
Hydrogenated rare-earth nickelates exhibit proton redistribution, lattice response, and polarization relaxation. We systematically integrate established reduced descriptions for H_x –NdNiO _3 : a one-dimensional structure-conditioned proton-transport model with local-detailed-balance rates and a variational two-mode free energy, a separately fitted phenomenological distributed-relaxation model, and a recoverability diagnostic for visible-state sufficiency. The effective Hessian has K_A=10.3784 eV A^-2 and K_R=36.6667 eV A^-2. At 300 K, the fitted dimensionless relaxation-spectrum slope is α _A=6.5289, with 95% profile-likelihood interval [5.8557, 7.4265]. Conditional on E_A=0.1740 eV, this maps to χ _A^cond=0.9698; using this coefficient in transport is a parameter-transfer hypothesis. On the complete 87-point holdout at 533 kV cm^-1, the relaxation model achieves root mean square error 0.02634 and coefficient of determination 0.99081, reducing error by 47.3% relative to the field-law stretched exponential. This comparison tests polarization prediction, while residual autocorrelation and unequal model flexibility limit mechanistic interpretation. A separate constant-field, spatially constant-diffusivity transport benchmark exhibits convergence order 2.029, a weighted spectral-gap bound, and relative-entropy dissipation. Recoverability diagnoses information discarded by the visible-state choice, with zero coupling to force or hopping rate. The contribution is a literature-constrained modular framework with task-specific evaluation; a dynamically calibrated concentration-to-polarization map remains to be established. Evidence synthesis for Hx-NdNiO3. External holdout, structural calibration, relaxation-spectrum identifiability, conditional transport sensitivity, matched forward continuum, recoverability transfer, and prospective measurement design.
We investigate the nonlinear dynamics of modulational instability in a quasi-one-dimensional Bose gas with effective-range interaction corrections. Based on the modified Gross–Pitaevskii equation that includes a nonlinear spatial-dispersion term, we obtain the MI gain spectrum through linear stability analysis. Furthermore, employing a three-mode truncation method, we establish a reduced canonical Hamiltonian model for the carrier and the first pair of sidebands and analyze the full modulational instability dynamics of weak periodic perturbations superimposed on a plane-wave background in combination with real-time numerical evolution. The results demonstrate that in the attractive short-range interaction regime, two distinct forms of instability occur: long-wavelength and short-wavelength instability. In the long-wavelength instability regime, regular homoclinic/heteroclinic Fermi–Pasta–Ulam–Tsingou recurrence and single-cycle growth-decay dynamics of the Akhmediev breather type can be sustained. Conversely, all short-wavelength instabilities lead to the destabilization of the plane wave and the emergence of complex dynamics. Notably, destabilization also occurs in the long-wavelength unstable region in the vicinity of the short-wave instability boundary. Under the repulsive short-range interaction regime, the system exhibits either modulational stability or short-wavelength instability, and such short-wave instability likewise triggers destabilization of the plane-wave background. These findings reveal the critical role of effective-range corrections in the modulational instability dynamics and structural destabilization of Bose–Einstein condensates.
Cuprate and nickelate superconductors occupy neighboring but non-equivalent regions of transition-metal-oxide pairing space. The decisive distinction is not simply planar versus interlayer pairing, but whether charge-transfer processes permit low-energy access to the vertical-orbital manifold. In infinite-layer nickelates, suppression of the vertical sector leaves a predominantly planar correlated state with cuprate-like magnetic and nodal characteristics, whereas bilayer geometry and pressure restore vertical hopping, bonding–antibonding splitting, and interlayer exchange. In cuprates, the reverse evolution is impeded because added charge is preferentially redistributed over ligand states and the Cu valence remains pinned, preventing formation of the two-orbital configuration required for nickelate-like interlayer pairing. This charge-transfer constraint links the planar and bilayer nickelate limits while explaining why the apparently inverse route is electronically inaccessible in cuprates. The cuprate–nickelate correspondence is therefore directional and is controlled by orbital accessibility set by charge-transfer energetics rather than by lattice geometry alone.
Attention mechanisms provide an effective means of enhancing the capability of physics-informed neural networks to learn complex non-linear structures by dynamically adjusting the contributions of different samples, regions or weights during training. In this work, the residual-based attention physics-informed neural network (RBA-PINN) is applied for the first time to the study data-driven localized wave solutions of the integrable Chen-Lee-Liu equation, and its data-driven solving capability is systematically investigated for a variety of non-linear localized wave structures, including bright soliton, dark soliton, breather, periodic wave and rogue periodic wave. By adaptively adjusting the local weights according to the evolution of the physical residuals at the collocation points, the RBA-PINN enables the network to place greater emphasis on high-residual and difficult-to-learn regions during training. Numerical results demonstrate that the RBA-PINN can accurately reconstruct the spatiotemporal evolution of different wave structures and exhibits favorable convergence and predictive accuracy in terms of the loss function, relative L_2 norm error and L_∞ error. In particular, for the rogue periodic wave example, the final loss of the RBA-PINN is reduced by more than two orders of magnitude compared with that of the standard PINN, while both the relative L_2 norm error and the L_∞ error are improved by approximately one order of magnitude. These results indicate that the RBA-PINN has clear advantages in resolving high-gradient, strongly non-linearity, and complex oscillatory dynamics, providing a high-accuracy and adaptive computational approach for the complex data-driven non-linear localized waves in integrable systems.
Physics-informed neural networks (PINNs) have achieved remarkable success in solving continuous partial differential equations (PDEs), yet their extension to discrete soliton equations—particularly coupled differential-difference systems—remains considerably less mature and faces substantial optimization challenges, especially when the residuals of different governing equations suffer from persistent magnitude imbalances. To address this issue, we propose the improved PINNs with interaction-term loss (ITL–PINNs) framework for a higher-order nonlinear self-dual equation. The proposed interaction-term loss explicitly couples the residual norms of the current and voltage equations and penalizes their magnitude discrepancy, thereby promoting a more balanced optimization process. This formulation is further combined with symmetric-difference enhancement, in which learnable symmetry-center parameters are introduced to incorporate spatial and temporal symmetry information into the network. The resulting ITL–PINNs with symmetric-difference enhancement (ITL–SDE–PINNs) framework is evaluated for forward prediction, temporal extrapolation, parameter identification, and physics-constrained denoising. For one- and two-soliton solutions, the proposed method achieves lower prediction errors and reduced run-to-run variability than standard PINNs and the corresponding single-component variants over the tested domains. In coefficient-identification problems, ITL–SDE–PINNs generally improves estimation accuracy under sparse and noisy observations, although isolated optimization failures remain in several challenging settings. The residual-coupled formulation is also applied to physics-constrained signal reconstruction, where it yields higher signal-to-noise ratios than a fourth-order Butterworth filter for the synthetic noisy signals considered. These results indicate that residual coupling and symmetry enhancement provide complementary mechanisms for improving physics-informed learning of coupled discrete nonlinear systems.An improved PINN’s framework augmented with symmetric difference enhancement and an interaction-term loss is developed for forward and inverse problems as well as signal denoising of a higher-order nonlinear self-dual network equation. Numerical results reveal that the proposed combined approach not only reproduces exact discrete soliton solutions with high accuracy and robustness, but also achieves more accurate parameter discovery in inverse problems and outperforms conventional methods in signal denoising.