- Research
- Open access
- Published:
A second-order low-regularity integrator for the nonlinear Schrödinger equation
Advances in Continuous and Discrete Models volume 2022, Article number: 23 (2022)
Abstract
In this paper, we analyze a new exponential-type integrator for the nonlinear cubic Schrödinger equation on the d dimensional torus \(\mathbb{T}^{d}\). The scheme has also been derived recently in a wider context of decorated trees (Bruned et al. in Forum Math. Pi 10:1–76, 2022). It is explicit and efficient to implement. Here, we present an alternative derivation and give a rigorous error analysis. In particular, we prove the second-order convergence in \(H^{\gamma }(\mathbb{T}^{d})\) for initial data in \(H^{\gamma +2}(\mathbb{T}^{d})\) for any \(\gamma > d/2\). This improves the previous work (Knöller et al. in SIAM J. Numer. Anal. 57:1967–1986, 2019).
The design of the scheme is based on a new method to approximate the nonlinear frequency interaction. This allows us to deal with the complex resonance structure in arbitrary dimensions. Numerical experiments that are in line with the theoretical result complement this work.
1 Introduction
The nonlinear Schrödinger equation (NLS) arises as a model equation in several areas of physics see, e.g., Sulem and Sulem [20]. In this paper, we are concerned with the numerical integration of the NLS equation on a d dimensional torus:
where \({\mathbb{T}}=(0,2\pi )\), \(\lambda =\pm 1\), \(u=u(t,\boldsymbol{x}):{\mathbb{R}}^{+}\times {\mathbb{T}}^{d}\to \mathbb{C}\) is the sought-after solution, and \(u_{0}\in {H^{\vartheta }({\mathbb{T}}^{d})}\) for some \({\vartheta \ge 0}\) is the given initial data. Here we only consider the case \(\lambda =1\); the case \(\lambda =-1\) can be treated in exactly the same way. Note that the well-posedness of the nonlinear Schrödinger equation in \({H^{\vartheta }({\mathbb{T}}^{d})}\) has been established for \({\vartheta >\frac{d}{2}-1}\). For details, we refer to [2].
Many authors studied numerical aspects of the NLS equation. A considerable amount of literature has been published on splitting and exponential integration methods. For a general introduction to these methods, we refer to [9–12, 18]. It is well known that schemes of an arbitrarily high order can be constructed assuming that the solution of (1.1) is smooth enough. For instance, the second-order convergence in \({H^{\vartheta }}\) was obtained by requiring four additional derivatives of the solution for the Strang splitting scheme [17]. Further convergence results for the semilinear Schrödinger equation can be found, e.g., in [1, 4–6, 8, 13, 14, 21].
For classical methods and their analysis, strong regularity assumptions are unavoidable. Recently, however, so-called low-regularity integrators have emerged as a powerful tool for reducing the regularity requirements. The first breakthrough was made in [19], where the authors introduced a new exponential-type numerical scheme and achieved first-order convergence in \({H^{\vartheta }(\mathbb{T}^{d})}\) for \({H^{\vartheta +1}(\mathbb{T}^{d})}\) initial data with \(\vartheta >\frac{d}{2}\). Later, a first-order integrator was proposed in [22]. It converges in \({H^{\vartheta }}(\mathbb{T})\) without any loss of regularity and conserves mass up to order five. A second-order Fourier-type integrator was given by Knöller, Ostermann, and Schratz [16]. The integrator is based on the variation-of-constants formula and uses certain resonance-based approximations in Fourier space. For the second-order convergence, the scheme requires two additional derivatives of the solution in one space dimension and three derivatives in higher space dimensions. In this paper, we present and analyze an improved integrator that enables us to get the desired second-order accuracy with only two additional bounded spatial derivatives in dimensions \(d\ge 1\).
There are two main difficulties in designing low-regularity integrators. The first one is to control the spatial derivatives in the approximation while keeping the nonlinearity point-wise defined in physical space rather than in Fourier space. The second one is to overcome the difficulties caused by the complicated structure of resonances in higher dimensions. To explain this, let
and consider the phase function
In [16], letting
the authors approximated the phase function by
where \(|\mathcal{R}_{1}(\alpha ,\beta ,s)|\lesssim s^{2}|\alpha ||\beta |\). This choice requires three additional derivatives in higher space dimensions for second-order convergence.
