
A new nonlinear fourth-order sin-Gordon equation with actual physical background is first created. Then, by introducing an auxiliary function, the nonlinear fourth-order sin-Gordon equation is decomposed into the nonlinear system of equations of second-order derivatives of spatial variables. Subsequently, the time derivative is discretized by using the Crank-Nicolson (CN) scheme to construct a new time semi-discretized mixed CN (TSDMCN) scheme. Thereafter, the spatial variables in the TSDMCN scheme are discretized by using a two-grid mixed finite element (MFE) method to construct a new two-grid CN MFE (TGCNMFE) method with unconditional stability and second-order time precision, which consists of a system of nonlinear MFE equations defined on coarser grids and a system of linear MFE equations defined on finer grids with sufficiently high precision, and is very easy to solve. The existence, stability, and error estimates of the TSDMCN and TGCNMFE solutions are strictly proved theoretically, and the superiorities of the TGCNMFE method and the correctness of theoretical results are verified by two sets of numerical experiments.
Citation: Yanjie Zhou, Xianxiang Leng, Yuejie Li, Qiuxiang Deng, Zhendong Luo. A novel two-grid Crank-Nicolson mixed finite element method for nonlinear fourth-order sin-Gordon equation[J]. AIMS Mathematics, 2024, 9(11): 31470-31494. doi: 10.3934/math.20241515
[1] | Yuelong Tang . Error estimates of mixed finite elements combined with Crank-Nicolson scheme for parabolic control problems. AIMS Mathematics, 2023, 8(5): 12506-12519. doi: 10.3934/math.2023628 |
[2] | Emadidin Gahalla Mohmed Elmahdi, Jianfei Huang . Two linearized finite difference schemes for time fractional nonlinear diffusion-wave equations with fourth order derivative. AIMS Mathematics, 2021, 6(6): 6356-6376. doi: 10.3934/math.2021373 |
[3] | Yangfang Deng, Zhifeng Weng . Barycentric interpolation collocation method based on Crank-Nicolson scheme for the Allen-Cahn equation. AIMS Mathematics, 2021, 6(4): 3857-3873. doi: 10.3934/math.2021229 |
[4] | Eunjung Lee, Dojin Kim . Stability analysis of the implicit finite difference schemes for nonlinear Schrödinger equation. AIMS Mathematics, 2022, 7(9): 16349-16365. doi: 10.3934/math.2022893 |
[5] | Haoran Sun, Siyu Huang, Mingyang Zhou, Yilun Li, Zhifeng Weng . A numerical investigation of nonlinear Schrödinger equation using barycentric interpolation collocation method. AIMS Mathematics, 2023, 8(1): 361-381. doi: 10.3934/math.2023017 |
[6] | Abdul-Majeed Ayebire, Saroj Sahani, Priyanka, Shelly Arora . Numerical study of soliton behavior of generalised Kuramoto-Sivashinsky type equations with Hermite splines. AIMS Mathematics, 2025, 10(2): 2098-2130. doi: 10.3934/math.2025099 |
[7] | Ajmal Ali, Tayyaba Akram, Azhar Iqbal, Poom Kumam, Thana Sutthibutpong . A numerical approach for 2D time-fractional diffusion damped wave model. AIMS Mathematics, 2023, 8(4): 8249-8273. doi: 10.3934/math.2023416 |
[8] | Haifa Bin Jebreen . On the numerical solution of Fisher's equation by an efficient algorithm based on multiwavelets. AIMS Mathematics, 2021, 6(3): 2369-2384. doi: 10.3934/math.2021144 |
[9] | Zhichao Fang, Ruixia Du, Hong Li, Yang Liu . A two-grid mixed finite volume element method for nonlinear time fractional reaction-diffusion equations. AIMS Mathematics, 2022, 7(2): 1941-1970. doi: 10.3934/math.2022112 |
[10] | Huiqin Zhang, Yanping Chen . Two-grid finite element method with an H2N2 interpolation for two-dimensional nonlinear fractional multi-term mixed sub-diffusion and diffusion wave equation. AIMS Mathematics, 2024, 9(1): 160-177. doi: 10.3934/math.2024010 |
A new nonlinear fourth-order sin-Gordon equation with actual physical background is first created. Then, by introducing an auxiliary function, the nonlinear fourth-order sin-Gordon equation is decomposed into the nonlinear system of equations of second-order derivatives of spatial variables. Subsequently, the time derivative is discretized by using the Crank-Nicolson (CN) scheme to construct a new time semi-discretized mixed CN (TSDMCN) scheme. Thereafter, the spatial variables in the TSDMCN scheme are discretized by using a two-grid mixed finite element (MFE) method to construct a new two-grid CN MFE (TGCNMFE) method with unconditional stability and second-order time precision, which consists of a system of nonlinear MFE equations defined on coarser grids and a system of linear MFE equations defined on finer grids with sufficiently high precision, and is very easy to solve. The existence, stability, and error estimates of the TSDMCN and TGCNMFE solutions are strictly proved theoretically, and the superiorities of the TGCNMFE method and the correctness of theoretical results are verified by two sets of numerical experiments.
Let Ω⊂Rd (d=2,3) be a connected bounded domain with the boundary ∂Ω. We propose a new nonlinear fourth-order sin-Gordon (NFOSG) equation, which is different from the standard sin-Gordon equation in [1,2] and is defined as follows:
Problem 1. Find ϖ:[0,te]→C4(ˉΩ), satisfying
{ϖtt(x,t)−Δϖ(x,t)+Δ2ϖ(x,t)+sin(ϖ(x,t))=0,(x,t)∈Ω×(0,te),ϖ(x,t)=Δϖ(x,t)=0,(x,t)∈∂Ω×(0,te),ϖ(x,0)=ϖ0(x),ϖt(x,0)=ϖ1(x),x∈Ω, | (1.1) |
where te is a specified time upper bound,
ϖtt(x,t)=∂2ϖ(x,t)/∂t2,x=(x1,x2,⋯,xd),Δ=d∑i=1∂2/∂x2i |
is the Laplace operator, both ϖ0(x) and ϖ1(x) are sufficiently smooth known initial functions, and
ϖt(x,t)=∂ϖ(x,t)/∂t. |
The NFOSG equation is obtained by adding a fourth-order term Δ2ϖ(x,t) to the standard sin-Gordon equation. It is well known (see, e.g., [3,4,5]) that the fourth-order term Δ2ϖ(x,t) can describe the deformation of an object under the action of an external force such as seismic waves and flying aircraft, while the nonlinear term sin(ω(x,t)) can describe the nonlinear behavior of a moving body/wave. Therefore, the NFOSG equation can not only describe the fluid mechanics of porous media [6,7,8], groundwater dynamics [9], seepage mechanics and groundwater hydraulics [10,11,12], etc., like the standard sin-Gordon equation, but also describe the nonlinear deformation behavior of moving bodies/waves, such as the movement of seismic waves or the movement of flying aircraft. Therefore, the NFOSG equation is an important mathematical physics model with a real application background and has a wider range of applications than the standard sin-Gordon equation.
However, as a result of the NFOSG equation including a strong nonlinear term sin(ω(x,t)), it is very difficult to solve analytically. A numerical method is the most effective choice to solve the NFOSG equation. Fortunately, the numerical methods of the NFOSG equation play a pivotal role in simulation of the earthquake wave, the motion of flying aircraft, as well as the fluid mechanics of porous media, groundwater dynamics, seepage mechanics, groundwater hydraulics, and other phenomena. Therefore, it is very important to study the numerical methods of the NFOSG equation.
A large number of numerical examples have showed that the two-grid finite element (FE) algorithm is one of the best numerical methods for solving nonlinear partial differential equations (PDEs), which consists of a nonlinear FE system of equations defined on coarser grids and a linear FE system of equations defined on finer grids with sufficiently high precision. Hence, it can simplify computation and enhance calculation efficiency. It was originally used to solve quasi-linear elliptic equations [13]. More recently, Shi's and Liu's teams have used it to solve some of the more complex nonlinear PDEs (see [14,15,16]).
However, to our knowledge, at the moment, there has been no report on the Crank-Nicolson (CN) mixed FE (MFE) (CNMFE) format for the NFOSG equation reduced with two-grid FE technique. Therefore, the main task of this paper is to develop a new two-grid CNMFE (TGCNMFE) method for the NFOSG equation (i.e., Problem 1). The TGCNMFE method has at least three benefits. First, by introducing an auxiliary function φ=Δϖ, the NFOSG equation is split into two second-order equations, which can be easily solved by lower degree FEs such as linear or quadratic FEs. Second, the TGCNMFE method has unconditional stability and second-order time accuracy. Third, the TGCNMFE method also consists of a nonlinear MFE equation defined on coarser grids and a linear MFE equation defined on finer grids, which greatly simplifies the calculation and improves the calculation efficiency.
Although some single-grid FE methods for the standard sin-Gordon equation with only second-order derivatives of spatial variables have been proposed in [10,11,12], they are completely different from the TGCNMFE method for the NFOSG equation with fourth-order derivatives of spatial variables. Both the TGCNMFE method and the NFOSG equation are far more complex than those in [10,11,12]. Therefore, both the establishment of the TGCNMFE method and the theoretical analysis of existence, stability, and errors of the TGCNMFE solutions herein are more difficult and require more skills than the single-grid FE methods in [10,11,12], but the NFOSG equation has a wider range of applications than the standard sin-Gordon equation, as mentioned above. Hence, it is very valuable to study the TGCNMFE method for solving the NFOSG equation.
