Abstract
In this paper, we present two linearized BDF2 Galerkin FEMs for the nonlinear and coupled Schrödinger-Helmholtz equations. Different from the standard linearized second-order Crank-Nicolson methodology, we employ backward differential concept to obtain second-order temporal accuracy at the time step tn (instead of the time instant tn+ 1/2) and apply semi-implicit or explicit treatment of nonlinear terms to formulate the decoupled schemes. We prove optimal error estimates for r-order FEM without any grid-ratio condition through a so-called temporal-spatial error splitting technique, and some sharp estimations to cope with the nonlinear terms. Finally, we provide two numerical experiments to illustrate the theoretical analysis and the efficiency of the proposed methods. Here, h is the spatial subdivision parameter, and τ is the time step.
We’re sorry, something doesn't seem to be working properly.
Please try refreshing the page. If that doesn't work, please contact support so we can address the problem.
Avoid common mistakes on your manuscript.
1 Introduction
Consider the following initial-boundary value problem of Schrödinger-Helmholtz equations
in which \(X=(x,y), T<+\infty \) and Ω is a convex bounded domain in \(\mathbb {R}^{d} (d=2,3)\) with the boundary ∂Ω. \(\mathrm {i}=\sqrt {-1}\), α,β are real nonnegative constants with α + β≠ 0. \(f:\mathbb {R}^{+}\rightarrow \mathbb {R}\) and \(u_{0}:{\Omega }\rightarrow \mathbb {C}\) are given functions. The complex-valued function u stands for the single particle wave function, the real-valued function ϕ(X,t) denotes the potential. The system (1.1) models many different physical phenomena in optics, quantum mechanics, and plasma physics, and so forth. When α = 0, the system (1.1) reduces to the Schrödinger-Poisson model [1,2,3,4,5]. And when β = 0, the system (1.1) degenerate to a generalized nonlinear Schrödinger equation [6, 7]. Besides, we refer [8, 9] for other Schrödinger type equations such as the Schrödinger-Poisson-Slater model. We can see that the model (1.1) conserves the total mass
and the total energy
when α = 1,f(s) = sσ− 1,σ ≥ 1.
A series of mathematical studies have been devoted for diverse Schrödinger type equations. For example, the existence and uniqueness of solution to the Schrödinger-Poisson type equations in \(\mathbb {R}^{d} (d=2,3)\) were investigated in [10, 11]. And in [12], a type of Schrödinger-Helmholtz system as a regularization of the generalized nonlinear Schrödinger equation was introduced, local and global existence of a unique solution of the system was studied. Along the numerical front, various numerical methods for Schrödinger type equations also have been proposed including finite difference methods [13,14,15,16,17,18], spectral or pseudo-spectral methods [19,20,21,22], and FEMs [23,24,25,26,27,28,29]. Especially, the linearized backward Euler Galerkin FEMs and Crank-Nicolson Galerkin FEMs were studied for Schrödinger-Helmholtz system in [30] and [31], respectively. Both of them derived optimal L2 error estimates for r-order FEMs without any grid-ratio restriction condition. Due to some pollution arising from the approximation used for the nonlinear terms ϕf(|u|)u and f(|u|)|u|2, only the error estimate at the time instant tn+ 1/2 instead of the time division node tn for the potential ϕ was derived in [31], because Schrödinger part u and the Helmholtz part ϕ are solved at different time step levels so as to decouple the strongly nonlinear and coupled of Schrödinger-Helmholtz equations.
Generally speaking, we can derive error estimates at the time instant tn of (linearized) Crank-Nicolson scheme for many different nonlinear PDEs (see [32,33,34]). But for some strong nonlinearity and coupled problems, it is not an easy thing to establish the linearized decoupled high accurate (in time) numerical schemes and obtain the error estimates at the time division node tn. To decouple the schemes, a nature strategy is to solve the coupled problems at different time step levels, see, e.g., [35] for the time-dependent nonlinear thermistor equations, and [36] for Cahn-Hilliard equation, where they achieved the error estimates at the time instant tn+ 1/2 for the electric potential ϕ in [35] and the chemical potential μ in [36], respectively. This phenomenon also produced in [37] for flux \(\vec {p}\), where a Crank-Nicolson mixed FEM was used for the nonlinear Sobolev equation. Whether we have the accuracy O(τ2) at the time division node tn for numerical solutions involved in a strongly nonlinear and coupled system such as Schrödinger-Helmholtz (1.1).