Now we explain our current approach, for which we consider a slightly more general situation. Assume that α has a “good” structure, which means \(\int _{0}^{\tau }\mathrm{e}^{is\alpha }\,ds\) is point-wise defined (as in the example above) while β has a “bad” structure but still has a low upper bound, e.g., consisting of mixed derivatives (as in the example above). In order to approximate
where
we employ the Taylor series expansion
Next, we use that
where
and replace \(i\beta \pi \) by \(\mathrm{e}^{i\beta \tau }-1\). Then, we obtain
where \(|\mathcal{R}_{2}(\alpha ,\beta ,\tau )|\lesssim \tau ^{3}|\beta |^{2}\). This bound will be proved in Lemma 2.2 below. Relying on this structure, the scheme requires only two additional derivatives for \(\tau ^{2}\), which gives convergence in \({H^{\vartheta }}(\mathbb{T}^{d})\) for initial data in \({H^{\vartheta +2}(\mathbb{T}^{d})}\).
Finally, it does not require any specific structure of β. In particular, \(\beta ^{-1}\) is not contained in the expression (1.5). This is another advantage compared to (1.2), for which the integration (or a further approximation) of \(\int _{0}^{\tau }\mathrm{e}^{is\beta }\,ds\) is needed.
Now we state the main result of this paper. We define the new low-regularity integrator with second-order accuracy as
for \(n\ge 0\). For this method, we have the following convergence result.
Theorem 1.1
Let \(u^{n}\) be the numerical solution (1.6) of the Schrödinger equation (1.1) up to some fixed time \(T>0\). Under the assumption that \(u_{0}\in H^{\gamma +2}({\mathbb{T}}^{d})\) for some \(\gamma >\frac{d}{2}\), there exist constants \(\tau _{0}, C>0\) such that for any \(0<\tau \leq \tau _{0}\), it holds
The constants \(\tau _{0}\) and C only depend on T and \(\|u\|_{L^{\infty }((0,T);H^{\gamma +2}({\mathbb{T}}^{d}))}\).
Having finished the analysis of this paper, we became aware of the recent work [3] by Bruned and Schratz, in which low-regularity integrators for dispersive equations are discussed in a broader context. In particular, using the formalism of decorated trees, various numerical methods for the nonlinear Schrödinger equation are proposed. The above method (1.6) is stated in formula (5.17). Nevertheless, we give here an alternative (and brief) derivation of the method because the employed approximations form the basis of our rigorous error analysis.
The paper is organized as follows: In Sect. 2, we introduce some notations and collect some useful lemmas. In Sect. 3, we discuss the construction of the method and analyze the accuracy and regularity requirements of each single approximation step. Collecting all these results, we prove our convergence result (Theorem 1.1) in Sect. 4. This theoretical result is illustrated with some numerical experiments in Sect. 5.
2 Preliminaries
In this section, we introduce some notations, recall a result from harmonic analysis, and give some elementary estimates. All of these will be used frequently in the following sections.
2.1 Some notations
We start with notations, some of which are borrowed from [7]. We write \(A\lesssim B\) or \(B\gtrsim A\) to denote the statement that \(A\leq CB\) for some constant \(C>0\). This constant may vary from line to line, but it is independent of τ or n. Further, we write \(A\sim B\) for \(A\lesssim B\lesssim A\), we denote
and define \((d\boldsymbol{\xi })\) to be the normalized counting measure on \(\mathbb{Z}^{d}\) such that
The Fourier transform of a function f on \(\mathbb{T}^{d}\) is defined by
Instead of f̂, we sometimes also write \(\mathcal{F}f\) or \(\mathcal{F}(f)\). The Fourier inversion formula takes the form
We recall the following properties of the Fourier transform:
For the Sobolev space \(H^{s}({\mathbb{T}}^{d})\), \(s\geq 0\), we consider the equivalent norm
where \(J^{s}=(1-\Delta )^{\frac{s}{2}}\).
