Optimal error bounds on the exponential wave integrator for the nonlinear Schrödinger equation with low regularity potential and nonlinearity

Weizhu Bao,Chushan Wang

DOI: https://doi.org/10.1137/23M155414X

2023-08-26

Abstract:We establish optimal error bounds for the exponential wave integrator (EWI) applied to the nonlinear Schrödinger equation (NLSE) with $ L^\infty $-potential and/or locally Lipschitz nonlinearity under the assumption of $ H^2 $-solution of the NLSE. For the semi-discretization in time by the first-order Gautschi-type EWI, we prove an optimal $ L^2 $-error bound at $ O(\tau) $ with $ \tau>0 $ being the time step size, together with a uniform $ H^2 $-bound of the numerical solution. For the full-discretization scheme obtained by using the Fourier spectral method in space, we prove an optimal $ L^2 $-error bound at $ O(\tau + h^2) $ without any coupling condition between $ \tau $ and $ h $, where $ h>0 $ is the mesh size. In addition, for $ W^{1, 4} $-potential and a little stronger regularity of the nonlinearity, under the assumption of $ H^3 $-solution, we obtain an optimal $ H^1 $-error bound. Furthermore, when the potential is of low regularity but the nonlinearity is sufficiently smooth, we propose an extended Fourier pseudospectral method which has the same error bound as the Fourier spectral method while its computational cost is similar to the standard Fourier pseudospectral method. Our new error bounds greatly improve the existing results for the NLSE with low regularity potential and/or nonlinearity. Extensive numerical results are reported to confirm our error estimates and to demonstrate that they are sharp.

Numerical Analysis

What problem does this paper attempt to address?

The problem that this paper attempts to solve is to establish optimal error bounds for the nonlinear Schrödinger equation (NLSE) with low - regularity potentials and/or nonlinearities. Specifically, the paper focuses on how to provide theoretical error estimates for the exponential wave integrator (EWI) method when the potential function \( V(x)\in L^{\infty}(\Omega) \) and the nonlinear term \( f(|\psi|^{2}) \) are locally Lipschitz continuous.

### Problem Background

The nonlinear Schrödinger equation (NLSE) has wide applications in quantum mechanics, nonlinear optics, and Bose - Einstein condensation, etc. For NLSE with smooth potentials and nonlinearities, many efficient numerical methods have been proposed and analyzed. However, when the regularity of the potential and the nonlinear term is low, existing numerical methods may encounter difficulties, especially in error estimation. This is mainly because low regularity will lead to a reduction in the order of the local truncation error and it is difficult to obtain stability estimates in high - order Sobolev spaces.

### Main Contributions of the Paper

1. **Time Semi - discretization**:
- For the first - order Gautschi - type EWI method, the paper proves that under the \( H^{2}\)-solution assumption, the optimal \( L^{2}\)-error bound \( O(\tau) \) can be obtained, and the numerical solution has uniform boundedness in the \( H^{2}\) space.

2. **Full - discretization**:
- After using the Fourier spectral method for spatial discretization, the paper proves that without any coupling conditions, the optimal \( L^{2}\)-error bound \( O(\tau + h^{2}) \) can be obtained.

3. **More General Regularity Conditions**:
- When the potential function belongs to \( W^{1,4} \) and the nonlinear term is slightly stronger, under the \( H^{3}\)-solution assumption, the optimal \( H^{1}\)-error bound can be obtained.

4. **Extended Fourier Pseudo - spectral Method**:
- When the potential function has low regularity but the nonlinear term is smooth enough, the paper proposes an extended Fourier pseudo - spectral method. This method has a computational cost similar to that of the standard Fourier pseudo - spectral method and can also obtain optimal error bounds.

### Improvements

Compared with the classical exponential wave integrator and time - splitting methods in existing literature, the results of this paper significantly reduce the regularity requirements for the potential function and the nonlinear term. Specifically:

- For the potential function, the differentiability requirement is reduced by two orders;
- For the nonlinear term, the differentiability requirement is reduced by one order.

In addition, this paper also improves the convergence order of the Crank - Nicolson Galerkin method in the \( L^{2}\)-norm and removes the coupling condition between the time step \( \tau \) and the mesh size \( h \).

Through these improvements, the paper provides a more accurate and effective numerical solution for NLSE with low - regularity potentials and nonlinearities.