In this paper, we shall give an affirmative answer to this question for Schrödinger-Helmholtz (1.1). Based on the second-order BDF temporal approximation framework, we present two BDF2 schemes to solve (1.1) instead of the Crank-Nicolson formula, because the BDF type scheme has the following striking advantages: (i) It is multi-step methods and unconditionally stable [38, 39]; (ii) It can achieve high-order accuracy without increasing the computation significantly [40]; (iii) This kind of scheme treats and approximates every term at time step tn (instead of the time instant tn+ 1/2). To overcome the strong nonlinearity and coupling in ϕf(|u|)u, we adopt semi-implicit or explicit treatment of ϕf(|u|)u to develop the decoupled schemes. Different from [30, 31], we solve Schrödinger equation for numerical solution \({u_{h}^{n}}\) firstly, and then to solve Helmholtz equation for numerical solution \({\phi _{h}^{n}}\) at the same time level tn for our first scheme (see (2.3)–(2.6)). We achieve optimal error estimates without any grid-ratio restriction condition by use of the approach of [41, 42] to split the error into two parts, i.e., the temporal error and the spatial error. Besides, some fine tricks are also applied to deal with the nonlinear terms. The novelty with respect to previous works is that our scheme decoupled the strong nonlinearity of (1.1), and we obtain second-order temporal accuracy at the time step tn (instead of the time instant tn+ 1/2) without time step constraint.
The outline of the article is arranged in the following way. In Section 2, two linearized BDF schemes are developed, and the mass and energy conservative laws are proved. In Section 3, a corresponding time-discrete system is proposed and the temporal error with order O(τ2) is deduced. In Section 4, optimal spatial error estimates with order O(h2) are derived for low-order elements (r = 1), and uniform boundedness of numerical solutions in \(L^{\infty }\)-norm are established, which lead to unconditional optimal L2 error estimates of the r-order (r ≥ 1) Galerkin FEMs in Section 5. Two numerical examples are given to confirm our theoretical analysis in Section 6 and a conclusion is presented in Section 7, respectively.
2 Linearized BDF Galerkin FEMs
In this section, we will construct two linearized BDF2 Galerkin FEMs for (1.1). For this purpose, we use the classical Sobolev spaces Ws,p(Ω) and their associated norms and semi-norms. We denote Hs(Ω) = Ws,2(Ω) with the corresponding norm ∥v∥s = ∥v∥s,2,Ω, seminorm |v|s = |u|s,2,Ω and ∥v∥0 = ∥u∥0,2,Ω defined by
In addition, for any two complex functions u,v ∈ L2(Ω), the inner product is defined by
in which v∗ denotes the conjugate of v. Besides, for a Banach space \(\mathcal {X}\) with the norm \(\|\cdot \|_{\mathcal {X}}\) the function space \(L^{s}(0, T ;\mathcal {X})\) consists of all strongly measurable functions \(f:[0, T ]\rightarrow \mathcal {X}\) with
Following the classical finite element theory [43], we define \(\mathcal {T}_{h}=\{\mathcal {K}\}\) to be a quasi-uniform partition of Ω into triangles or rectangles (in \(\mathbb {R}^{2}\)) or hexahedrons (in \(\mathbb {R}^{3}\)) with mesh size \(h=\max \limits \nolimits _{\forall \mathcal {K}\in \mathcal {T}_{h}}\{\text {diam} \mathcal {K}\}\) and 0 < h < 1. We introduce the finite element space functions defined as (see, e.g., [44])
where \(\mathcal {P}_{r}(\mathcal {K})\) denotes the polynomial space of degree ≤ r(r ≥ 1) and \(\mathcal {Q}_{r}(\mathcal {K})\) denotes the polynomial space of degree ≤ r in each variable.
To proceed, we define the Ritz projection operator \(R_{h}:{H_{0}^{1}}({\Omega })\rightarrow V_{h0}\) by [43, 44]:
which satisfies
Here and later, C with or without superscripts and subscripts, denotes a generic positive constant, not necessarily the same at different occurrences, which is always dependent on the solution and the given data but independent of h and τ.
For a given positive integer N, let {tn : tn = nτ;0 ≤ n ≤ N} be a uniform partition of [0,T] with the time step τ := T/N, nodes tn := nτ for n = 0,1,2,…,N, and intermediate nodes \(t_{n+1/2}:=t_{n}+\frac {\tau }{2}\). Let
and
For the time derivative Dτ, the inner product (Dτσn,σn) can be written by
Here and later \(P(\sigma ^{n}):=\|\sigma ^{n}\|_{0}^{2}+\|2\sigma ^{n}-\sigma ^{n-1}\|_{0}^{2},~ Q(\sigma ^{n}):=\|\sigma ^{n}-2\sigma ^{n-1}+\sigma ^{n-2}\|_{0}^{2}\).
With these notations, we define the linearized BDF2 Galerkin FEM to (1.1) as: to find \(({u_{h}^{n}},{\phi _{h}^{n}})\in \widetilde {V}_{h0}\times V_{h0}\) for all n ≥ 2 such that
where \(\widetilde {V}_{h0}:=V_{h0}\oplus \mathrm {i}V_{h0}\). To ensure the second-order accuracy for the temporal direction, we adopt a predictor corrector method to compute \(({u_{h}^{1}},{\phi _{h}^{1}})\):
in which \((u_{h}^{1,0},\phi _{h}^{1,0})\) is computed by
with the initial value \({u_{h}^{0}}=R_{h}u_{0}\) and \({\phi _{h}^{0}}\) satisfies
for all \((v_{h},\varphi _{h})\in \widetilde {V}_{h0}\times V_{h0}\).