2.2 Some estimates
First, we recall the following inequalities, which were originally proved in [15].
Lemma 2.1
(The Kato-Ponce inequality, [15])
The following inequalities hold:
-
(i)
For any \(\gamma >\frac{d}{2}\) and \(f,g\in H^{\gamma }\), we have
$$\begin{aligned} \bigl\Vert J^{\gamma }(fg) \bigr\Vert _{L^{2}}\lesssim \Vert f \Vert _{H^{\gamma }} \Vert g \Vert _{H^{\gamma }}. \end{aligned}$$ -
(ii)
For any \(\delta \ge 0\), \(\gamma >\frac{d}{2}\) and \(f\in H^{\delta +\gamma }\), \(g\in H^{\delta }\), we have
$$\begin{aligned} \bigl\Vert J^{\delta }(fg) \bigr\Vert _{L^{2}}\lesssim \Vert f \Vert _{H^{\delta +\gamma }} \Vert g \Vert _{H^{ \delta }}. \end{aligned}$$
The next lemma plays a crucial role in the analysis of this paper.
Lemma 2.2
Let \(\alpha ,\beta \in \mathbb{R}\). Then, the following properties hold:
-
(i)
For ψ defined as in (1.4), we have
$$\begin{aligned} \frac{1}{\tau } \int _{0}^{\tau }s \mathrm{e}^{is\alpha }\,ds =-\tau \psi (i \tau \alpha ). \end{aligned}$$(2.1) -
(ii)
There exists a function \(\mathcal{R}_{2}(\alpha ,\beta ,\tau )\) such that
$$\begin{aligned} \int _{0}^{\tau }\mathrm{e}^{is(\alpha +\beta )}\,ds=\tau \varphi (i \tau \alpha )-\tau \bigl(\mathrm{e}^{i\tau \beta }-1 \bigr)\psi (i\tau \alpha )+\mathcal{R}_{2}(\alpha ,\beta ,\tau ) \end{aligned}$$(2.2)with \(|\mathcal{R}_{2}(\alpha ,\beta ,\tau )|\lesssim \tau ^{3}|\beta |^{2}\).
Proof
(i) We first note that
Using integration by parts, we then find that
which proves (i).
(ii) From (2.2), we obtain that the remainder \(\mathcal{R}_{2}\) satisfies
Using (2.1) and (2.3), we rewrite (2.5) in the following way
First, we decompose
and thus get
Finally, note that
Therefore, (2.6) can be controlled by \(C\tau ^{3}|\beta |^{2}\). □
3 Construction of the method
Now we derive a second-order numerical method for (1.1). Since the employed approximations form the basis of our error analysis, we present some construction details. For an alternative derivation of this method, we refer to [3].
Let \(\tau >0\) be the time step size and \(t_{n}=n\tau \), \(n\ge 0\) the temporal grid points. First, by employing the twisted variable \(v=\mathrm{e}^{-it\Delta }u\) and Duhamel’s formula, we get
Then, freezing the nonlinear interaction by approximating \(\mathrm{e}^{i(t_{n}+\rho )\Delta }\approx \mathrm{e}^{i(t_{n}+\sigma ) \Delta }\) and \(v(t_{n}+\rho )\approx v(t_{n})\), we get
The remainder term \(\mathcal{R}_{3}^{n}(v,\sigma )\) satisfies the following estimate.
Lemma 3.1
Let \(\gamma >\frac{d}{2}\), \(\sigma \in [0,\tau ]\) and \(v\in L^{\infty }((0,T);H^{\gamma +2})\). Then,
We postpone the proof of the lemma to Sect. 3.1.
Next, we derive a second-order expansion of Duhamel’s formula
Replacing \(v(t_{n}+\sigma )\) by (3.2), we infer that
where
The remainder term \(\mathcal{R}_{4}^{n}(v)\) can be bounded as stated in the following lemma. Again, the proof of this lemma is postponed to Sect. 3.1.
