Hybrid quantum-classical algorithms for complex nonlinear partial differential equations with Ginzburg-Landau potential and vortex motion laws

Abstract: We propose quantum algorithms for complex-valued nonlinear partial differential equations in the strongly nonlinear regime, where the dynamics is governed by vortex cores, phase singularities, and nonlinear vortex interactions. Examples include the complex-valued nonlinear Schrödinger equation, as well as nonlinear heat and wave equations with Ginzburg--Landau-type nonlinearity. In the strongly nonlinear regime, the solutions to these equations are asymptotically governed by, in leading order, linear elliptic equations, coupled with low-dimensional vortex dynamics, where the vortex cores correspond to topological defects in superconductors. Our hybrid quantum-classical algorithms utilize this asymptotic property, in which the vortex dynamic is advanced classically while the boundary-value problem of linear elliptic equation is handled by quantum algorithms. For the two-dimensional nonlinear Schrödinger equation, we also combine quantum BPX preconditioning with Schrödingerization to estimate physically relevant observables in the small-output regime. This yields, already in two dimensions, an {\it exponential} improvement in the dependence on the spatial problem size, while the dependence on the target accuracy remains essentially linear up to polylogarithmic factors. We further show that the same principle extends to dissipative Ginzburg--Landau vortex dynamics and to vortex filaments in three-dimensional superconductivity. Numerical results support the validity of this PDE reduction and the effectiveness of the proposed approach.

• The paper develops a hybrid method that evolves low-dimensional vortex locations classically while using a quantum BPX-preconditioned Schrödingerization solver for harmonic or London-field corrections.
• The framework estimates only local gradients or other linear observables, achieving conditional per-observable complexity of nearly O(ε⁻¹) up to polylogarithmic factors instead of classical spatial scaling O(ε⁻ᵈ log(1/ε)).
• Numerical tests show approximately O(ε) reconstruction errors, with the boundary-aware vortex model M2 outperforming the free-space model M1 and delivering about half the final masked error at ε = 0.025.

Overview and motivation

This paper by Jin, Liu, and Ma develops hybrid quantum-classical algorithms for complex-valued nonlinear PDEs in a strongly nonlinear regime, where the dynamics is dominated by vortex cores, phase singularities, and nonlinear vortex interactions (2604.14079). The model problems are the two-dimensional nonlinear Schrödinger equation (NLS) with Ginzburg–Landau potential, dissipative nonlinear heat (Ginzburg–Landau) equations, and the three-dimensional time-dependent Ginzburg–Landau system for superconductivity. The central observation is that, rather than attempting to linearize or embed the full nonlinear PDE—a strategy that generally fails outside perturbative regimes—the authors exploit rigorous asymptotic reductions showing that in the small-core limit $\varepsilon \to 0$ the solution decomposes into a low-dimensional classical evolution of vortex locations coupled to a high-dimensional but linear boundary-value problem for a harmonic correction field. The quantum computer is used only for this linear subproblem, while the genuinely nonlinear content is carried by an $O(M)$-dimensional ODE system advanced classically.

This design contrasts sharply with existing approaches to quantum algorithms for nonlinear PDEs: exact linearizing transformations (Cole–Hopf, level sets), Koopman–von Neumann lifting after spatial discretization, Carleman-type embeddings, and hydrodynamic reformulations all either apply only to special equations, weak nonlinearities, or short times, or lose quantum advantage because the discretized nonlinear system's dimension scales with resolution.

Asymptotic reduction to vortices plus a linear correction

The starting point is the NLS equation $$i\partial_t u^\varepsilon = -\Delta u^\varepsilon + \varepsilon^{-2}(1-|u^\varepsilon|^2)u^\varepsilon$$ on a bounded simply connected domain $\Omega \subset \mathbb{R}^2$, with boundary data of degree $M > 0$. Under assumptions (H1)–(H3)—same-sign unit vortices and energy bounded by $M\pi\log(1/\varepsilon) + C_0$—the solution converges weakly to

$$u_{\bm a}(\bm x) = e^{i h_{\bm a}(\bm x)} \prod_{j=1}^M \frac{\bm x - \bm a_j}{|\bm x - \bm a_j|},$$

where $\Theta_{\bm a} = \sum_j \arg(\bm x - \bm a_j)$ carries the topological singularities and $h_{\bm a}$ is a harmonic correction determined by the Dirichlet problem $\Delta h_{\bm a} = 0$ in $O(M)$0, $O(M)$1 on $O(M)$2. This follows from the weak convergence of the linear momentum to an incompressible Euler velocity field $O(M)$3 satisfying the tangential boundary condition derived from the data.

