Arbitraryorder energypreserving exponential integrators for the cubic Schrödinger equation
Abstract
In this paper we derive and analyse new and efficient energypreserving exponential integrators of arbitrarily high order to solve the cubic Schrödinger Cauchy problem on a dimensional torus. Energy preservation is a key feature of the cubic Schrödinger equation. It is proved that the novel integrators can be of arbitrarily high order which exactly preserve the continuous energy of the original continuous system. The existence and uniqueness, regularity, global convergence, nonlinear stability of the new integrators are studied in detail. One of the new energypreserving exponential integrators is constructed and two numerical experiments are included. The numerical results illustrate the efficiency of the new integrator in comparison with existing numerical methods in the literature.
Keywords: Cubic Schrödinge equationEnergy preservationExponential integrators
MSC: 65P1035Q5565M1265M70
1 Introduction
It is well known that one of the cornerstones of quantum physics is the Schrödinger equation. The nonlinear Schrödinger equation has long been used to approximately describe the dynamics of complicated systems, such as Maxwell¡¯s equations for the description of nonlinear optics or the equations describing surface water waves, including rogue waves which appear from nowhere and disappears without a trace (see, e.g. [1, 2, 9]). This paper is devoted to designing and analysing novel numerical integrators to preserve the continuous energy of the cubic Schrödinger equation
(1) 
where is denoted the onedimensional torus and for the given positive integer , denotes the dimensional torus. In this paper, we consider this equation as a Cauchy problem in time (no space discretisation is made). The solutions of this equation have conservation of the following energy
(2) 
It is of great interest to devise numerical schemes which can conserve the continuous version of this important invariant, and the aim of this paper is to formulate a novel kind of exponential integrators with this fundamental property.
Schrödinger equations frequently arise in a wide variety of applications including several areas of physics, fiber optics, quantum transport and other applied sciences (see, e.g. [3, 24, 35, 43, 53]). Many efficient and effective methods have been proposed for the numerical integration of Schrödinger equations, such as splitting methods (see, e.g. [5, 10, 23, 26, 44, 46, 54]), exponentialtype integrators (see, e.g. [14, 15, 19, 21, 49]), multisymplectic methods (see, e.g. [6, 50]) and other efficient methods (see, e.g. [8, 27, 34, 36]).
In recent decades, structurepreserving algorithms of differential equations have been received much attention and for the related work, we refer the reader to [25, 29, 39, 55, 57, 59, 56, 58, 60, 62] and references therein. It is well known that structurepreserving algorithms are able to exactly preserve some structural properties of the underlying continuous system. Amongst the typical subjects of structurepreserving algorithms are energypreserving schemes, which exactly preserve the energy of the underlying system. There have been a lot of studies on this topic for Hamiltonian partial differential equations (PDEs). In [52], finite element methods were introduced systematically for numerical solution of PDEs. The authors in [20] researched discrete gradient methods for PDEs. The work in [16] investigated the average vector field (AVF) method for discretising Hamiltonian PDEs. Hamiltonian Boundary Value Methods (HBVMs) were studied for the semilinear wave equation in [12] and were recently researched for nonlinear Schrödinger equations with wave operator in [13]. The adapted AVF method for Hamiltonian wave equations was analysed in [41, 42]. Other related work is referred to [11, 18, 22, 37, 40, 45, 47, 51]. On the other hand, exponential integrators have been widely introduced and developed for solving firstorder ODEs, and we refer the reader to [29, 30, 31, 32, 48] for example. This kind of methods has also been studied in the numerical integration of Schrödinger equations (see, e.g. [14, 15, 19, 21, 49]). However, until now, energy preserving exponential integrators for Schrödinger equations that exactly preserve the continuous energy, do not seem to have been studied in the literature, which motivates this paper.
With this premise, this paper is mainly concerned with arbitraryorder energypreserving exponential integrators for solving cubic Schrödinger equations. The remainder of this paper is organized as follows. We first present some notations and preliminaries in Section 2. Then the scheme of energypreserving exponential integrators is formulated in Section 3. Section 4 discusses the implementation issues. In Section 5, we analyse the existence, uniqueness and smoothness of the integrators. Section 6 pays attention to the regularity. The convergence of the integrators is studied in Section 7 and the nonlinear stability is discussed in Section 8. Section 9 is devoted to constructing one of practical exponential integrators in the light of the approach proposed in this paper, and reporting two numerical experiments to demonstrate the excellent qualitative behavior of the new exponential integrator. Section 10 focuses on the concluding remarks.
2 Notations and preliminaries
In this paper, we use the following notations.

For , we denote For and , denotes the function on defined by

We denote by (or simply ) the set of (classes of) complex functions on such that , endowed with the norm

For all functions and all , denote by the Fourier coefficient

For , denote by (or simply ) the space of (classes of) complex functions such that endowed with the norm Note that with the same norm.

