwined spin, charge and pair correlations in the two-dimensional Hubbard model in the thermodynamic limit. Proc. Natl. Acad. Sci. U.S.A. 119, e2112806119 (2022).DOI
  29. [29]D. Bourgund et al., Formation of individual stripes in a mixed-dimensional cold-atom Fermi–Hubbard system. Nature 637, 57–62 (2025).DOI
  30. [30]L. H. Kendrick et al., Pseudogap in a Fermi–Hubbard quantum simulator. Nature 656, 602–608 (2026).DOI
  31. [31]H. Xu, C.-M. Chung, M. Qin, U. Schollwöck, S. R. White, S. Zhang, Coexistence of superconductivity with partially filled stripes in the Hubbard model. Science 384, eadh7691 (2024).DOI
  32. [32]G. Carleo, M. Troyer, Solving the quantum many-body problem with artificial neural networks. Science 355, 602–606 (2017).DOI
  33. [33]D. Luo, B. K. Clark, Backflow transformations via neural networks for quantum many-body wave functions. Phys. Rev. Lett. 122, 226401 (2019).DOI
  34. [34]D. Wu et al., Variational benchmarks for quantum many-body problems. Science 386, 296–301 (2024).DOI
  35. [35]R. Rende, L. L. Viteritti, A. Georges, Superconductivity in the t-t′ Hubbard model from symmetry-preserving neural-network quantum states. arXiv:2608.12465 (2026).arxiv.org/abs/2608.12465
  36. [36]C. Roth, A. Chen, A. Sengupta, A. Georges, Superconductivity in the two-dimensional Hubbard model revealed by neural quantum states. arXiv:2511.07566 (2025).arxiv.org/abs/2511.07566
  37. [37]A. Chen, Z.-Q. Wan, A. Sengupta, A. Georges, C. Roth, Neural network-augmented Pfaffian wave-functions for scalable simulations of interacting fermions. Proc. Natl. Acad. Sci. U.S.A. 123, e2535288123 (2026).DOI
  38. [38]L. L. Viteritti, R. Rende, C. Roth, A. Sengupta, G. Carleo, A. Georges, Beyond variational bias: Resolving intertwined orders in the Hubbard model. arXiv:2604.21978 (2026).arxiv.org/abs/2604.21978
  39. [39]K. Loehr, B. K. Clark, Enhancing neural network backflow. arXiv:2510.26906 (2025).arxiv.org/abs/2510.26906
  40. [40]Y. Gu et al., Solving the Hubbard model with neural quantum states. Nat. Commun. 17, 7838 (2026).
  41. [41]Y. Gu, Z. Han, W. Li, Z. Xiao, T. Xiang, M. Qin, L. Wang, D. Lv, Pareto frontier of neural quantum states: Scalable, affordable, and accurate convolutional backflow for strongly correlated lattice fermions. PRX Intelligence 1, 023001 (2026).arxiv.org/abs/2604.25775
  42. [42]C. A. Jiménez-Hoyos, R. Rodríguez-Guzmán, G. E. Scuseria, Multi-component symmetry-projected approach for molecular ground state correlations. J. Chem. Phys. 139, 204102 (2013).
  43. [43]S. Zhang, J. Carlson, J. E. Gubernatis, Pairing correlations in the two-dimensional Hubbard model. Phys. Rev. Lett. 78, 4486–4489 (1997).DOI
  44. [44]C. N. Yang, Concept of off-diagonal long-range order and the quantum phases of liquid He and of superconductors. Rev. Mod. Phys. 34, 694–704 (1962).DOI
  45. [45]W. Kohn, Theory of the insulating state. Phys. Rev. 133, A171–A181 (1964).DOI
  46. [46]D. J. Scalapino, S. R. White, S. Zhang, Insulator, metal, or superconductor: The criteria. Phys. Rev. B 47, 7995–8007 (1993).pubmed.ncbi.nlm.nih.gov/10004809

Paper details

vlnc:W_qahr3wkg
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 10

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Physics and Astronomy

Not recorded for this paper

View statement
Looking for related papers…
Contents