The reduced vortex motion is governed by the renormalized energy $O(M)$4 via the Hamiltonian law $O(M)$5. In the same-sign unit-vortex regime this becomes the explicit model M2,

$O(M)$6

combining pairwise Kirchhoff interaction with a boundary-induced drift through $O(M)$7 at the vortex sites. When boundary effects are negligible over the time horizon, the free-space approximation M1 drops the drift term entirely; the authors note that M1 is only an auxiliary approximation, since the rigorous bounded-domain law is M2. A structural point that matters computationally: even in the fully coupled model, the classical update requires only the local gradients $O(M)$8, not the full field.

Quantum treatment of the harmonic correction via Schrödingerization

The linear subproblem is solved on a quantum device using Schrödingerization combined with BPX multilevel preconditioning. After discretization, the harmonic correction yields $O(M)$9, where the discrete Laplacian $$i\partial_t u^\varepsilon = -\Delta u^\varepsilon + \varepsilon^{-2}(1-|u^\varepsilon|^2)u^\varepsilon$$0 is independent of the vortex configuration—all time dependence enters through the right-hand side. With the BPX factorization $$i\partial_t u^\varepsilon = -\Delta u^\varepsilon + \varepsilon^{-2}(1-|u^\varepsilon|^2)u^\varepsilon$$1, the symmetrically preconditioned operator $$i\partial_t u^\varepsilon = -\Delta u^\varepsilon + \varepsilon^{-2}(1-|u^\varepsilon|^2)u^\varepsilon$$2 has mesh-uniform spectral bounds and admits a structure-aware block-encoding with normalization $$i\partial_t u^\varepsilon = -\Delta u^\varepsilon + \varepsilon^{-2}(1-|u^\varepsilon|^2)u^\varepsilon$$3. The relaxation dynamics toward the steady state is embedded into an augmented linear ODE, lifted via the warped phase transform to a Schrödinger-type equation in one extra dimension, and implemented coherently by Hamiltonian simulation, QFTs in the auxiliary register, postselected recovery of the steady-state block, and overlap estimation against a measurement state prepared from $$i\partial_t u^\varepsilon = -\Delta u^\varepsilon + \varepsilon^{-2}(1-|u^\varepsilon|^2)u^\varepsilon$$4.

Crucially, the algorithm never reconstructs the full vector $$i\partial_t u^\varepsilon = -\Delta u^\varepsilon + \varepsilon^{-2}(1-|u^\varepsilon|^2)u^\varepsilon$$5; it estimates only linear observables $$i\partial_t u^\varepsilon = -\Delta u^\varepsilon + \varepsilon^{-2}(1-|u^\varepsilon|^2)u^\varepsilon$$6 via Hadamard tests or amplitude estimation. Under the stated spectral and state-preparation assumptions, the query complexity is $$i\partial_t u^\varepsilon = -\Delta u^\varepsilon + \varepsilon^{-2}(1-|u^\varepsilon|^2)u^\varepsilon$$7 per observable. These assumptions are nontrivial: they require mesh-uniform conditioning of $$i\partial_t u^\varepsilon = -\Delta u^\varepsilon + \varepsilon^{-2}(1-|u^\varepsilon|^2)u^\varepsilon$$8 and efficient oracle preparation of both input and measurement states, so the complexity claim is conditional on those primitives being available.

Two operating modes are formulated: decoupled mode M1, where the vortex trajectory is computed entirely classically and the quantum routine is invoked only at output times, and coupled mode M2, where quantum-estimated feedback values $$i\partial_t u^\varepsilon = -\Delta u^\varepsilon + \varepsilon^{-2}(1-|u^\varepsilon|^2)u^\varepsilon$$9 enter each classical update step. In both modes the classical–quantum communication remains limited to a few scalar observables, preserving the small-output character of the framework.

Extensions to Ginzburg–Landau dynamics and three-dimensional filaments

For the logarithmically rescaled nonlinear heat equation, the same outer ansatz holds, but the vortex law changes from Hamiltonian to gradient flow, $\Omega \subset \mathbb{R}^2$0, which in bounded domains takes the form $\Omega \subset \mathbb{R}^2$1. The harmonic correction enters the same elliptic problem as before, so the identical BPX–Schrödingerization machinery applies; the authors note this is a quasi-static description on the logarithmically accelerated vortex-motion scale.