The rest of this paper consists of the following four sections. In Section 2, by introducing an auxiliary function φ=Δϖ, we split the NFOSG equation into two second-order equations, and by using the time CN scheme to discretize them, we design a new time semi-discretization mixed CN (TSDMCN) scheme and discuss the existence, stability, and error estimates of the TSDMCN solutions. In Section 3, by using the two-grid MFE method to discretize the TSDMCN scheme, we construct a new TGCNMFE method for the NFOSG equation and analyze the existence, stability, and errors of the TGCNMFE solutions. In Section 4, we use two sets of numerical experiments to confirm the correctness of the obtained theoretical results and show the superiorities of the TGCNMFE method. Finally, we give the main conclusions of this article and future study prospects in Section 5.
The Sobolev spaces and norms used in this context are classical (see [5,17,18]). Let
W=H10(Ω)andφ=−Δϖ. |
Thus, by Green's formula, we can derive the following mixed variational format of Problem 1.
Problem 2. For any t∈(0,te), find (ϖ,φ)∈W×W, meeting
{(ϖtt,υ)+(∇φ,∇υ)+(φ,υ)=(f(ϖ),υ),∀υ∈W,(∇ϖ,∇ϑ)=(φ,ϑ),∀ϑ∈W,ϖ(x,0)=ϖ0(x),ϖt(x,0)=ϖ1(x),φ(x,0)=−Δϖ0(x),x∈Ω, | (2.1) |
where (⋅,⋅) denotes the L2 inner product, and
f(ϖ)=−sin(ϖ). |
Using the proof method of Theorem 3.3.1 in [5] or the proof method of the following Theorem 1, we can demonstrate the existence and stability of the generalized solutions for Problem 2.
In order to establish the TGCNMFE method, we first set up a new TSDMCN scheme. Therefore, we assume that K>0 is an integer,
Δt=te/K |
is the time step, and ϖk and φk are the approximations to ϖ(x,t) and φ(x,t) at
tk=kΔt(0⩽k⩽K), |
respectively.
Using an implicit scheme of time to discretize the first equation in Problem 2 yields
1Δt2(ϖk+1−2ϖk+ϖk−1,υ)+(∇φk+1,∇υ)+(φk+1,υ)=(f(ϖk+1),υ),∀υ∈W. | (2.2) |
Using an explicit scheme of time to discretize the first equation in Problem 2 yields
1Δt2(ϖk+1−2ϖk+ϖk−1,υ)+(∇φk−1,∇υ)+(φk−1,υ)=(f(ϖk−1),υ),∀υ∈W. | (2.3) |
By adding formulas (2.2) and (2.3), we obtain the following brand-new TSDMCN format, which is different from the existing time semi-discrete formats, including that in [1].
Problem 3. Find {ϖk,φk}∈W×W (1⩽k⩽K), meeting
{1Δt2(ϖk+1−2ϖk+ϖk−1,υ)+12(∇(φk+1+φk−1),∇υ)+12(φk+1+φk−1,υ)=12(f(ϖk+1)+f(ϖk−1),υ),∀υ∈W,1⩽k⩽K−1,(∇ϖk,∇ϑ)=(φk,ϑ),∀ϑ∈W,0⩽k⩽K,ϖ0=ϖ0,ϖ1=ϖ0+Δtϖ1,φ0=−Δϖ0,φ1=φ0−ΔtΔϖ1,inΩ. | (2.4) |
For Problem 3, we obtain the following results:
Theorem 1. Problem 3 has a unique series of solutions
{ϖk,φk}Kk=1⊂W×W |
meeting the following boundness, i.e., stability:
‖∇ϖk‖0+‖∇φk‖0⩽c(‖ϖ0‖1+‖ϖ1‖1),1⩽k⩽K. | (2.5) |
The c used in this context is a generical positive constant independent of Δt. And, when ϖ0(x) and ϖ1(x) are adequately smooth, the series of solutions {ϖk,φk}Kk=1 meets the error estimates
‖∇(ϖ(tk)−ϖk)‖0+‖∇(φ(tk)−φk)‖0⩽cΔt2,1⩽k⩽K, | (2.6) |
where
ϖ(tk)=ϖ(x,tk)andφ(tk)=φ(x,tk)(1⩽k⩽K). |
Proof. The demonstration of Theorem 1 consists of the following three parts.
(1) The existence and uniqueness of series of solutions of Problem 3.
Taking
υ=ϖk+1−ϖk−1 |
in the first subsystem of equations of (2.4), by the second subsystem of equations of (2.4), the Hölder and Cauchy inequalities, and the Lagrange differential mean formula (LDMF), we obtain
‖ϖk+1−ϖk‖20−‖ϖk−ϖk−1‖20+Δt22(‖φk+1‖20−‖φk−1‖20)+Δt22(‖∇ϖk+1‖20−‖∇ϖk−1‖20)=Δt22(f(ϖk+1)+f(ϖk−1),ϖk+1−ϖk−1)⩽cΔt2(‖ϖk+1−ϖk‖20+‖ϖk−ϖk−1‖20),1⩽k⩽K−1. | (2.7) |
By summing (2.7) from 1 to k (k⩽K−1) and using the third equation of (2.4), we obtain
‖ϖk+1−ϖk‖20+Δt22(‖φk+1‖20+‖φk‖20)+Δt22(‖∇ϖk+1‖20+‖∇ϖk‖20)⩽‖ϖ1−ϖ0‖20+Δt22(‖φ1‖20+‖φ0‖20)+Δt22(‖∇ϖ1‖20+‖∇ϖ0‖20)+cΔt2k∑j=0‖ϖj+1−ϖj‖20⩽cΔt2(‖ϖ0‖21+‖ϖ1‖21)+cΔt2k∑j=0‖ϖj+1−ϖj‖20,1⩽k⩽K−1. | (2.8) |
When Δt is adequately small satisfying cΔt2⩽1/4, by simplifying (2.8), we obtain
‖ϖk+1−ϖk‖20+Δt2‖φk+1‖20+Δt2‖∇ϖk+1‖20⩽cΔt2(‖ϖ0‖21+‖ϖ1‖21)+cΔt2k−1∑j=0‖ϖj+1−ϖj‖20,1⩽k⩽K−1. | (2.9) |
By applying Gronwall's inequality (see [5, Lemme 3.1.9]) to (2.9), we obtain
‖ϖk+1−ϖk‖20+Δt2‖φk+1‖20+Δt2‖∇ϖk+1‖20⩽cΔt2(‖ϖ0‖21+‖ϖ1‖21)exp(ckΔt2)⩽cΔt2(‖ϖ0‖21+‖ϖ1‖21),1⩽k⩽K−1. | (2.10) |
Thereupon, we obtain
‖φk‖0+‖∇ϖk‖0⩽c(‖ϖ0‖1+‖ϖ1‖1),1⩽k⩽K. | (2.11) |
Further, by the first and second subsystems of equations of (2.4), the Hölder and Cauchy inequalities, and LDMF, we obtain
‖∇(ϖk+1−ϖk)‖20−‖∇(ϖk−ϖk−1)‖20+Δt22(‖∇φk+1‖20−‖∇φk−1‖20)+Δt22(‖φk+1‖20−‖φk−1‖20)=(ϖk+1−2ϖk+ϖk−1,φk+1−φk−1)+Δt22(∇(φk+1+φk−1),∇(φk+1−φk−1))+Δt22(φk+1+φk−1,φk+1−φk−1)=Δt22(f(ϖk+1)+f(ϖk−1),φk+1−φk−1)⩽cΔt(‖ϖk+1−ϖk‖20+‖ϖk−ϖk−1‖20)+cΔt3(‖φk+1‖20+‖φk−1‖20),1⩽k⩽K−1. | (2.12) |
By summing (2.12) from 1 to k (k⩽K−1) and using the third equation of (2.4), we obtain
‖∇(ϖk+1−ϖk)‖20+Δt22(‖∇φk+1‖20+‖∇φk‖20)+Δt22(‖φk+1‖20+‖φk‖20)⩽cΔtk∑j=0‖ϖj+1−ϖk‖20+cΔt3k∑j=0‖φj+1‖21+cΔt2(‖ϖ0‖21+‖ϖ1‖21),1⩽k⩽K−1. | (2.13) |
When Δt is adequately small satisfying cΔt2⩽1/4, by simplifying (2.13), we obtain
‖∇(ϖk+1−ϖk)‖20+Δt2‖∇φk+1‖20+Δt2‖φk+1‖20⩽cΔtk−1∑j=0‖ϖj+1−ϖk‖20+cΔt3k−1∑j=0‖φj+1‖21+cΔt2(‖ϖ0‖21+‖ϖ1‖21),1⩽k⩽K−1. | (2.14) |
By applying Gronwall's inequality (see [5, Lemme 3.1.9]) to (2.14), we obtain
‖∇(ϖk+1−ϖk)‖20+Δt2‖∇φk+1‖20+Δt2‖φk+1‖20⩽cΔt2(‖ϖ0‖21+‖ϖ1‖21)exp(ckΔt),1⩽k⩽K−1. | (2.15) |
Thereupon, we obtain
‖∇φk+1‖0+‖φk+1‖0⩽c(‖ϖ0‖1+‖ϖ1‖1),1⩽k⩽K−1. | (2.16) |
Thus, when
ϖ0=ϖ1=0, |
by (2.11) and (2.16), we obtain
φk=ϖk=0. |
This signifies that Problem 3 has at least one series of solutions {ϖk,φk}Nk=1.