Theorem 1
The discrete scheme (2.3)–(2.3) is mass conservative in the recursive sense
where
with
Proof
Taking \(v_{h}={u_{h}^{1}}+{u_{h}^{0}}\) and \(v_{h}=u_{h}^{1,0}+{u_{h}^{0}}\) in (2.3) and (2.3), respectively, and choosing the imaginary parts of the resulting equations, we derive
Then, taking \(v_{h}={u_{h}^{n}}\) in (2.3) and extracting the imaginary part, we get
which leads to
Therefore, the proof is completed. □
Theorem 2
The discrete scheme (2.3)–(2.3) is energy conservative in the recursive sense
where
with
Proof
Taking \(v_{h}=\bar {\partial }_{\tau }{u_{h}^{1}}\) and \(v_{h}=\frac {u_{h}^{1,0}-{u_{h}^{0}}}{\tau }\) in (2.3) and (2.3), respectively, and choosing the real parts of the resulting equations, we can obtain (2.14) and (2.13), respectively. Finally, taking \(v_{h}=D_{\tau }{u_{h}^{n}}\) in (2.3) and extracting the real part, we have
Thus, we can see that
which leads to the desired results (2.13)–(2.15). □
The scheme (2.3)–(2.6) can be seen as a semi-decoupled scheme because one only needs to solve a linear system for \({u_{h}^{n}}\) firstly, and then for \({\phi _{h}^{n}}\) at each time step. This is different from that in [31] where one needs to solve \(\phi _{h}^{n-1/2}\) firstly, and then to solve for \({u_{h}^{n}}\). However, by use of an explicit treatment of the nonlinear term of (2.3), it allow us to define the following fully decoupled linearized BDF2 scheme: to seek \(({u_{h}^{n}},{\phi _{h}^{n}})\in \widetilde {V}_{h0}\times V_{h0}\) for all n ≥ 2 such that
with \(({u_{h}^{1}},{\phi _{h}^{1}})\) and \((u_{h}^{1,0},\phi _{h}^{1,0})\) is computed by (2.3) and (2.3), respectively, the initial value \({u_{h}^{0}}=R_{h}u_{0}\) and \({\phi _{h}^{0}}\) satisfies (2.6). We point out that we can solve the two (2.20a)–(2.20b) for \({u_{h}^{n}}\) and \({\phi _{h}^{n}} (n\geq 2)\) in parallel at each time step.
Similar to Theorem 1 and Theorem 2, we also have the following conservative laws for scheme (2.20).
Theorem 3
The discrete scheme (2.20) conserve the following conservative laws in the recursive sense
where
with
In this paper, we only give out the error estimates for the linearized scheme (2.3)–(2.6). The analysis of the second linearized scheme (2.20) can be derived analogously, which will be confirmed numerically in Section 6. Like [31], we assume that \(f:\mathbb {R}\rightarrow \mathbb {R}\) is locally Lipschitz continuous, i.e., for any γ1,γ2 ∈ [−K∗,K∗],
where \(L_{K^{\ast }}\) is the Lipschitz constant dependent on K∗. Besides, we assume that the solution to the problem (1.1) exists and satisfies
where M is a positive constant depends only on Ω.
In our analysis, we need the following lemma which can be found in [45] for the details.
Lemma 1
Assume g ∈ Hm(Ω) and ∂Ω is Cm+ 2 for any nonnegative integer m. Suppose ψ is the unique solution of the boundary value problem
Then ψ ∈ Hm+ 2(Ω) satisfies
where \(\widetilde {C}\) depending on m,Ω and α,β. Especially, (2.29) holds for convex domains when m = 0.
3 Error estimates for temporal discretization
In this section, we first introduce a time-discrete system, then estimate the error functions un − Un and ϕn −Φn, as well as the boundedness of the time-discrete solutions in some norms.