In three dimensions, the matched-asymptotic analysis of the gauge-fixed time-dependent Ginzburg–Landau superconductivity system (London gauge) shows that vortex filaments move by curvature flow $\Omega \subset \mathbb{R}^2$2, while the leading outer magnetic field satisfies the compact London equation

$\Omega \subset \mathbb{R}^2$3

Discretizing gives $\Omega \subset \mathbb{R}^2$4, and a lemma proved via the discrete Poincaré inequality establishes that the shifted BPX-preconditioned operator $\Omega \subset \mathbb{R}^2$5 retains mesh-uniform spectral bounds whenever $\Omega \subset \mathbb{R}^2$6 does. Hence the three-dimensional London problem falls into exactly the same framework. The authors are careful to state that the quantum target here is not the full gauge-dependent triple $\Omega \subset \mathbb{R}^2$7 but the filament geometry and the leading outer magnetic field $\Omega \subset \mathbb{R}^2$8; the electric potential is not singled out as a linear observable in the reduction.

Complexity results

The main quantitative claim is a unified comparison for one time step of the linear correction stage. Classically, a BPX-preconditioned iterative solver costs $\Omega \subset \mathbb{R}^2$9 operations per right-hand side. On a uniform grid with first-order accuracy, matching discretization error to tolerance gives $M > 0$0, so the classical cost becomes $M > 0$1, whereas the quantum cost per observable is $M > 0$2. The paper states this as an exponential improvement in the dependence on the spatial problem size already in two dimensions, with essentially linear dependence on target accuracy up to polylogarithmic factors. Two qualifications are stated explicitly: the advantage applies only when a fixed number of observables is needed—not to full-field reconstruction—and it rests on the availability of the block-encoding and state-preparation oracles with the assumed costs.

Numerical verification

The experiments validate the PDE reduction itself rather than the quantum solver. On the unit square with degree-2 boundary data and two initial unit vortices, the full NLS is evolved by Strang splitting and compared against outer reconstructions built from M1 and M2 trajectories, with errors measured on a masked region excluding disks of radius $M > 0$3 around the reference vortex positions and aligned by a global constant phase. For $M > 0$4, the final-time masked relative $M > 0$5 errors decrease at roughly $M > 0$6 for both models, consistent with the asymptotic theory. Notably, M2 consistently outperforms M1, and at $M > 0$7 its final masked error is approximately half that of M1, demonstrating that the boundary-induced $M > 0$8-drift provides a quantitatively meaningful improvement in the small-core regime—even though at the trajectory level the two models remain close over the short horizon $M > 0$9. Incorporating the feedback does not alter the hybrid structure, since the quantum component still returns only local gradient values.

Limitations and open questions

Several restrictions are acknowledged directly. The asymptotic reduction assumes the almost-minimizing regime, in which the defect measure vanishes and vortex motion is fully determined by the renormalized energy; mixed-sign and higher-degree vortices are excluded from the exposition for simplicity, though the underlying theory permits them. The free-space model M1 neglects radiation and vortex–sound interactions, effects known from prior numerical studies to cause deviations of reduced laws from full simulations. The exponential-in-$M\pi\log(1/\varepsilon) + C_0$0 advantage is confined to the small-output regime and does not extend to reconstructing the full outer field; moreover, the complexity bounds presuppose oracle access to input states, measurement states, and a low-normalization block-encoding of the preconditioned operator, none of which is demonstrated end-to-end on hardware. The numerical validation covers a single configuration, two vortices, and a short time horizon, leaving open how the reduction performs under stronger boundary feedback, longer times, or many-vortex regimes. Whether the framework extends to broader classes of nonlinear PDEs beyond Ginzburg–Landau-type complex scalar fields remains an open question posed by the authors.

Conclusion

The paper establishes a principled nonlinear-to-linear decomposition for strongly nonlinear complex scalar field equations: rigorous asymptotics split the dynamics into a low-dimensional classical vortex system and a high-dimensional linear elliptic (or London-type) subproblem amenable to BPX-preconditioned Schrödingerization. In the small-output regime this yields an exponential improvement in the scaling with spatial degrees of freedom, at essentially linear cost in accuracy up to polylogs, and the mechanism transfers across Hamiltonian NLS vortices, dissipative Ginzburg–Landau gradient flows, and three-dimensional curvature-driven filaments. Numerical evidence confirms the $M\pi\log(1/\varepsilon) + C_0$1 accuracy of the outer reconstruction and the value of the boundary-induced drift term, supporting the proposed hybrid framework as a concrete route to quantum algorithms for a class of genuinely nonlinear PDEs.