If Problem 3 has another series of solutions {˜ϖk,˜φk}Nk=1, it should satisfy the following system of equations:
{1Δt(˜ϖk+1−2˜ϖk+˜ϖk−1,υ)+12(∇(˜φk+1+˜φk−1),∇υ)+12(˜φk+1+˜φk−1,υ)=12(f(˜ϖk+1)+f(˜ϖk−1),υ),∀υ∈W,1⩽k⩽K−1,(∇˜ϖk,∇ϑ)=(˜φk,ϑ),∀ϑ∈W,0⩽k⩽K,˜ϖ0=ϖ0,˜φ0=−Δϖ0,˜ϖ1=ϖ1,˜φ1=φ0−ΔtΔϖ1,inΩ. | (2.17) |
Let
Ek=ϖk−˜ϖkandek=φk−˜φk. |
Subtracting (2.17) from (2.4) yields
{1Δt(Ek+1−2Ek+Ek−1,υ)+12(∇(ek+1+ek−1),∇υ)+12(ek+1+ek−1,υ)=12(f(ϖk+1)+f(ϖk−1)−f(˜ϖk+1)−f(˜ϖk−1),υ),∀υ∈W,1⩽k⩽K−1,(∇Ek,∇ϑ)=(ek,ϑ),∀ϑ∈W,0⩽k⩽K,E0=E1=0,e0=e1=0,inΩ. | (2.18) |
Taking
υ=Ek−Ek−1 |
in the first equation of (2.18), by the second equation of (2.18), the Hölder and Cauchy inequalities, and the LDMF, we obtain
2(‖Ek+1−Ek‖20−‖Ek−Ek−1‖20)+Δt2(‖ek+1‖20−‖ek−1‖20)+Δt2(‖∇Ek+1‖20−‖∇Ek−1‖20)=Δt2(f(ϖk+1)−f(˜ϖk+1)+f(ϖk−1)−f(˜ϖk−1),Ek+1−Ek−1)=Δt2(Ek+1f′(ξk)+Ek−1f′(ξk−1),Ek+1−Ek−1)⩽cΔt2(‖Ek+1−Ek‖20+‖Ek−Ek−1‖20),1⩽k⩽K−1, | (2.19) |
where ξi lies between ϖi and ˜wi (i=k,k−1).
Summing (2.19) from 1 to k (k⩽K−1) and noting that
E1=e1=E0=e0=0, |
we obtain
2‖Ek+1−Ek‖20+Δt2(‖ek+1‖20+‖ek‖20)+Δt2(‖∇Ek+1‖20+‖∇Ek‖20)⩽cΔt2k∑i=0‖Ei+1−Ei‖20,1⩽k⩽K−1. | (2.20) |
When Δt is adequately small, satisfying cΔt2⩽1, by simplifying (2.20), we obtain
‖Ek+1−Ek‖20+Δt2‖ek+1‖20+Δt2‖∇Ek+1‖20⩽cΔt2k−1∑i=0‖Ei+1−Ei‖20,1⩽k⩽K−1. | (2.21) |
By applying Gronwall's inequality to (2.21), we obtain
‖Ek+1−Ek‖20+Δt2‖ek+1‖20+Δt2‖∇Ek+1‖20⩽0,1⩽k⩽K−1. | (2.22) |
Thereupon, we obtain
˜ϖk=˜ϖkandφk=φk(1⩽k⩽K). |
Hence, Problem 3 has a unique series of solutions {ϖk,φk}Kk=1.
(2) Discuss the boundness, i.e., stability of solutions {ϖk,φk}Kk=1 of Problem 3.
When Problem 3 has a unique series of {ϖk,φk}Kk=1, by (2.11) and (2.16), we claim that it is bounded, i.e., stable, namely (2.5) holds.
(3) Estimate the errors of solutions {ϖk,φk}Kk=1 of Problem 3.
Via the Taylor expansion, we obtain
v(tk+1)=v(tk)+Δtv′(tk)+Δt22v″(tk)+Δt36v‴(tk)+Δt424v‴(tk)+⋯, | (2.23) |
v(tk−1)=v(tk)−Δtv′(tk)+Δt22v″(tk)−Δt36v‴(tk)+Δt424v‴(tk)+⋯, | (2.24) |
f(v(tk+1))=f(v(tk))+[v(tk+1)−v(tk)]f′(v(tk))+12(v(tk+1)−v(tk))2f″(v(tk))+⋯, | (2.25) |
f(v(tk−1))=f(v(tk))+[v(tk−1)−v(tk)]f′(v(tk))+12(v(tk−1)−v(tk))2f″(v(tk))+⋯. | (2.26) |
Therefore, we obtain
ϖ″(tk)=ϖ(tk+1)−2ϖ(tk)+ϖ(tk−1)Δt2+Δt224ϖ‴(ηk),tk−1⩽ζk⩽tk+1, | (2.27) |
φ(tk)=φ(tk+1)+φ(tk−1)2−Δt2φ″(ςk),tk−1⩽ςk⩽tk+1, | (2.28) |
f(ϖ(tk))=f(ϖ(tk+1))+f(ϖ(tk−1))2−Δt2R(x,t), | (2.29) |
where R(x,t) is a bounded remainder function, which is obtained by (2.23)–(2.26).
Thereupon, subtracting the first equation of (2.4) from the first equation of (2.1) after taking t=tk and setting
ϱk=ϖ(x,tk)−ϖkand˜ϱk=φ(x,tk)−φk, |
we obtain the following system of error equations:
1Δt2(ϱk+1−2ϱk+ϱk−1,υ)+12(∇(˜ϱk+1+˜ϱk−1),∇υ)+12(˜ϱk+1+˜ϱk−1,υ)=12(f(ϖ(tk+1))−f(ϖk+1)+f(ϖ(tk−1))−f(ϖk−1),υ)+Δt2(R(x,t),υ),∀υ∈W,1⩽k⩽K−1, | (2.30) |
(∇ϱk,∇ϑ)=(˜ϱk,ϑ),∀ϑ∈W,0⩽k⩽K, | (2.31) |
ϱ0=ϱ1=˜ϱ0=˜ϱ1=0, | (2.32) |
where ˜R(x,t) is also a bounded remainder function, which is determined by substituting (2.27)–(2.29) into (2.1).