When n ≥ 2, we introduce the following auxiliary equations:
Similar to (2.3)–(2.3), we compute (U1,Φ1) and (U1,0,Φ1,0) by
and
respectively, where U0 = u0 in Ω and Φ0 = ϕ0 satisfies
In what follows, we analyze the error functions un − Un and ϕn −Φn, respectively. To this end, under the regularity assumption (2.28), we define
and let
From (2.3)–(2.5) and (3.1)–(3.3), we have the error equations for n ≥ 2:
and
where
and
By Taylor expansion formula, we have
Theorem 4
Suppose that the system (1.1) has unique solution (u,ϕ) satisfying (2.28). Then there exists τ0 > 0 such that when τ ≤ τ0, the time-discrete system (3.1)–(3.3) is uniquely solvable for n = 1,...,N, satisfying
and
Proof
Since the system (3.1)–(3.3) are linear elliptic equations, the classical theory of elliptic PDEs ensure that the solution of system (3.1)–(3.3) is unique solvable. In what follows, we prove (3.9)–(3.10) by mathematical induction. For the initial time step, multiplying (3.7a) by \((e_{u}^{1,0})^{\ast }\), and integrating it over Ω, then taking the imaginary parts, we find
which together with (3.8) shows that
Similarly, multiplying (3.7a) by \(({\Delta } e_{u}^{1,0})^{\ast }\), integrating it over Ω and summing the real and imaginary parts together, we can see that
which together with (3.12), one has
Combining (3.12) and (3.13), we derive
which implies that
when τ ≤ τ1 = 1/CC3. By use of Lemma 1 and (3.12), we can see that
Therefore
when τ ≤ τ2 = 1/(CC4)1/2.
Next, multiplying (3.7a) by \(({e_{u}^{1}})^{\ast }\), and integrating it over Ω, then taking the imaginary parts, we obtain
which together with (3.12) and (3.16), we conclude
when τ ≤ τ3 = 1/2C.
Moreover, multiplying (3.7a) by \(({\Delta } {e_{u}^{1}})^{\ast }\), integrating it over Ω and taking the real and imaginary parts, respectively. Then summing them together to get
which together with (3.12), (3.16) and (3.18), we find that
This yields
and
when τ ≤ τ4 = 1/(CC7)2/3. By use of Lemma 1 again, and (3.18), we obtain
Therefore
when τ ≤ τ5 = 1/(CC8)1/2.
By mathematical induction, we assume that (3.9)–(3.10) holds for m ≤ n − 1. Then there exists positive τ6 = (CC0)2/3, such that when τ ≤ τ6, we have
Now we prove (3.9)–(3.10) holds for m = n. Multiplying (3.5a) by \(({e_{u}^{n}})^{\ast }\), and taking the imaginary part, it follows that
and summing up from 2 to n, we get
By use of the Gronwall’s inequality, we have
when τ ≤ τ7. The above estimate further shows that
which together with (3.27), and from (3.5a), we obtain
By using Gagliardo-Nirenberg inequality [31], we can see that
when \(\tau \leq \tau _{8}=1/(CC_{K_{0}})^{4/5}\). Therefore,
By use of Lemma 1 and (3.27), we obtain from (3.5) that
which further shows that
when τ ≤ τ9 = (CC11)1/2. Thus (3.9)–(3.10) holds for m = n. The induction is closed. Furthermore, we obtain
for all n = 1,2,...,N. Taking \(\tau _{0}\leq \min \limits _{i=1}^{9}\tau _{i}\) and \(C_{0}\geq \max \limits _{i=1}^{11}C_{i}\), the proof of Theorem 4 is completed. □
4 Error estimates for spatial discretization
In this section, we obtain a τ-independent error estimate for \(U^{n}-{u_{h}^{n}}\) and \({\Phi }^{n}-{\phi _{h}^{n}}\). To do so, we split the errors as follows
Theorem 5
Let (Un,Φn) and \(({u_{h}^{n}},{\phi _{h}^{n}})\) be the solutions of (3.1)–(3.3) and (2.3)–(2.6) respectively for n = 1,2,...,N. Then there exists \(\tau _{0}^{\prime }>0,~h^{\prime }_{0}>0\), such that when \(\tau \leq \tau _{0}^{\prime },~h\leq h_{0}^{\prime }\),
Proof
Since \(\|R_{h}{\Phi }^{n}\|_{0,\infty }\leq C\|{\Phi }^{n}\|_{2}\) and \(\|R_{h}U^{n}\|_{0,\infty }\leq C\|U^{n}\|_{2}\), and by use of (3.11), we can see that \(\|R_{h}{\Phi }^{n}\|_{0,\infty }\) and \(\|R_{h}U^{n}\|_{0,\infty }\) are uniformly bounded, thus we denote by \(K^{\prime }_{0}:=1+\max \limits _{n=0}^{N}\{\|R_{h}{\Phi }^{n}\|_{0,\infty }+\|R_{h}U^{n}\|_{0,\infty }\}\). First, we estimate the initial error. Since \({u_{h}^{0}}=R_{h}u_{0}\), by using (2.1) and (3.11), we have
From (2.6) and (3.4), we obtain
When β = 0, (4.3) leads to
When β≠ 0, (4.3) yields to
By use of the Aubin-Nitsche techniques, we can derive
which shows that
Combining (4.2), (4.4) and (4.7), and employing the inverse inequality, we have
when \(h\leq h_{1}^{\prime }=\min \limits \{\frac {1}{(CC_{1}^{\prime })^{2/(4-d)}},\frac {1}{(CC_{3}^{\prime })^{2/(4-d)}}\}\).