Lemma 3.2
Let \(\gamma >\frac{d}{2}\) and \(0<\tau \leq 1\). Then, for \(v\in L^{\infty }((0,T);H^{\gamma +2})\),
where the constant C only depends on \(\|v\|_{L^{\infty }((0,T);H^{\gamma +2})}\).
Due to the complexity of the phase functions
we note that the terms in \(I_{1}\) and \(I_{2}\) cannot be easily expressed in physical space.
Therefore, we consider \(I_{1}\) first in Fourier space. Using
we get
The main problem concerns the handling of the phase \(\mathrm{e}^{is\phi _{3}}\). Defining
allows us to write
Applying the formulas presented in Lemma 2.2, we get
where the remainder term \(\mathcal{R}_{5}^{n}(v)\) obeys the bound given in the following lemma. Its proof will be postponed to Sect. 3.1.
Lemma 3.3
Let \(\gamma >\frac{d}{2}\) and \(v\in L^{\infty }((0,T);H^{\gamma +2})\). Then,
Using \(\beta =\phi _{3}-\alpha \) and (3.6), we transform (3.7) back to physical space to get
The term \(I_{2}\) is of higher order in τ. Therefore, it is sufficient to freeze the linear flow and approximate the term as
where the remainder term \(\mathcal{R}_{6}^{n}(v)\) obeys the bound given in the following lemma. Again, its proof will be postponed to Sect. 3.1.
Lemma 3.4
Let \(\gamma >\frac{d}{2}\) and \(v\in L^{\infty }((0,T);H^{\gamma +2})\). Then
Now combining (3.4), (3.8), and (3.10), we have that
where the operator \(\Phi ^{n}\) is defined by
Our second-order low-regularity integrator is obtained by dropping the remainder terms \(\mathcal{R}_{4}^{n}\), \(\mathcal{R}_{5}^{n}\), \(\mathcal{R}_{6}^{n}\) in (3.11). The method for the twisted variable is summarized as follows: let \(v^{0}=u_{0}\) and
Finally, setting \(u^{n}=\mathrm{e}^{it_{n}\Delta }v^{n}\), we obtain the announced numerical scheme (1.6) for the NLS equation (1.1).
3.1 Estimates of the remainder terms
Now we prove Lemmas 3.1 to 3.4.
Proof of Lemma 3.1
By (3.2), we have that
Note that from (3.1), Lemma 2.1(i), and the Sobolev embedding, we get
Moreover, for any \(f\in H^{\gamma }\),
Applying these two estimates, we obtain
and thus
This is the desired result. □
Proof of Lemma 3.2
Inserting (3.2) with \(\sigma =\rho \) in (3.1) and using (3.4), we find that the remainder \(\mathcal{R}_{4}^{n}(v)\) consists of terms of the form
where
By Lemma 3.1 and Lemma 2.1(i), we thus get
This finishes the proof of the lemma. □
Proof of Lemma 3.3
Without loss of generality, we may assume that \(\hat{v}(t_{n})\) and \(\hat{\bar{v}}(t_{n})\) are positive (otherwise, one may replace them with their absolute values).
From Lemma 2.2, we have
and further
By symmetry, we may assume that \(|\boldsymbol{\xi }_{1}|\ge |\boldsymbol{\xi }_{2}|\ge | \boldsymbol{\xi }_{3}|\). This yields
Using this estimate, we get
Therefore, by Plancherel’s identity and Lemma 2.1(ii) with \(\delta =0\), we obtain that for any \(\gamma _{1}>\frac{d}{2}\),
Since \(\gamma >\frac{d}{2}\), choosing \(\gamma _{1}=\gamma \), we get the desired result. □
Proof of Lemma 3.4
By (3.5) and (3.9), we have that
Then, the claimed result follows directly from (3.14) and Lemma 2.1(i). □
4 Proof of Theorem 1.1
Taking the difference between the numerical scheme (3.13) and the exact solution gives
where \(\mathcal{L}^{n}=\Phi ^{n} (v(t_{n}) )-v(t_{n+1})\) is the local error.
4.1 Local error
The following bound on the local error holds.
Lemma 4.1
Let \(\gamma >\frac{d}{2}\) and \(0<\tau \leq 1\). Then,
where the constant C only depends on \(\|v\|_{L^{\infty }((0,T);H^{\gamma +2})}\).