Taking
υ=ϱk+1−ϱk−1 |
in (2.30) and using Eq (2.31), the Hölder and Cauchy inequalities, Green's formula, and the LDMF, we obtain
‖ϱk+1−ϱk‖20−‖ϱk−ϱk−1‖20+Δt22(‖˜ϱk+1‖20−‖˜ϱk−1‖20)+Δt22(‖∇ϱk+1‖20−‖∇ϱk−1‖20)=Δt22(f(ϖ(tk+1))−f(ϖk+1)+f(ϖ(tk−1))−f(ϖk−1),ϱk+1−ϱk−1)+Δt4(R(x,t),ϱk+1−ϱk−1)⩽cΔt(‖ϱk+1−ϱk‖20+‖ϱk−ϱk−1‖20)+cΔt7,1⩽k⩽K−1. | (2.33) |
Summing (2.33) from 1 to k (k⩽K−1), and noting that
ϱ0=ϱ1=˜ϱ0=˜ϱ1=0, |
we get
‖ϱk+1−ϱk‖20+Δt22(‖˜ϱk+1‖20+‖˜ϱk‖20)+Δt22(‖∇ϱk+1‖20+‖∇ϱk‖20)⩽cΔtk∑i=0‖ϱi+1−ϱi‖20+ckΔt7,1⩽k⩽K−1. | (2.34) |
Thus, when Δt is adequately small, satisfying cΔt⩽1/4, by simplifying (2.34), we get
‖ϱk+1−ϱk‖20+Δt2‖˜ϱk+1‖20+Δt2‖∇ϱk+1‖20⩽cΔtk−1∑i=0‖ϱi+1−ϱi‖20+cΔt6,1⩽k⩽K−1. | (2.35) |
By applying Gronwall's inequality to (2.35), we obtain
‖ϱk+1−ϱk‖20+Δt2‖˜ϱk+1‖20+Δt2‖∇ϱk+1‖20⩽cΔt6exp(ckΔt)⩽cΔt6,1⩽k⩽K−1. | (2.36) |
Thereupon, we obtain
‖φ(tk)−φk‖0+‖∇(ϖ(tk)−ϖk)‖0⩽cΔt2,1⩽k⩽K. | (2.37) |
Further, by (2.30) and (2.31), the Hölder and Cauchy inequalities, and the LDMF, we obtain
2(‖∇(˜ρk+1−˜ρk)‖20−‖∇(˜ρk−˜ρk−1)‖20)+Δt2(‖∇˜ρk+1‖20−‖∇˜ρk−1‖20)+Δt2(‖˜ρk+1‖20−‖˜ρk−1‖20)=(ρk+1−2ρk+ρk−1,˜ρk+1−˜ρk−1)+Δt2(∇(˜ρk+1+˜ρk−1),∇(˜ρk+1−˜ρk−1))+Δt2(˜ρk+1+˜ρk−1,˜ρk+1−˜ρk−1)=Δt2(f(ϖ(tk+1))−f(ϖk+1)+f(ϖ(tk−1))−f(ϖk−1),˜ρk+1−˜ρk−1)+Δt4(R(x,t),˜ρk+1−˜ρk−1)⩽cΔt(‖∇(˜ρk+1−˜ρk)‖20+‖∇(˜ρk−˜ρk−1)‖20)+Δt3(‖˜ρk+1‖20+‖˜ρk−1‖20)+cΔt7,1⩽k⩽K−1. | (2.38) |
By summing (2.38) from 1 to k (k⩽K−1) and noting that
ϱ0=ϱ1=˜ϱ0=˜ϱ1=0, |
we obtain
2‖∇(˜ρk+1−˜ρk)‖20+Δt2(‖∇˜ρk+1‖20+‖∇˜ρk‖20)+Δt2(‖˜ρk+1‖20+‖˜ρk‖20)⩽cΔtk∑j=0‖∇(˜ρj+1−˜ρj)‖20+cΔt3k∑j=0‖˜ρj+1‖20+ckΔt7,1⩽k⩽K−1. | (2.39) |
Thus, when Δt is adequately small, satisfying cΔt⩽1/2, by simplifying (2.39), we get
‖∇(˜ρk+1−˜ρk)‖20+Δt2‖∇˜ρk+1‖20+Δt2‖˜ρk+1‖20⩽cΔtk−1∑j=0‖∇(˜ρj+1−˜ρj)‖20+cΔt3k−1∑j=0‖˜ρj+1‖20+cΔt6,1⩽k⩽K−1. | (2.40) |
By applying Gronwall's inequality to (2.40), we obtain
‖∇(˜ρk+1−˜ρk)‖20+Δt2‖∇˜ρk+1‖20+Δt2‖˜ρk+1‖20⩽cΔt6exp(ckΔt)⩽cΔt6,1⩽k⩽K−1. | (2.41) |
Thereupon, we obtain
‖∇(φ(tk)−φk)‖0+‖φ(tk)−φk‖0⩽cΔt4,1⩽k⩽K−1. | (2.42) |
Combining (2.37) with (2.42) yields (2.6). Theorem 1 is proved.
Remark 1. Theorem 1 shows that the TSDMCN solutions are stable and their H1 norm error estimates can reach second-order accuracy, which is the optimal order error estimates.
In order to construct the TGCNMFE format, it is necessary further to discretize the spatial variables in Problem 3 by using the two-grid MFE method. To this end, we assume that ℑH is a coarse grid of quasi-uniform partition on ˉΩ, which is formed by two-dimensional triangles or quadrangles and three-dimensional tetrahedrons or hexahedrons, and H denotes the maximum diameter of all elements in ℑH. Thus, the FE space defined on the coarse grids is expressed by
WH={vH∈C(ˉΩ)∩W:vH|E∈Pl(E),∀E∈ℑH}, |
where Pl(E) (l⩾1) denotes the space of polynomials with degree ⩽l defined on the coarse grid element E∈ℑH.
Further, we assume that ℑh is a fine grid of quasi-uniform partition on ˉΩ and h denotes the maximum diameter of all elements in ℑh (h≪H). Likewise, the FE space defined on the fine grids ℑh is denoted by
Wh={vh∈C(ˉΩ)∩W:vh|e∈Pl(e),∀e∈ℑh}. |
Thereupon, a new TGCNMFE formulation can be created as follows.
Problem 4. Step 1. Find
(ϖkH,φkH)∈WH×WH(1⩽k⩽K) |
defined on the coarse grid ℑH, satisfying the nonlinear system of equations:
{(ϖk+1H−2ϖkH+ϖk−1H,υH)+Δt22(∇(φk+1H+φk−1H),∇υH)+Δt22(φk+1H+φk−1H,υH)=Δt2(f(ϖk+1H)+f(ϖk−1H),υH),∀υH∈WH,1⩽k⩽K−1,(∇ϖkH,∇ϑH)=(φkH,ϑH),∀ϑH∈WH,0⩽k⩽K,ϖ0H=RHϖ0,ϖ1H=RHϖ1,φ0H=RHφ0,φ1H=RHφ1,inΩ. | (3.1) |
Step 2. Find
(ϖkh,φkh)∈Wh×Wh(1⩽k⩽K) |
defined on the fine grid ℑh, satisfying the linear system of equations:
{(ϖk+1h−2ϖkh+ϖk−1h,υh)+Δt2(∇(φk+1h+φk−1h),∇υh)+Δt2(φk+1h+φk−1h,υh)=Δt2(f(ϖk+1H)+f′(ϖk+1H)(ϖk+1h−ϖk+1H)+f(ϖk−1h),υh),∀υh∈Wh,1⩽k⩽K−1,(∇ϖkh,∇ϑh)=(φkh,ϑh),∀ϑh∈Wh,0⩽k⩽K,ϖ0h=Rhϖ0,ϖ1h=Rhϖ1,φ0h=Rhφ0,φ1h=Rhφ1,inΩ. | (3.2) |
The above operators Rδ: W→Wδ (δ=H,h) denote the Ritz projection; i.e., for any φ∈W, there exist two unique Rδφ∈Wδ satisfying
(∇(φ−Rδφ),∇ϑδ)=0,∀ϑδ∈Wδ,δ=H,h, | (3.3) |
and the following error estimates:
|φ−Rδφ|r⩽Cδl+1−r,ifφ∈W∩Hl+1(Ω),δ=H,h,r=−1,0,1. | (3.4) |
For Problem 4, we obtain the following results:
Theorem 2. Problem 4 has a unique set of solutions
{(ϖkH,φkH)}Kk=1⊂WH×WH |
defined on the coarse grid ℑH and a unique set of solutions
{(ϖkh,φkh)}Kk=1⊂Wh×Wh |
defined on the fine grid ℑh, respectively, meeting the following unconditional boundness, i.e., unconditional stability:
‖ϖkH‖1+‖ϖkh‖1+‖φkH‖0+‖φkh‖0⩽c(‖ϖ0‖1+‖ϖ1‖1),1⩽k⩽K. | (3.5) |
The c that appears here and after is also a positive constant independent of H, h, and Δt. Further, when
h=O(H1+1/l), |
they meet the following error estimates:
‖ϖ(tk)−ϖkH‖0+‖φ(tk)−φkH‖0+H‖∇(ϖ(tk)−ϖkH)‖0+H‖∇(φ(tk)−φkH)‖0⩽c(Δt2+Hl+1), | (3.6) |
‖ϖ(tk)−ϖkh‖0+‖φ(tk)−φkh‖0+h‖∇(ϖ(tk)−ϖkh)‖0+h‖∇(φ(tk)−φkh)‖0⩽c(Δt2+hl+1), | (3.7) |
where 1⩽k⩽K.
Proof. The demonstration of Theorem 2 consists of the following two parts:
(1) Prove the existence and unconditional stability of TGCNMFE solutions.
(ⅰ) Consider the existence and unconditional stability of the TGCNMFE solutions defined on the coarse grid ℑH.
Noting that the system of Eq (3.1) has the same form as the system of Eq (2.4), by using the same approach as proving Theorem 2, we can demonstrate that the nonlinear system of Eq (3.1) has a unique set of solutions
{(ϖkH,φkH)}Kk=1⊂WH×WH, |
satisfying
‖∇ϖkH‖0+‖∇φkH‖0⩽c(‖ϖ0‖1+‖ϖ1‖1),1⩽k⩽K. | (3.8) |
(ⅱ) Consider the existence and unconditional stability of the TGCNMFE solutions defined on the fine grid ℑh.