For the first time step, from (2.5) and (3.3), we derive
Choosing vh = ξ1,0 in (4.10a), and taking the imaginary and real parts, then adding them together, one has
Since \(\|\frac {U^{1,0}-U^{0}}{\tau }\|_{2}\leq \|\frac {U^{1,0}-u^{1}}{\tau }\|_{2}+\|\frac {u^{1}-u^{0}}{\tau }\|_{2}\leq C\), we have
and
Thus, (4.11) leads to
which implies that
and
when \(\tau \leq \tau _{1}^{\prime }=1/2C,h\leq h_{2}^{\prime }=\frac {1}{(CC_{4}^{\prime })^{2/(4-d)}}\). Moreover, taking φh = 𝜃1,0 in (4,10b), we derive
which together with (4.15)–(4.16), and by use of the Aubin-Nitsche techniques again, we have
this further implies that
when \(h\leq h_{3}^{\prime }=\frac {1}{(CC_{5}^{\prime })^{2/(4-d)}}\).
On the other hand, from (2.4) and (3.2), we obtain
Likewise, we can see that
which implies that
when \(\tau \leq \tau _{2}^{\prime }=1/2C,~h\leq h_{4}^{\prime }=\frac {1}{(CC_{6}^{\prime })^{2/(4-d)}}\).
We prove (4.1) by mathematical induction. By mathematical induction, we assume that the result (4.1) holds for m ≤ n − 1(n ≥ 2), then there exists \(h_{5}^{\prime }=\frac {1}{(CC_{0}^{\prime })^{2/(4-d)}}\), when \(h\leq h_{5}^{\prime }\) such that
Now we prove (4.1) holds for m = n. By (2.3) and (2.1), one can derive the following error equations
Setting vh = ξn in (4.24a) and taking the imaginary part, we have
Replacing n by i and summing up the equation from 2 to n, we obtain
On the other hand, setting φh = 𝜃n in (4.24b), if β≠ 0, we derive
This shows that
If β = 0, we also obtain from (4.24b) that
Substituting (4.28) into (4.25), then by use of the Gronwall’s inequality, there exists positive constants \(\tau _{3}^{\prime }, C_{7}^{\prime }\), such that when \(\tau \leq \tau _{3}^{\prime }\)
Then, from (4.28) and (4.29), we also have
Further
when \(h\leq h_{6}^{\prime }=\frac {1}{(CC_{7}^{\prime })^{2/(4-d)}}\). Thus, (4.1) holds for m = n if we take \(C_{0}^{\prime }\geq \sum \nolimits _{i=1}^{7}C_{i}^{\prime },~\tau _{0}\leq \min \limits _{i=1}^{3}\tau _{i}^{\prime }\) and \(h_{0}\leq \min \limits _{i=1}^{6}h_{i}^{\prime }\). The proof is completed. □
Remark 1
Clearly, one can see that the error estimate in Theorem 5 is optimal in L2-norm for linear Galerkin FEM, and we can derive optimal H1 error estimate
Furthermore, from the proof of Theorem 5, we can see that the following supercloseness result can be derived when β≠ 0
5 Optimal error estimates for the fully discrete scheme
In this section, we will derive L2 optimal error estimates for the r-order (r ≥ 1) Galerkin FEM by using the results in the above sections.
From (2.1), (3.9), and (4.1), we have optimal error estimates for the linear Galerkin FEM (r = 1) as follows.
Similarly, we derive
and
For r > 1, the above estimates are not optimal for the r-order Galerkin FEM. However, we can derive the uniform bounds of the numerical solutions in \(L^{\infty }\)-norm from Theorem 2 as:
for n = 0,1,⋅⋅⋅,N when \(\tau \leq \tau _{0}^{\prime }, h\leq h_{0}^{\prime }\). By the above uniform bounds, we can obtain optimal L2 error estimates given in the following Theorem.
Theorem 6
Let (Un,Φn) and \(({u_{h}^{n}},{\phi _{h}^{n}})\) be the solutions of (1.1) and (2.3)–(2.6) respectively for n = 1,2,...,N. Then there holds
Proof
Let \({\xi _{u}^{n}}=R_{h}u^{n}-{u_{h}^{n}},\theta _{\phi }^{n}=R_{h}\phi ^{n}-{\phi _{h}^{n}}\). At the time step t = t1/2, we can easily get \(\|{\xi _{u}^{1}}\|_{0}\leq C(h^{r+1}+\tau ^{2})\) and \(\|\theta _{\phi }^{1}\|_{0}\leq C(h^{r+1}+\tau ^{2})\). Thus, we only analyze the errors \(u^{n}-{u_{h}^{n}}\) and \(\phi ^{n}-{\phi _{h}^{n}}\) for 1 ≤ n ≤ N in the following. From (1.1) and (2.3), we obtain
Setting \(v_{h}={\xi _{u}^{n}}\) in (5.7a) and taking the imaginary part to obtain
Replacing n by i and summing up the equation from 2 to n, and noting that
On the other hand, for any n ≥ 1, setting \(\varphi _{h}=\theta _{\phi }^{n}\) in (5.7b), we derive
By the same technique used in the proof of estimates (4.3)–(4.7), we can derive
Thus, with (5.11) and (5.13), there exists a positive \(\tau ^{\prime \prime }\) such that when \(\tau \leq \tau ^{\prime \prime }\), we have
Therefore, by (2.1) and the triangle inequality, we derive (5.6), which completes the proof. □
6 Numerical results
In this section, we present two numerical examples to confirm the efficiency and accuracy of the proposed numerical schemes. In our test, we choose linear and quadratic basis functions on triangular and rectangular finite elements to derive numerical solutions.