For all , we call a linear operator acting on as diagonal, if is in the domain of and there exists a such that . For example, the Laplace operator acting on is diagonal since for all , Another example is the identity operator on which will be denoted by .

For all such operator and all functions from to itself, denote by the diagonal linear operator acting on whose domain is the set of linear combinations of functions defined for all by

For all and all linear operators from to itself, we denote
The following result given in [21] will be useful for the analysis of this paper.
Proposition 1
(See [21]) With the above notations, if is a function from to bounded by on , then for all and , is a bounded linear operator from to itself with . For example, for all , it is true that
In order to derive the energypreserving exponential integrators, we will use the idea of continuous time finite element methods in a generalised function space. To this end, we first present the following three definitions.
Definition 1
Define the generalised function space =span on as follows:
where is an integer satisfying and are supposed to be linearly independent on and sufficiently smooth. In this paper, we choose and as follows:
with for Then choose a time stepsize and define and on as
(3) 
where for . It is noted that throughout this paper, the notations and are referred to as and for all the functions, respectively.
Definition 2
The inner product for the time is defined by
where and are two integrable functions for (scalarvalued or vectorvalued) on , and if they are both vectorvalued functions, ‘’ denotes the entrywise multiplication operation.
Definition 3
A projection onto is defined as
(4) 
where is a continuous twodimensional vector function for .
With regard to the projection operation , the following property is important.
Lemma 1
The projection can be explicitly expressed as
where and is a standard orthonormal basis of under the inner product given by
for
Proof The standard orthonormal basis of is obtained immediately from the choice of . Since , it can be expressed as By taking in (4) for and , we obtain
which gives By the standard orthonormal basis , this result can be formulated as
Then one has
which proves the result.
3 Formulation of energypreserving schemes
Define the linear differential operator by
and let . Then the system (1) is identical to
(5) 
The solutions of this equation satisfy Duhamel’s formula
(6) 
Let and then the equation (1) can be rewritten as a pair of real initialvalue problems
(7)  
In this case, the energy of this system is expressed by
(8) 
Accordingly, the system (7) can be formulated as the following infinitedimensional real Hamiltonian system
(9) 
where and
With the preliminaries described above, we first derive the energypreserving exponential integrators for solving the realvalued equation (9) and then present the integrators for solving the cubic Schrödinger equation (1).
Find with , satisfying that
(10) 
for any , where From the definition (4) of the operator , it follows that
which yields the following result by considering Lemma 1
(11) 
Applying Duhamel’s formula (6) to (11) leads to
This yields the following definition of energypreserving exponential integrators for (9) in this paper.
Definition 4
The energypreserving exponential integrator for solving the real Hamiltonian initialvalue problem (9) is defined by
(12) 
where is a time stepsize and
(13) 
Theorem 1
Proof For each function , it follows from (10) that
for , where is the th vector of units and denotes the th entry of a vector. Then, one arrives at
(14)  
Since , we obtain that and . Letting in (14) gives
Because is skew symmetric, we have
Therefore, it is true that
On the basis of the analysis stated above, we now present the energypreserving scheme for solving the cubic Schrödinger equation (1).
Definition 5
The energypreserving exponential integrator (denoted as EPEIr) with a time stepsize for the cubic Schrödinger equation (1) is defined by
(15) 
where
(16) 
Theorem 2
Proof In what follows, we prove that the schemes (15) and (12) share the same result for solving (1). With careful calculations, we obtain
On the other hand, we have
These formulae show that the above two different expressions share the same result for solving (1). Moreover, it is clear that and then and yield the same result. According to the above analysis and looking closer at the two schemes (15) and (12), it can be confirmed that they are the same integrator when applied to (1).
4 Implementation issues
Introduce
with respect to distinct points . Then is a basis of satisfying It is assumed that since . Choosing for and using , one gets which leads to Let and denote . The integrator (15) now becomes
(17) 
It is clear that in (17), inserting the appearing in the two integrals by the first equation yields the scheme of EPEIr integrators. Consequently, we have the following definition for energypreserving exponential integrators in practice.
Definition 6
5 The existence, uniqueness and smoothness
In the remainder of this paper, we use the following assumptions on the exact solution of (1) and on the nonlinearity .
Assumption 1
It is assumed that the Schrödinger equation (1) admits an exact solution which is sufficiently smooth. In particular, there exists such that for all , where
and .
Assumption 2
We assume that the mapping is sufficiently smooth.
Assumption 3
The thorder derivative of is assumed that . And we denote for
Note that these assumptions are fulfilled in the case of the cubic nonlinear Schrödinger equation (see [4] for example, Chapter II, Proposition 2.2 ), at least when and .
According to Proposition 1, one gets that the coefficients and of our integrators for and are uniformly bounded. Hence, we let
(19) 
Theorem 3
Under the above assumptions, if satisfies
(20) 
then the EPEIr integrator (15) admits a unique solution which smoothly depends on .
Proof By setting and defining
(21) 
we get a function series If is uniformly convergent, is a solution of the EPEIr integrator (15).