Let
A((ϖ,φ),(ϑ,υ))=(ϖ,υ)−(φ,ϑ)+(∇ϖ,∇ϑ)+Δt22(∇φ,∇υ)−Δt22(f′(ϖk+1H)ϖ,υ)+Δt22(φ,υ),F(υ,ϑ)=(2ϖkh−ϖk−1h,υ)−Δt22(∇φk−1h,∇υ)−Δt22(φk−1h,υ)+Δt22(f(ϖk+1H)−f′(ϖk+1H)ϖk+1H+f(ϖk−1h),υ). |
The linear system of equations (3.2) can be rewritten into as follows:
Find
(ϖkh,φkh)∈Wh×Wh(1⩽k⩽K) |
satisfying the following linear system of equations:
{A((ϖk+1h,φk+1h),(ϑh,υh))=F(ϑh,υh),∀(ϑh,υh)∈Wh×Wh,1⩽k⩽K−1,ϖ0h=Rhϖ0(x),ϖ1h=Rhϖ1(x),φ0h=Rhφ0(x),φ1h=Rhφ1(x),x∈Ω. | (3.9) |
By Poincaré's inequality, we claim that there is a constant θ0>0 such that
‖ϑ‖0⩽‖ϑ‖1⩽θ0‖∇ϑ‖0(∀ϑ∈W=H10(Ω)) |
and
|f′(ϖk+1H)|=|−cos(ϖk+1H)|⩽1. |
Thereupon, when Δt is adequately small meeting
Δt2θ0<4, |
there exists a constant
α0=min{Δt2/2,1−Δt2θ0/4}/θ20 |
satisfying
A((ϖ,φ),(ϖ,φ))=(ϖ,φ)−(φ,ϖ)+(∇ϖ,∇ϖ)+Δt22(∇φ,∇φ)+Δt22(φ,φ)−Δt22(cos(ϖkH)ϖ,φ)⩾(1−Δtθ04)‖∇ϖ‖20+Δt2‖∇φ‖20+Δt2‖φ‖20−Δt4‖φ‖20⩾α0‖(ϖ,φ)‖21,∀(ϖ,φ)∈Wh×Wh, | (3.10) |
where
‖(ϖ,φ)‖1=(‖ϖ‖21+‖φ‖21)1/2 |
is the norm in W×W. Thus, A((ϖ,φ),(ϑ,υ)) is positive definite, and the bilinear functional A((ϖ,φ),(ϑ,υ)) and the linear functional F(υ,ϑ) are evidently bounded in Wh×Wh for given ϖk+1H, ϖkh, ϖk−1h, and φk−1h. Thereupon, by the Lax-Milgram theorem (see [5, Theorem 1.2.1]), we assert that the linear system of Eq (3.9), namely Step 2 for Problem 4, has a unique set of solutions
{(ϖkh,φkh)}Kk=1⊂Wh×Wh |
satisfying
‖∇ϖkh‖0+‖∇φkh‖0⩽c(‖ϖ0‖1+‖ϖ1‖1),1⩽k⩽K. | (3.11) |
This signifies that the series of solutions
{(ϖkh,φkh)}Kk=1⊂Wh×Wh |
for Problem 4 defined on the fine grid ℑh is unconditionally bounded, namely it is unconditionally stable. Combining (3.8) with (3.11) yields (3.5).
(2) Estimate the errors of the TGCNMFE solutions of Problem 4.
(a) Estimate the errors of the solutions {(ϖkH,φkH)}Kk=1 of Problem 4 defined on ℑH.
By subtracting (3.1) from (2.4), and taking
υ=υHandϑ=ϑH, |
as well as setting
EkH=ϖk−ϖkH,ρkH=ϖk−RHϖk,ϱkH=RHϖk−ϖkH,˜EkH=φk−φkH,˜ρkH=φk−RHφk |
and
˜ϱkH=RHφk−φkH, |
we obtain
1Δt2(Ek+1H−2EkH+Ek−1H,υH)+12(∇(˜Ek+1H+˜Ek−1H),∇υH)+12(˜Ek+1H+˜Ek−1H,υH)=12(f(ϖk+1)−f(ϖk+1H)+f(ϖk−1)−f(ϖk−1H),υH),∀υH∈WH,1⩽k⩽K−1, | (3.12) |
(∇EkH,∇ϑH)=(˜EkH,ϑH),∀ϑH∈WH,1⩽k⩽K, | (3.13) |
E0H=ϖ0−RHϖ0,E1H=ϖ1−RHϖ1,˜E0H=φ0−RHφ0,˜E1H=φ1−RHφ1. | (3.14) |
By (3.2)–(3.4), (3.12), (3.13), Taylor's formula, LDMF, and the Hölder and Cauchy inequalities, when
h=O(H1+1/l), |
we obtain
1Δt2(‖∇(Ek+1H−EkH)‖20−‖∇(EkH−Ek−1H)‖20)+12(‖∇˜Ek+1H‖20−‖∇˜Ek−1H‖20)+12(‖˜Ek+1H‖20−‖˜Ek−1H‖20)=1Δt2(∇(ρk+1H−2ρkH+ρk−1H),∇(ρk+1H−ρk−1H))+1Δt2(˜Ek+1H−2˜EkH+˜Ek−1H,ϱk+1H−ϱk−1H)+12(∇(˜ρk+1H+˜ρk−1H),∇(˜ρk+1H−˜ρk−1H))+12(∇(˜Ek+1H+˜Ek−1H),∇(˜ϱk+1H−˜ϱk−1H))+12(˜Ek+1H+˜Ek−1H,˜ρk+1H−˜ρk−1H)+12(˜Ek+1H+˜Ek−1H,˜ϱk+1H−˜ϱk−1H)=1Δt2(∇(ρk+1H−2ρkH+ρk−1H),∇(ρk+1H−ρk−1H))+12(∇(˜ρk+1H+˜ρk−1H),∇(˜ρk+1H−˜ρk−1H))+12(˜Ek+1H+˜Ek−1H,˜ρk+1H−˜ρk−1H)+12(f(ϖk+1)−f(ϖk+1H)+f(ϖk−1)−f(ϖk−1H),˜ϱk+1H−˜ϱk−1H)⩽1Δt(‖∇(Ek+1H−EkH)‖20+‖∇(EkH−Ek−1H)‖20)+cΔtH2l+cΔt(‖˜Ek+1H‖20+‖˜Ek−1H‖20),1⩽k⩽K−1. | (3.15) |
By summing (3.15) from 1 to k (k⩽K−1), and using (3.4) and (3.14), we obtain
1Δt2‖∇(Ek+1H−EkH)‖20+12(‖∇˜Ek+1H‖20+‖∇˜EkH‖20)+12(‖˜Ek+1H‖20+‖˜EkH‖20)⩽cΔtk∑i=0‖∇(Ei+1H−EiH)‖20+cH2l+cΔtk∑i=0‖˜Ei+1H‖20,1⩽k⩽K−1. | (3.16) |
Thus, when Δt is adequately small, satisfying cΔt⩽1/4, by simplifying (3.16), we get
1Δt2‖∇(Ek+1H−EkH)‖20+‖∇˜Ek+1H‖20+‖˜Ek+1H‖0⩽cΔtk−1∑i=0‖∇(Ei+1H−EiH)‖20+cH2l+cΔtk−1∑i=0‖˜Ei+1H‖20,1⩽k⩽K−1. | (3.17) |
By applying Gronwall's inequality to (3.17), we obtain
1Δt2‖∇(Ek+1H−EkH)‖20+‖∇˜Ek+1H‖20+‖˜Ek+1H‖20⩽cH2lexp(ckΔt),1⩽k⩽K−1. | (3.18) |
Thereupon, we obtain
‖∇˜EkH‖0+‖˜EkH‖0⩽cHl,1⩽k⩽K. | (3.19) |
By the Nitsche technique (see [5, Theorem 1.3.9]) and (3.19), we immediately obtain the following error estimates:
‖φk−φkH‖0+H‖∇(φk−φkH)‖0⩽cHl+1,1⩽k⩽K. | (3.20) |
Further, by (3.2), (3.3), (3.12), (3.13), Taylor's formula, the LDMF, the Hölder and Cauchy inequalities, we obtain
1Δt2(‖Ek+1H−EkH‖20−‖EkH−Ek−1H‖20)+12(‖˜Ek+1H‖20−‖˜Ek−1H‖20)+12(‖∇Ek+1H‖20−‖∇Ek−1H‖20)=1Δt2(Ek+1H−2EkH+Ek−1H,ρk+1H−ρk−1H)+1Δt(Ek+1H−2EkH+Ek−1H,ϱk+1H−ϱk−1H)+12(˜Ek+1H+˜Ek−1H,˜ρk+1H−˜ρk−1H)+12(∇(ρkH+ρk−1H),∇(ρk+1H−ρk−1H))+12(˜Ek+1H+˜Ek−1H,˜ϱk+1H−˜ϱk−1H)+12(∇(EkH+Ek−1H),∇(ϱk+1H−ϱk−1H))=1Δt2(Ek+1H−2EkH+Ek−1H,ρk+1H−ρk−1H)+12(∇(ρkH+ρk−1H),∇(ρk+1H−ρk−1H))+12(∇(ρkH+ρk−1H),∇(ρk+1H−ρk−1H))+12(f(ϖk)−f(ϖkH)+f(ϖk−1)−f(ϖk−1H),ϱkH−ϱk−1H)⩽cΔt(‖Ek+1H−EkH‖20+‖EkH−Ek−1H‖20)+cΔtH2l+cΔt(‖˜Ek+1H‖20+‖˜Ek−1H‖20),1⩽k⩽K−1. | (3.21) |
By summing (3.21) from 1 to k, and using (3.4) and (3.14), we obtain
1Δt2‖Ek+1H−EkH‖20+12(‖˜Ek+1H‖20+‖˜EkH‖20)+12(‖∇Ek+1H‖20+‖∇EkH‖20)⩽cΔtk∑i=0‖Ei+1H−EiH‖20+cH2l+cΔtk∑i=0‖˜Ei+1H‖20,1⩽k⩽K−1. | (3.22) |
Thus, when Δt is adequately small, satisfying cΔt⩽1/4, by simplifying (3.22), we get
1Δt2‖Ek+1H−EkH‖20+‖˜Ek+1H‖20+‖∇Ek+1H‖20⩽cΔtk−1∑i=0‖Ei+1H−EiH‖20+cH2l+cΔtk−1∑i=0‖˜Ei+1H‖20,1⩽k⩽K−1. | (3.23) |
By applying Gronwall's inequality to (3.23), we obtain
1Δt2‖Ek+1H−EkH‖20+‖˜Ek+1H‖20+‖∇Ek+1H‖20⩽cH2lexp(ckΔt),1⩽k⩽K−1. | (3.24) |
Thereupon, we obtain
‖∇(ϖk−ϖkH)‖0⩽cHl,1⩽k⩽K. | (3.25) |
Further, by the Nitsche technique, we can obtain the following error estimates:
‖ϖk−ϖkH‖0+H‖∇(ϖk−ϖkH)‖0⩽cHl+1,1⩽k⩽K. | (3.26) |
Combining (3.20) and (3.26) with Theorem 1 yields (3.6).