Example 1
We consider the following Schrödinger-Helmholtz equation [30, 31].
where Ω = (0,1)2,α = β = 1 in (6.1) and the final time is chosen as T = 2 in the computations. f1,f2 and u0(X) are chosen correspondingly to the exact solutions
Now, we solve the problem (6.1) by the linearized BDF2 schemes (2.3)–(2.6) and (2.20), respectively, with linear triangular element on triangular (P1) and quadratic element on rectangular (Q2) approximation. We choose τ = 5h for P1 element and τ = h3/2 for Q2 element, respectively, and divide the domain Ω into M + 1 nodes in each direction for P1 element with different M = 5,10,20,40, and different mesh-grids m × n = 5 × 5,10 × 10,20 × 20 and 40 × 40 for Q2 element, respectively. The numerical results are listed in Tables 1, 2, 3 and 4 at time t = 0.5,1.0 and 2.0. It can be observed that the errors in L2 norm are proportional to h2 for P1 element and h3 for Q2 element, which are consistent with the theoretical analysis. Additionally, we also observe that the semi-implicit or explicit treatment of the nonlinear terms in the (1.1) has little impact on the convergence of the whole scheme.
To show the unconditional convergence of the linearized BDF2 scheme (2.2)–(2.6) and (2.20) , respectively, we solve the problem (6.1) for each \(\tau =\frac {1}{5},\frac {1}{10},\frac {1}{20},\) with different mesh-grids at time t = 1.0. The numerical results are presented in Figs. 1, 2, 3 and 4 for P1 and Q1 elements. We can see that the numerical errors tend to be a constant as \(\frac {\tau }{h}\rightarrow \infty \) for each fixed τ, which show that grid-ratio condition is unnecessary.
Example 2
Here, we consider a high-order Schrödinger-Poisson-Slater system:
in which Ω = (0,1)2,T = 1. g1,g2 and u0(X) are chosen correspondingly to the exact solutions
We solve this problem by the two linearized BDF2 schemes given in Section 2, with above P1 and Q2 elements. To show the convergence in L2 norm, we adopt the same mesh generation as Example 6.1. The numerical results are presented in Tables 5, 6, 7 and 8 at time t = 0.5 and 1.0. We can see that the errors in L2 norm are in line with the theoretical analysis. On the other hand, to verify unconditional stability of schemes, we also list numerical results at time t = 1.0. in Figs. 5, 6, 7 and 8 for each \(\tau =\frac {1}{5},\frac {1}{10},\frac {1}{20}\) with different mesh-grids. We can see that for a fixed τ, the errors in L2 norm converge to a small constant when the mesh refine gradually, which show that the two proposed schemes are unconditionally stable and the grid-ratio condition is unnecessary.
7 Conclusion
In this paper, we have presented two linearized BDF2 schemes with Galerkin finite elements approximation for the nonlinear Schrödinger-Helmholtz equations. Different from the existing second accurate (in time) numerical schemes for coupled equations, we derive optimal error estimates at the time step tn (instead of the time instant tn+ 1/2) for the proposed schemes without grid-ratio condition. At last, two numerical examples are provided to confirm the theoretical analysis. On the other hand, there are some interesting works on the variable-step BDF2 method for self-adaptive time stepping integrations for long-time simulations of phase field models, such as [46,47,48]. The analytic method in this paper can be extended to analyze other nonlinear physical models, such as the time-dependent nonlinear thermistor equations [35], Cahn-Hilliard equation [36], Keller-Segel system [49], and so forth.