Proof
By (3.11), we get that
Thus, the desired estimate follows from Lemmas 3.2, 3.3, and 3.4. □
4.2 Stability
The main result in this subsection is the following stability estimate.
Lemma 4.2
Let \(\gamma >\frac{d}{2}\). Then,
where the constant C only depends on \(\|v\|_{L^{\infty }((0,T);H^{\gamma })}\).
Proof
For short, we denote \(g_{n}=v^{n}-v(t_{n})\). Then, using (3.12), we have
where
Note that by the definition of φ and ψ in (1.4), we have that
Hence, by Lemma 2.1(i),
where C only depends on \(\|v \|_{L^{\infty }((0,T); H^{\gamma })}\).
Similarly, we get that
Combining the above estimates, we finally obtain
which is the desired result. □
4.3 Proof of Theorem 1.1
Now, combining the local error estimate with the stability result, we prove Theorem 1.1. From Lemma 4.1 and Lemma 4.2, we infer that there exists a constant C depending only on \(\|v\|_{L^{\infty }((0,T);H^{\gamma +2})}\) such that for \(0<\tau \leq 1\), we have
By recursion, we get from this the bound
From this estimate, we infer that there exist positive constants \(\tau _{0}\) and C such that for any \(\tau \in [0,\tau _{0}]\),
Note that the constants \(\tau _{0}\) and C only depend on T and \(\|u\|_{L^{\infty }((0,T);H^{\gamma +2})}\). This proves Theorem 1.1.
5 Numerical experiments
In this section, we carry out some numerical experiments to illustrate our convergence result in two space dimensions. For this purpose, we consider the nonlinear Schrödinger equation (1.1) with initial data
where γ is used to set the regularity of the data. This choice guarantees that \(u^{0}\in H^{\gamma }(\mathbb{T}^{2})\). In the experiment, we set \(\varepsilon =0\).
In order to be able to use FFT techniques, we discretize space by equidistant grid points
The numerical approximation, obtained with step size τ at \(t=t_{n}=n\tau \) on this grid will be denoted by \(u^{n}_{\tau ,N}\). We choose \(N=2^{7}\), i.e., 214 grid points, and measure the temporal discretization error \(w = u(t_{n},\cdot )-u^{n}_{\tau ,N}\), defined on the grid by \(w(t_{n},x_{1}^{j},x_{2}^{m})=u(t_{n},x_{1}^{j},x_{2}^{m})-u^{n}_{ \tau ,N}(j,m)\). We consider this matrix as an element of the linear space \(L^{2}_{N}\) with norm \(\|\cdot \|_{L^{2}_{N}}\) defined by
The discrete \(H^{\gamma }_{N}\) spaces are then defined in the usual way with the help of the discrete Fourier transform, i.e.,
Our results for initial data \(u_{0}\in H^{\gamma +2}(\mathbb{T}^{2})\) are presented in Fig. 1. We choose the three different values \(\gamma =1\), 1.5, 2 to illustrate the convergence rate. In the left panel, we present the results for the standard Strang splitting. As expected, the Strang splitting shows a strong order reduction and irregular error behavior. For our scheme (1.6), the results are given in the right panel. As expected, the slopes of the error curves are 2 whenever γ is bigger than 1. The slope of the curve for \(\gamma =1\) is slightly less regular. This is also expected because the value \(\gamma =1\) is the limit case in two space dimensions. Thus, the results agree well with the corresponding results of the theoretical analysis given in Theorem 1.1.
Availability of data and materials
Not applicable.