(b) Estimate the errors of solutions (ϖkh,φkh) (1⩽k⩽K) of Problem 4 defined on the fine grid ℑh.
Subtracting (3.2) from (2.4), taking
υ=υhandϑ=ϑh, |
and setting
Ekh=ϖk−ϖkh,ρkh=ϖk−Rhϖk,ϱkh=Rhϖk−ϖkh,˜Ekh=φk−φkh,˜ρkh=φk−Rhφk |
and
˜ϱkh=Rhφk−φkh, |
we obtain
1Δt2(Ek+1h−2Ekh+Ek−1h,υh)+12(∇(˜Ek+h+˜Ek−1h),∇υh)+12(˜Ek+1h+˜Ek−1h,υh)=12(f(ϖk+1)−f(ϖk+1H)+f(ϖk−1)−f(ϖk−1h)−f′(ϖk+1H)(ϖkhk+1−ϖk+1H),υh),∀υh∈Wh,1⩽k⩽K−1, | (3.27) |
(∇Ekh,∇ϑh)=(˜Ekh,ϑh),∀ϑH∈Wh,1⩽k⩽K, | (3.28) |
E0h=ϖ0−Rhϖ0,E1h=ϖ1−Rhϖ1,˜E0h=φ0−Rhφ0,˜E1h=φ1−Rhφ1. | (3.29) |
By (3.3), (3.4), (3.27), (3.28), Taylor's formula, the LDMF, the Hölder and Cauchy inequalities, and (3.6) or (3.26), when
h=O(H1+1/l), |
we obtain
1Δt2(‖∇(Ek+1h−Ekh)‖20−‖∇(Ekh−Ek−1h)‖20)+12(‖∇˜Ek+1h‖20−‖∇˜Ek−1h‖20)+12(‖˜Ek+1h‖20−‖˜Ek−1h‖20)=1Δt2(∇(ρk+1h−2ρkh+ρk−1h),∇(ρk+1h−ρk−1h))+12(∇(˜ρk+1h+˜ρk−1h),∇(˜ρkh−˜ρk−1h))+12(˜Ek+1h+˜Ek−1h,˜ρk+1h−˜ρk−1h)+1Δt2(˜Ek+1h−2˜Ekh+˜Ek−1h,ϱk+1h−ϱk−1h)+12(∇(˜Ek+1h+˜Ek−1h),∇(˜ϱk+1h−˜ϱk−1h))+12(∇(˜Ekh+˜Ek−1h),∇(˜ϱk+1h−˜ϱk−1h))=1Δt2(∇(ρk+1h−2ρkh+ρk−1h),∇(ρk+1h−ρk−1h))+12(∇(˜ρk+1h+˜ρk−1h),∇(˜ρkh−˜ρk−1h))+12(˜Ek+1h+˜Ek−1h,˜ρk+1h−˜ρk−1h)−12(f′(ϖk+1H)(ϖk+1h−ϖk+1H),˜ϱk+1h−˜ϱk−1h)+12(f(ϖk+1)−f(ϖk+1H)+f(ϖk−1)−f(ϖk−1h),˜ϱk+1h−˜ϱk−1h)⩽2Δt(‖∇(Ek+1h−Ekh)‖20+‖∇(Ekh−Ek−1h)‖20)+cΔth2l+cΔt(‖˜Ek+1h‖20+‖˜Ek−1h‖20),1⩽k⩽K−1. | (3.30) |
By summing (3.30) from 1 to k (k⩽K−1), and using (3.4) and (3.29), we obtain
1Δt2‖∇(Ek+1h−Ekh)‖20+12(‖∇˜Ek+1h‖20+‖∇˜Ekh‖20)+12(‖˜Ek+1h‖20+‖˜Ekh‖20)⩽2Δtk∑i=0‖∇(Ei+1h−Ekh)‖20+cH2l+2+cΔtk∑i=0‖˜Ei+1h‖20,1⩽k⩽K−1. | (3.31) |
Thus, when Δt is adequately small, satisfying cΔt⩽1/4, by simplifying (3.31), we obtain
1Δt2‖∇(Ek+1h−Ekh)‖20+‖∇˜Ek+1h‖20+‖˜Ek+1h‖20⩽2Δtk−1∑i=0‖∇(Ei+1h−Ekh)‖20+cH2l+2+cΔtk−1∑i=0‖˜Ei+1h‖20,1⩽k⩽K−1. | (3.32) |
By applying Gronwall's inequality to (3.32), we obtain
1Δt2‖∇(Ek+1h−Ekh)‖20+‖∇˜Ek+1h‖20+‖˜Ek+1h‖20⩽ch2lexp(ckΔt)⩽ch2l,1⩽k⩽K−1. | (3.33) |
Thereupon, we obtain
‖∇(φk−φkh)‖0⩽chl,1⩽k⩽K. | (3.34) |
By the Nitsche technique (see [5, Theorem 1.3.9]), we easily obtain the following error estimates:
‖φk−φkh‖0+h‖∇(φk−φkh)‖0⩽chl+1,1⩽k⩽K. | (3.35) |
Further, by (3.3), (3.27), (3.28), Taylor's formula, the LDMF, the Hölder and Cauchy inequalities, and (3.6) or (3.26), we get
1Δt2(‖Ek+1h−Ekh‖20−‖Ekh−Ek−1h‖20)+12(‖˜Ek+1h‖20−‖˜Ek−1h‖20)+12(‖∇Ek+1h‖20−‖∇Ek−1h‖20)=1Δt2(Ek+1h−2Ekh+Ek−1h,ρk+1h−ρk−1h)+12(˜Ek+1h+˜Ek−1h,˜ρk+1h−˜ρk−1h)+12(∇(ρkh+ρk−1h),∇(ρk+1h−ρk−1h))+1Δt2(Ek+1h−2Ekh+Ek−1h,ϱk+1h−ϱk−1h)+12(˜Ek+1h+˜Ek−1h,˜ϱk+1h−˜ϱk−1h)+12(∇(Ek+1h+Ek−1h),∇(ϱk+1h−ϱk−1h))=1Δt2(Ek+1h−2Ekh+Ek−1h,ρk+1h−ρk−1h)+12(˜Ek+1h+˜Ek−1h,˜ρk+1h−˜ρk−1h)+12(∇(ρkh+ρk−1h),∇(ρk+1h−ρk−1h))−12(f′(ϖk+1H)(ϖk+1h−ϖk+1H),ϱk+1h−ϱk−1h)+12(f(ϖk+1)−f(ϖk+1H)+f(ϖk−1)−f(ϖk−1h),ϱk+1h−ϱk−1h)⩽cΔt(‖Ek+1H−EkH‖20+‖EkH−Ek−1H‖20)+cΔth2l+cΔt(‖˜Ek+1h‖20+‖˜Ek−1h‖20),1⩽k⩽K−1. | (3.36) |
By summing (3.36) from 1 to k (k⩽K−1), and using (3.4) and (3.29), we obtain
1Δt2‖Ek+1h−Ekh‖20+12(‖˜Ek+1h‖20+‖˜Ekh‖20)+12(‖∇Ek+1h‖20+‖∇Ekh‖20)⩽cΔtk∑i=0‖∇(Ei+1h−Ekh)‖20+cH2l+2+cΔtk∑i=0‖˜Ei+1h‖20,1⩽k⩽K−1. | (3.37) |
Thus, when Δt is adequately small, satisfying cΔt⩽1/4, by simplifying (3.37), we obtain
1Δt2‖∇(Ek+1h−Ekh)‖20+‖∇Ek+1h‖20+‖˜Ek+1h‖20⩽cΔtk−1∑i=0‖∇(Ei+1h−Ekh)‖20+cH2l+2+cΔtk−1∑i=0‖˜Ei+1h‖20,1⩽k⩽K−1. | (3.38) |
By applying Gronwall's inequality to (3.38), we obtain
1Δt2‖∇(Ek+1h−Ekh)‖20+‖∇Ek+1h‖20+‖˜Ek+1h‖20⩽ch2lexp(ckΔt)⩽ch2l,1⩽k⩽K−1. | (3.39) |
Thereupon, we obtain
‖∇(ϖk−ϖkh)‖0⩽chl,1⩽k⩽K. | (3.40) |
By the Nitsche technique (see [5, Theorem 1.3.9]), we easily deduce the following error estimates
‖ϖk−ϖkh‖0+h‖∇(ϖk−ϖkh)‖0⩽chl+1,1⩽k⩽K. | (3.41) |
Thereupon, inequality (3.7) is obtained by combining (3.35) and (3.41) with Theorem 1. This completes the proof for Theorem 2.