References
Bohun, S., Illner, R., Lange, H., Zweifel, P.F.: Error estimates for Galerkin approximations to the periodic Schrödinger-Poisson system, ZAMMZ. Angew. Math. Mech 76(1), 7–13 (1996)
Harrison, R., Moroz, I., Tod, K.P.: A numerical study of the Schrödinger-Newton equations. Nonlinearity 16(1), 101–122 (2003)
Leo, M.D., Rial, D.: Well posedness and smoothing effect of Schrödinger-Poisson equation. J. Math. Phys 48, 093509 (2007)
Lubich, C.: On splitting methods for Schrödinger-Poisson and cubic nonlinear Schrödinger equations. Math. Comput 77, 2141–2153 (2008)
Masaki, S.: Energy solution to a Schrödinger-Poisson system in the two-dimensional whole space. SIAM J. Math. Anal 43(6), 2719–2731 (2011)
Antoine, X., Besse, C., Klein, P.: Absorbing boundary conditions for general nonlinear Schrödinger equations. SIAM J. Sci. Comput 33(2), 1008–1033 (2011)
Pelinovsky, D.E., Afanasjev, V.V., Kivshar, Y.S.: Nonlinear theory of oscillating, decaying, and collapsing solitons in the generalized nonlinear Schrödinger equation. Phys. Rev. E 53(2), 1940–1953 (1996)
Bao, W., Mauser, N.J., Stimming, H.P.: Effective one particle quantum dynamics of electrons: a numerical study of the Schrödinger-Poisson-Xα model. Commun. Math. Sci 1(4), 809–828 (2003)
Zhang, Y., Dong, X.: On the computation of ground state and dynamics of Schrödinger-Poisson-Slater system. J. Comput. Phys 230, 2660–2676 (2011)
Mauser, N.J.: The Schrödinger-Poisson-Xα equation. Appl. Math. Lett 14, 759–763 (2001)
Stimming, H.P.: The IVP for the Schrödinger-Poisson-Xα equation in one dimension. Math. Models Methods Appl. Sci 15(8), 1169–1180 (2005)
Cao, Y., Musslimani, Z.H., Titi, E.S.: Nonlinear Schrödinger-Helmholtz equation as numercal regularization of the nonlinear Schrödinger equation. Nonlinearity 21, 879–898 (2008)
Bao, W., Cai, Y.: Uniform error estimates of finite difference methods for the nonlinear Schrödinger equation with wave operator. SIAM J. Numer. Anal 50(2), 492–521 (2012)
Bratsos, A.G.: A modified numerical scheme for the cubic Schrödinger equation. Numer. Methods Partial Differ. Equ 27(3), 608–620 (2011)
Liao, H., Sun, Z., Shi, H.: Error estimate of fourth-order compact scheme for linear Schrödinger equations. SIAM J. Numer Anal 47(6), 4381–4401 (2010)
Zhang, L.: Convergence of a conservative difference schemes for a class of Klein-Gordon-Schrödinger equations in one space dimension. Appl. Math. Comput 163(1), 343–355 (2005)
Wang, T., Jiang, Y.: Point-wise errors of two conservative difference schemes for the Klein-Gordon-Schrödinger equation. Commun. Nonlinear Sci. Numer. Simulat 17(12), 4565–4575 (2012)
Wang, T., Zhao, X., Jiang, J.: Unconditional and optimal H2-error estimates of two linear and conservative finite difference schemes for the Klein-Gordon-Schrödinger equation in high dimensions. Adv. Comput. Math 44(5), 477–503 (2018)
Xiang, X.: Spectral method for solving the system of equations of Schrödinger-Klein-Gordon field. J. Comput. Appl. Math 21 (2), 161–171 (1988)
Kong, L., Wang, L., Jiang, S., Duan, Y.: Multisymplectic Fourier pseudo-spectral integrators for Klein-Gordon-Schrödinger equations. Sci. China Math 56, 915–932 (2013)
Hong, Q., Wang, Y., Wang, J.: Optimal error estimate of a linear Fourier pseudo-spectral scheme for two dimensional Klein-Gordon-Schrödinger equations. J. Math. Anal. Appl 468(2), 817–838 (2018)
Fu, Y., Cai, W., Wang, Y.: Structure-preserving algorithms for multi-dimensional fractional Klein-Gordon-Schrödinger equation. Appl. Numer. Math. 156, 77–93 (2020)
Wang, J.: A New Error Analysis of Crank-Nicolson Galerkin FEMs for a Generalized Nonlinear Schrödinger Equation. J. Sci. Comput 60, 390–407 (2014)
Shi, D., Wang, P., Zhao, Y.: Superconvergence analysis of anisotropic linear triangular finite element for nonlinear Schrödinger equation. Appl. Math. Lett 38, 129–134 (2014)
Shi, D., Wang, J.: Unconditional Superconvergence analysis of a Crank-Nicolson Galerkin FEM for nonlinear Schrödinger equation. J. Sci. Comput 72 (3), 1093–1118 (2017)
Zhang, H., Wang, J.: Superconvergence analysis of Crank-Nicolson Galerkin FEMs for a generalized nonlinear Schrödinger equation. Numer. Methods Partial Differ. Equ 34(2), 799–820 (2018)
Zhang, H., Shi, D., Li, Q.: Nonconforming finite element method for a generalized nonlinear Schrödinger equation. Appl. Math. Comput 377, 125141 (2020)
