Eddy de León and Caroline Lasser from Technical University of Munich have analyzed a technique to solve time-dependent Schrödinger equations that achieves first- and second-order convergence rates under certain assumptions, offering potential improvements over existing methods. Their work centers on residual minimization, a technique for approximating solutions within low-dimensional spaces, and analyzes two distinct formulations for time-stepping: one that discretizes then parametrizes the equation, and another that discretizes the Dirac, Frenkel variational principle. The researchers demonstrate a unified error analysis for both approaches, yielding bounds that separate the effects of time discretization and residual minimization. Importantly, when applied to Gaussian approximation manifolds with polynomial operators, this framework allows for explicit closed-form expressions of residual norms and gradients, enabling efficient implementation without spatial discretization.

For the variational formulation, the analysis reveals that stability is linked to the conditioning of the parametrization map, providing insight into optimal parameter selection. The framework is applied to Gaussian approximation manifolds, for which residual norms and gradients admit explicit closed-form expressions when polynomial operators are involved. This enables efficient implementation without spatial discretization. Numerical experiments for time-dependent Schrödinger equations illustrate the theoretical convergence rates and the influence of residual accuracy on conservation properties.

The pursuit of efficient numerical solutions for complex, time-dependent equations has led researchers to explore methods that move beyond traditional grid-based approaches. A recent analysis focuses on a class of techniques employing parametrized approximations, representing solutions within low-dimensional nonlinear manifolds and advancing them in time by minimizing residuals, the difference between the approximate and actual solution. This work establishes a unified framework for analyzing the errors inherent in two distinct approaches to this time-stepping process. The core of the investigation lies in understanding how errors accumulate from both the time discretization itself and the process of minimizing the residual within the chosen manifold. Researchers have demonstrated that the global error can be separated into these two contributions, yielding explicit convergence rates. Specifically, the analysis reveals that both first- and second-order convergence rates are achievable under conditions of Lipschitz continuity, one-sided Lipschitz continuity, and dissipativity of the underlying system.

Nonlinear Manifold Approximation of Parametric Solutions

Eddy de León and Caroline Lasser are analyzing techniques for modeling complex systems by focusing on how to accurately simulate their evolution over time, a challenge particularly acute in fields like quantum dynamics and wave propagation. Their work, detailed in a recent paper, centers on representing solutions not as detailed grids, but as parameters evolving on a lower-dimensional space capturing essential features. This approach sidesteps the computational bottlenecks of traditional methods as dimensionality increases. The researchers are analyzing two distinct strategies for implementing this: one that discretizes the equation before parameterizing the solution, and another that discretizes the underlying variational principle governing the parameter dynamics. A key advancement is a unified error analysis applicable to both approaches within the family of -methods, allowing for a clearer understanding of how errors accumulate from both time-stepping and the manifold approximation itself.

The framework yields first- and second-order convergence rates under certain assumptions. The framework proves particularly efficient when applied to Gaussian approximation manifolds, for which residual norms and gradients admit explicit closed-form expressions when polynomial operators are involved. This analytical tractability, as the authors state, “enables efficient implementation without spatial discretization,” a significant computational advantage. The researchers emphasize that while their initial tests used one-dimensional Schrödinger equations, encompassing harmonic, double-well, and hyperbolic cosine potentials as examples, the framework is broadly applicable to a wider range of evolution equations, offering a versatile tool for modeling complex phenomena.

While many numerical methods for complex equations rely on increasingly fine spatial grids, a different approach focuses on representing solutions within constrained, low-dimensional spaces. Recent work by Eddy de León and Caroline Lasser establishes a rigorous framework for analyzing the accuracy of time-stepping methods that operate on these, offering a path toward simulations that avoid computationally expensive spatial discretization. Their analysis centers on two distinct strategies for building these time-stepping schemes, each with its own strengths and weaknesses. A key focus is understanding how errors accumulate when minimizing “residuals”, measures of how well the approximate solution satisfies the original equation, at each time step. The resulting bounds reveal both first- and second-order convergence rates under certain assumptions.

This dual analysis provides a robust framework for assessing the accuracy and stability of different time-stepping strategies within the family of -methods. The authors demonstrate that these closed-form expressions are not merely theoretical curiosities; they facilitate efficient computation, concentrating computational effort on low-dimensional optimization problems. Numerical experiments focusing on time-dependent Schrödinger equations, including harmonic, double-well, and hyperbolic cosine potentials as examples, confirm the predicted convergence rates and highlight the direct correlation between residual accuracy and the preservation of key conservation properties, such as energy. The framework is applied to Gaussian approximation manifolds, for which residual norms and gradients admit explicit closed-form expressions when polynomial operators are involved.

Their findings, detailed in a recent publication, demonstrate that first- and second-order convergence rates are achievable for time-dependent Schrödinger equations under specific conditions, namely, Lipschitz continuity, one-sided Lipschitz behavior, and dissipativity assumptions. A key aspect of their framework is the consideration of two distinct residual formulations. This analytical tractability stems from the specific properties of Gaussian manifolds when polynomial operators are involved, allowing for streamlined calculations of crucial quantities. The unified error analysis developed separates the effects of time discretization from those of residual minimization, providing a clearer understanding of the factors influencing accuracy.

The team analyzed two distinct residual formulations for time-stepping. This duality offers researchers flexibility in computational strategy, allowing them to select the method best suited to a given problem. Numerical experiments with one-dimensional Schrödinger equations, including harmonic, double-well, and hyperbolic cosine potentials as examples, validated the theoretical convergence rates and highlighted the direct link between residual accuracy and the conservation of physical properties.

👉 More information
🗞 Residual-Based Time Discretization on Nonlinear Approximation Manifolds: Analysis and Gaussian Applications
✍️ Eddy de Leon and Caroline Lasser
🧠 ArXiv: https://arxiv.org/abs/2607.15086

More like this