Remark 2. Theorem 2 shows that the TGCNMFE solutions are unconditionally stable and their theoretical errors reach optimal order. In Section 4, we use the numerical tests to verify the correctness of the obtained theoretical error estimates.
In this section, we provide two sets of numerical experiments to verify the correctness of our theoretical results and show the superiorities of the TGCNMFE method.
For the two-dimensional case, we take
ˉΩ=[0,1]×[0,1]⊂R2 |
and the initial functions
ϖ0(x)=ϖ1(x)=4sin(πx1)sin(πx2) |
in the NFOSG equation (i.e., Problem 1), in which it may be considered that there is an earthquake focus at center (0.5,0.5) of the region Ω.
The fine grid division ℑh is composed of squares with diagonal
h=√2/1000 |
and all squares parallel to the coordinate axis. When l=1, in order to satisfy
h=O(H1+1/l), |
i.e.,
h=O(H2), |
the coarse grid division ℑH is composed of the squares with diagonal
H=4√2/√1000 |
and all squares also parallel to the coordinate axis. According to Theorem 2, the L2 norm error estimates of the TGCNMFE solutions of the NFOSG equation can theoretically reach O(10−6) when
Δt=1/1000. |
We first find the TGCNMFE solutions ϖkh and φkh by the TGCNMFE format at t=10, and their contours are respectively shown in Figures 1a and 2a.
In order to show the advantages of the TGCNMFE format, we also find single-grid CNMFE (SGCNMFE) solutions ˆϖkh and ˆφkh at t=10 by the following SGCNMFE format, and their contours are respectively shown in Figures 1b and 2b.
Problem 5. Find
(ˆϖkh,ˆφkh)∈Wh×Wh(1⩽k⩽K) |
defined on the single fine grid ℑh, satisfying the nonlinear system of equations:
{(ˆϖk+1h−2ˆϖkh+ˆϖk−1h,υh)+Δt2(∇(ˆφk+1h+ˆφk−1h),∇υh)+Δt2(ˆφk+1h+ˆφk−1h,υh)=Δt2(f(ˆϖk+1h)+f(ˆϖk−1h),υh),∀υh∈Wh,1⩽k⩽K−1,(∇ˆϖkh,∇ϑh)=(ˆφkh,ϑh),∀ϑh∈Wh,0⩽k⩽K,ˆϖ0h=Rhϖ0(x),ˆφ0h=Rhφ0,ˆϖ1h=Rhϖ1(x),x∈Ω. | (4.1) |
By comparing the contours of each pair of graphs in Figures 1 and 2, it can be easy to see that the TGCNMFE solutions are very close to the SGCNMFE solutions at t=10.
To truly showcase the benefits of the TGCNMFE format, we record the CPU running time for finding the TGCNMFE and SGCNMFE solutions and their errors when t=2.0, 4.0, 6.0, 8.0, and 10.0, where errors are estimated respectively by ‖ϖkh−ϖk−1h‖0+‖φkh−φk−1h‖0 and ‖ˆϖkh−ˆϖk−1h‖0+‖ˆφkh−ˆφk−1h‖0, shown in Table 1.
t | SGCNMFE solutions errors | TGCNMFE solutions errors | SGCNMFE method CPU running-time | TGCNMFE method CPU running-time |
2.0 | 2.2316×10−6 | 3.0273×10−6 | 215.332 s | 112.056 s |
4.0 | 2.4187×10−6 | 3.1438×10−6 | 216.662 s | 113.112 s |
6.0 | 2.5665×10−6 | 3.2662×10−6 | 217.153 s | 114.224 s |
8.0 | 2.7782×10−6 | 3.3861×10−6 | 218.709 s | 115.451 s |
10.0 | 2.8864×10−6 | 3.4861×10−6 | 219.618 s | 116.872 s |
The data in Table 1 show that the numerical errors of TGCNMFE and SGCNMFE solutions reach O(10−6) when t=2.0, 4.0, 6.0, 8.0, and 10.0, which coincide with the obtained theoretical errors, but the CPU running time for finding SGCNMFE solutions is almost twice as long as that for finding SGCNMFE solutions. Therefore, the TGCNMFE method is visibly superior to the SGCNMFE method, and the TGCNMFE method is feasible and effective to solve the NFOSG equation.
For the three-dimensional NFOSG equation, we take
ˉΩ=[0,1]×[0,1]×[0,1]⊂R3 |
and the initial functions
ϖ0(x)=ϖ1(x)=sin(πx1)sin(πx2)sin(πx3), |
where it may be considered that there is an earthquake focus at center (0.5,0.5,0.5) of the region Ω.
The fine grid division ℑh is composed of cubes with diagonal
h=√3/1000 |
and all cubes parallel to the coordinate axis. When l=1, in order to satisfy
h=O(H1+1/l), |
i.e.,
h = O(H^{2}), |
the coarse grid division \Im_H is composed of cubes with diagonal
H = \sqrt[4]{3}/\sqrt{1000} |
and all cubes also parallel to the coordinate axis. According to Theorem 2, the L^2 norm errors of the TGCNMFE solutions of the NFOSG equation can also theoretically reach O(10^{-6}) when
\Delta t = 1/1000. |
We find the TGCNMFE solutions \varpi_h^k and \varphi_h^k by the TGCNMFE format at t = 1 , and their contours are respectively shown in Figures 3a and 4a.
To show that the TGCNMFE method is superior to the SGCNMFE method, we also find the SGCNMFE solutions \hat{\varpi}_h^k and \hat{\varphi}_h^k at t = 1 by the SGCNMFE format above (Problem 5), which are respectively shown in Figures 3b and 4b.
By comparing the each pair of graphs in Figures 3 and 4, it can also be easy to see that the TGCNMFE solutions are almost same as the SGCNMFE solutions at t = 1 .
To truly showcase the benefits of the TGCNMFE format, we also record the CPU running time for finding the TGCNMFE and SGCNMFE solutions and their errors when t = 0.2 , 0.4 , 0.6 , 0.8 , and 1.0 , where errors are also respectively estimated by \|\varpi_{h}^{k}-\varpi_{h}^{k-1}\|_0+\|\varphi_{h}^{k}-\varphi_{h}^{k-1}\|_0 and \|\hat{\varpi}_{h}^{k}-\hat{\varpi}_{h}^{k-1}\|_0+\|\hat{\varphi}_{h}^{k}-\hat{\varphi}_{h}^{k-1}\|_0 , shown in Table 2.
t | SGCNMFE solutions errors | TGCNMFE solutions errors | SGCNMFE method CPU running-time | TGCNMFE method CPU running-time |
0.2 | 2.5781 \times 10^{-6} | 3.1563 \times 10^{-6} | 422.562 s | 211.731 s |
0.4 | 2.6436 \times 10^{-6} | 3.2253 \times 10^{-6} | 424.826 s | 212.813 s |
0.6 | 2.7841 \times 10^{-6} | 3.3265 \times 10^{-6} | 426.716 s | 213.832 s |
0.8 | 2.8862 \times 10^{-6} | 3.4453 \times 10^{-6} | 428.261 s | 214.764 s |
1.0 | 2.9631 \times 10^{-6} | 3.5626 \times 10^{-6} | 429.142 s | 215.635 s |
The data in Table 2 also show that when t = 0.2 , 0.4 , 0.6 , 0.8 , and 1.0 , the numerical errors of the SGCNMFE and TGCNMFE solutions coincide with the theoretical errors O(10^{-6}) , but the CPU running time for finding SGCNMFE solutions is also almost twice as long as that for finding SGCNMFE solutions. It is further shown that the TGCNMFE method is indeed better than the SGCNMFE method, and the TGCNMFE method is indeed feasible and effective to solve the NFOSG equation.
Above, we have proposed a new NFOSG equation, a new TSDMCN scheme, and a new TGCNMFE method for the NFOSG equation, and have strictly proved the existence, stability, and error estimates of the TSDMCN and TGCNMFE solutions theoretically. We have also employed the two sets of numerical experiments to confirm the correctness of the obtained theoretical results and showed the superiorities for the TGCNMFE method. The TSDMCN scheme and the TGCNMFE method for the NFOSG equation are first proposed in this paper. They are completely different from the existing methods in [10,11,12]. Therefore, they are original and indeed fire-new.
Although we only study the TGCNMFE method for the NFOSG equation, the method of this paper can be extended to the more complex nonlinear PDEs, for example, the nonlinear Cahn-Hilliard equation and the Schrödinger equation in [19,20], in addition to the actual engineering nonlinear problems. Therefore, the TGCNMFE method has a wide range of applications.