Cai, W., He, D., Pan, K.: A linearized energy-conservative finite element method for the nonlinear Schrödinger equation with wave operator. Appl. Numer. Math 140, 183–198 (2019)
Li, M., Shi, D., Wang, J.: Unconditional superconvergence analysis of the conservative linearized Galerkin FEMs for nonlinear Klein-Gordon-Schrödinger equation. Appl. Numer. Math 142, 47–63 (2019)
Yang, Y., Jiang, Y.: Unconditional optimal error estimates of linearized backward Euler Galerkin FEMs for nonlinear Schrödinger-Helmholtz equations. Numer. Algor 86(4), 1495–1522 (2021)
Wang, J.: Unconditional stability and convergence of Crank-Nicolson Galerkin FEMs for a nonlinear Schrödinger-Helmholtz system. Numer. Math 139, 479–503 (2018)
Li, D., Wang, J.: Unconditionally optimal error analysis of Crank-Nicolson Galerkin FEMs for a strongly nonlinear Parabolic system. J. Sci. Comput 72, 892–915 (2017)
Sun, W., Wang, J.: Optimal error analysis of Crank-Nicolson schemes for a coupled nonlinear Schrödinger system in 3D. J. Comput. Appl. Math 317, 685–699 (2017)
Shi, D., Li, C.: Superconvergence analysis of two-grid methods for bacteria equations. Numer. Algor 86(1), 123–152 (2021)
Li, B., Gao, H., Sun, W.: Unconditionally optimal error estimates of a Crank-Nicolson Galerkin method for the nonlinear Thermistor equations. SIAM J. Numer. Anal 52(2), 933–954 (2014)
Amanda, E., Wang, C., Steven, M.: Stability and convergence of a second order mixed finite element method for the Cahn-Hilliard Equation. IMA J. Numer. Anal 36, 1867–1897 (2016)
Shi, D., Yan, F., Wang, J.: Unconditional superconvergence analysis of a new mixed finite element method for nonlinear Sobolev equation. Appl. Math. Comput 274, 182–194 (2016)
Gear, C.: Numerical Initial Value Problems in Ordinary Differential Equations. Prentice-Hall, Englewood Cliffs (1971)
Lambert, J.: Numerical Methods in Ordinary Differential Systems: the Initial Value Problems. Wiley, Chichester (1991)
Gao, H.: Unconditional optimal error estimates of BDF-Galerkin FEMs for nonlinear thermistor equations. J. Sci. Comput 66, 504–527 (2016)
Li, B., Sun, W.: Error analysis of linearized semi-implicit Galerkin finite element methods for nonlinear parabolic equations. Int. J. Numer. Anal. Model 10, 622–633 (2013)
Li, B., Sun, W.: Unconditional convergence and optimal error estimates of a Galerkin-mixed FEM for incompressible miscible flow in porous media. SIAM J. Numer. Anal 51, 1959–1977 (2013)
Thomée, V.: Galerkin Finite Element Methods for Parabolic Problems. Springer Series in Computational Mathematics, Sweden (2000)
Lin, Q., Lin, J.: Finite element methods: Accuracy and improvement. Science Press, Beijing (2006)
Evans, L.C.: Partial Differential Equations, 2nd edn. AMS, Providence (2010)
Liao, H., Song, X., Tang, T., et al.: Analysis of the second order BDF scheme with variable steps for the molecular beam epitaxial model without slope selection. Sci. China Math 64, 887–902 (2021)
Liao, H., Ji, B., Zhang, L.: An adaptive BDF2 implicit time-stepping method for the phase field crystal model. IMA J. Numer. Anal. in review arXiv:2008.00212v1 (2020)
Liao, H., Tang, T., Zhou, T.: On energy stable, maximum-principle preserving, second-order BDF scheme with variable steps for the Allen–Cahn equation. SIAM J. Numer. Anal 58(4), 2294–2314 (2020)
Zhou, G., Saito, N.: Finite volume methods for a Keller-Segel system: Discrete energy, error estimates and numerical blow-up analysis. Numer. Math 135, 265–311 (2017)
Funding
This work was supported by the National Natural Science Foundation of China (No. 12071443); the Key Scientific Research Projects of Henan Colleges and Universities (No. 20B110013).
Author information
Authors and Affiliations
Corresponding author
Ethics declarations
Conflict of interest
The authors declare no competing interests.
Additional information
Data availability
Data sharing not applicable to this article as no datasets were generated or analyzed during the current study.
Publisher’s note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Rights and permissions
About this article
Cite this article
Shi, D., Zhang, H. Unconditional error estimates of linearized BDF2-Galerkin FEMs for nonlinear coupled Schrödinger-Helmholtz equations. Numer Algor 92, 1679–1705 (2023). https://doi.org/10.1007/s11075-022-01360-5
Received:
Accepted:
Published:
Issue Date:
DOI: https://doi.org/10.1007/s11075-022-01360-5