References
Besse, C., Bidégaray, B., Descombes, S.: Order estimates in time of splitting methods for the nonlinear Schrödinger equation. SIAM J. Numer. Anal. 40, 26–40 (2002)
Bourgain, J.: Fourier transform restriction phenomena for certain lattice subsets and applications to nonlinear evolution equations. I. Schrödinger equations. Geom. Funct. Anal. 3, 107–156 (1993)
Bruned, Y., Schratz, K.: Resonance based schemes for dispersive equations via decorated trees. Forum Math. Pi 10, 1–76 (2022)
Cano, B., González-Pachón, A.: Exponential time integration of solitary waves of cubic Schrödinger equation. Appl. Numer. Math. 91, 26–45 (2015)
Celledoni, E., Cohen, D., Owren, B.: Symmetric exponential integrators with an application to the cubic Schrödinger equation. Found. Comput. Math. 8, 303–317 (2008)
Cohen, D., Gauckler, L.: One-stage exponential integrators for nonlinear Schrödinger equations over long times. BIT Numer. Math. 52, 877–903 (2012)
Colliander, J., Keel, M., Staffilani, G., Takaoka, H., Tao, T.: Multilinear estimates for periodic KdV equations, and applications. J. Funct. Anal. 211, 173–218 (2004)
Dujardin, G.: Exponential Runge–Kutta methods for the Schrödinger equation. Appl. Numer. Math. 59, 1839–1857 (2009)
Faou, E.: Geometric Numerical Integration and Schrödinger Equations. Eur. Math. Soc., Zurich (2012)
Hairer, E., Lubich, C., Wanner, G.: Geometric Numerical Integration. Structure-Preserving Algorithms for Ordinary Differential Equations, 2nd edn. Springer, Berlin (2006)
Hochbruck, M., Ostermann, A.: Exponential integrators. Acta Numer. 19, 209–286 (2010)
Holden, H., Karlsen, K.H., Lie, K.-A., Risebro, N.H.: Splitting for Partial Differential Equations with Rough Solutions. Eur. Math. Soc., Zurich (2010)
Ignat, L.I.: A splitting method for the nonlinear Schrödinger equation. J. Differ. Equ. 250, 3022–3046 (2011)
Jahnke, T., Lubich, C.: Error bounds for exponential operator splittings. BIT Numer. Math. 40, 735–744 (2000)
Kato, T., Ponce, G.: Commutator estimates and the Euler and Navier–Stokes equations. Commun. Pure Appl. Math. 41, 891–907 (1988)
Knöller, M., Ostermann, A., Schratz, K.: A Fourier integrator for the cubic nonlinear Schrödinger equation with rough initial data. SIAM J. Numer. Anal. 57, 1967–1986 (2019)
Lubich, C.: On splitting methods for Schrödinger–Poisson and cubic nonlinear Schrödinger equations. Math. Comp. 77, 2141–2153 (2008)
McLachlan, R.I., Quispel, G.R.W.: Splitting methods. Acta Numer. 11, 341–434 (2002)
Ostermann, A., Schratz, K.: Low regularity exponential-type integrators for semilinear Schrödinger equations. Found. Comput. Math. 18, 731–755 (2018)
Sulem, C., Sulem, P.-L.: The Nonlinear Schrödinger Equation. Self-Focusing and Wave Collapse. Springer, New York (1999)
Thalhammer, M.: Convergence analysis of high-order time-splitting pseudo-spectral methods for nonlinear Schrödinger equations. SIAM J. Numer. Anal. 50, 3231–3258 (2012)
Wu, Y., Yao, F.: A first-order Fourier integrator for the nonlinear Schrödinger equation on \(\mathbb{T}\) without loss of regularity. Math. Comp. https://doi.org/10.1090/mcom/3705
Acknowledgements
Parts of the research were carried out during a research visit of F.Y. at the University of Innsbruck.
Funding
Y.W. was partially supported by the NSFC grants 12171356 and 11771325. F.Y. acknowledges financial support by the China Scholarship Council.
Author information
Authors and Affiliations
Contributions
The authors declare that the study was realized in collaboration with equal responsibility. All authors read and approved the final manuscript.
Corresponding author
Ethics declarations
Competing interests
The authors declare that they have no competing interests.
Rights and permissions
Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/.
About this article
Cite this article
Ostermann, A., Wu, Y. & Yao, F. A second-order low-regularity integrator for the nonlinear Schrödinger equation. Adv Cont Discr Mod 2022, 23 (2022). https://doi.org/10.1186/s13662-022-03695-8
Received:
Accepted:
Published:
DOI: https://doi.org/10.1186/s13662-022-03695-8