Although the TGCNMFE method here can greatly simplify calculation, save CPU operating-time, and improve computational efficiency, when it is applied to settling practical engineering computations, it usually contains many (often more than tens of millions) unknowns and needs to take a long-time to obtain results on a computer. Thus, after a long computer operating time, owing to the accumulation of computing errors, the obtained TGCNMFE solutions could deviate from right solutions, or could even generate floating-point overflow, resulting in erroneous calculation results. Hence, in future study, we will use a proper orthogonal decomposition method (see [5,21,22]) to lessen the unknowns of the TGCNMFE method and develop some new reduced-dimensionality methods for the NFOSG equation.
Yanjie Zhou: conceptualization, investigation, methodology, validation, writing-original draft, formal analysis; Xianxiang Leng: conceptualization, methodology, formal analysis, writing-review and editing; Yuejie Li: validation, visualization, writing-review and editing; Qiuxiang Deng: inspection, writing-review and editing; Zhendong Luo: conceptualization, investigation, methodology, formal analysis, writing-original draft, writing-review, editing and proofreading. All authors have read and agreed to the published version of the manuscript.
This work is jointly supported by the National Natural Science Foundation of China (Nos. 42330801, 42374143 and 11671106) and National Natural Science Foundation Joint project Key fund project cooperation project (No. U1839206).
The authors declare that they have no conflicts of interest to report regarding the present study.
[1] |
A. Barone, F. Esposito, C. J. Magee, A. C. Scott, Theory and applications of the sine-Gordon equation, Riv. Nuovo Cim., 1 (1971), 227–267. https://doi.org/10.1007/BF02820622 doi: 10.1007/BF02820622
![]() |
[2] |
H. Eleuch, Y. V. Rostovtsev, Analytical solution to sine-Gordon equation, J. Math. Phys., 51 (2010), 093515. https://doi.org/10.1063/1.3491776 doi: 10.1063/1.3491776
![]() |
[3] | F. Brezzi, M. Fortin, Mixed and hybrid finite element methods, Springer-Verlag, 1991. https://doi.org/10.1007/978-1-4612-3172-1 |
[4] | P. G. Ciarlet, The finite element method for elliptic problems, North-Holland, 1978. https://doi.org/10.1137/1.9780898719208 |
[5] | Z. Luo, Finite element and reduced dimension methods for partial differential equations, Springer and Science Press of China, 2024. https://doi.org/10.1007/978-981-97-3434-4 |
[6] | A. M. Yousef, S. Z. Rida, H. R. Ibrahim, Approximate solution of fractional-order nonlinear sine-Gordon equation, Elixir Appl. Math., 82 (2015), 32549–32553. |
[7] | Z. Ren, L. Wei, An ADI finite element method for the sine-Gordon equation in two dimensions, Int. Math. Forum, 5 (2010), 1219–1225. |
[8] |
P. L. Christiansen, O. H. Olsen, Return effect for rotationally symmetric solitary wave solutions to the sine-Gordon equation, Phys. Lett. A, 68 (1978), 185–188. https://doi.org/10.1016/0375-9601(78)90798-3 doi: 10.1016/0375-9601(78)90798-3
![]() |
[9] | J. Cuevas-Maraver, P. G. Kevrekidis, F. Williams, The sine-Gordon model and its applications, Springer International Publishing, 2014. https://doi.org/10.1007/978-3-319-06722-3 |
[10] |
D. Shi, F. Zhang, A class of low order nonconforming finite element analysis for sine-Gordon equation, Math. Numer. Sin., 33 (2011), 289–297. https://doi.org/10.12286/jssx.2011.3.289 doi: 10.12286/jssx.2011.3.289
![]() |
[11] |
F. Wang, D. Shi, A new superconvergence analysis and extrapolation of Hermite-type finite element for nonlinear sine-Gordon equations, Acta Math. Appl. Sin., 35 (2012), 777–788. https://doi.org/10.12387/C2012061 doi: 10.12387/C2012061
![]() |
[12] |
D. Shi, F. Wang, M. Fan, Y. Zhao, A new approach of the lowest-order anisotropic mixed finite element high-accuracy analysis for nonlinear sine-Gordon equations, Math. Numer. Sin., 37 (2015), 148–161. https://doi.org/10.12286/jssx.2015.2.148 doi: 10.12286/jssx.2015.2.148
![]() |
[13] |
J. Xu, A novel two-grid method for semilinear elliptic equations, SIAM J. Sci. Comput., 15 (1994), 231–237. https://doi.org/10.1137/0915016 doi: 10.1137/0915016
![]() |
[14] |
D. Shi, Q. Liu, An efficient nonconforming finite element two-grid method for Allen-Cahn equation, Appl. Math. Lett., 98 (2019), 374–380. https://doi.org/10.1016/j.aml.2019.06.037 doi: 10.1016/j.aml.2019.06.037
![]() |
[15] |
D. Shi, R. Wang, Unconditional superconvergence analysis of a two-grid finite element method for nonlinear wave equations, Appl. Numer. Math., 150 (2020), 38–50. https://doi.org/10.1016/j.apnum.2019.09.012 doi: 10.1016/j.apnum.2019.09.012
![]() |
[16] |
Y. Liu, Y. Du, H. Li, J. Li, S. He, A two-grid mixed finite element method for a nonlinear fourth-order reaction-diffusion problem with time-fractional derivative, Comput. Math. Appl., 70 (2015), 2474–2492. https://doi.org/10.1016/j.camwa.2015.09.012 doi: 10.1016/j.camwa.2015.09.012
![]() |
[17] |
H. Li, R. Yang, Analysis of two spectral Galerkin approximation schemes for solving the perturbed FitzHugh-Nagumo neuron model, Comput. Math. Appl., 143 (2023), 1–9. https://doi.org/10.1016/j.camwa.2023.04.033 doi: 10.1016/j.camwa.2023.04.033
![]() |
[18] |
Z. Song, D. Li, D. Wang, H. Li, A modified Crank-Nicolson finite difference method preserving maximum-principle for the phase-field model, J. Math. Anal. Appl., 526 (2023), 127271. https://doi.org/10.1016/j.jmaa.2023.127271 doi: 10.1016/j.jmaa.2023.127271
![]() |
[19] |
S. Shen, Z. Yang, Z. Pang, Y. Ge, The complex-valued astigmatic cosine-Gaussian soliton solution of the nonlocal nonlinear Schrödinger equation and its transmission characteristics, Appl. Math. Lett., 125 (2022), 107755. https://doi.org/10.1016/j.aml.2021.107755 doi: 10.1016/j.aml.2021.107755
![]() |
[20] |
J. Li, Z. Yang, S. Zhang, Periodic collision theory of multiple cosine-Hermite-Gaussian solitons in Schrödinger equation with nonlocal nonlinearity, Appl. Math. Lett., 140 (2023), 108588. https://doi.org/10.1016/j.aml.2023.108588 doi: 10.1016/j.aml.2023.108588
![]() |
[21] |
H. Li, Y. Li, Y. Zeng, Z. Luo, A reduced-dimension extrapolation two-grid Crank-Nicolson finite element method of unknown solution coefficient vectors for spatial fractional nonlinear Allen-Cahn equations, Comput. Math. Appl., 167 (2024), 110–122. https://doi.org/10.1016/j.camwa.2024.05.007 doi: 10.1016/j.camwa.2024.05.007
![]() |
[22] |
Y. Li, H. Li, Y. Zeng, Z. Luo, A preserving accuracy two-grid reduced-dimensional Crank-Nicolson mixed finite element method for nonlinear wave equation, Appl. Numer. Math., 202 (2024), 1–20. https://doi.org/10.1016/j.apnum.2024.04.01 doi: 10.1016/j.apnum.2024.04.01
![]() |
t | SGCNMFE solutions errors | TGCNMFE solutions errors | SGCNMFE method CPU running-time | TGCNMFE method CPU running-time |
2.0 | 2.2316 \times 10^{-6} | 3.0273 \times 10^{-6} | 215.332 s | 112.056 s |
4.0 | 2.4187 \times 10^{-6} | 3.1438 \times 10^{-6} | 216.662 s | 113.112 s |
6.0 | 2.5665 \times 10^{-6} | 3.2662 \times 10^{-6} | 217.153 s | 114.224 s |
8.0 | 2.7782 \times 10^{-6} | 3.3861 \times 10^{-6} | 218.709 s | 115.451 s |
10.0 | 2.8864 \times 10^{-6} | 3.4861 \times 10^{-6} | 219.618 s | 116.872 s |
t | SGCNMFE solutions errors | TGCNMFE solutions errors | SGCNMFE method CPU running-time | TGCNMFE method CPU running-time |
0.2 | 2.5781 \times 10^{-6} | 3.1563 \times 10^{-6} | 422.562 s | 211.731 s |
0.4 | 2.6436 \times 10^{-6} | 3.2253 \times 10^{-6} | 424.826 s | 212.813 s |
0.6 | 2.7841 \times 10^{-6} | 3.3265 \times 10^{-6} | 426.716 s | 213.832 s |
0.8 | 2.8862 \times 10^{-6} | 3.4453 \times 10^{-6} | 428.261 s | 214.764 s |
1.0 | 2.9631 \times 10^{-6} | 3.5626 \times 10^{-6} | 429.142 s | 215.635 s |
t | SGCNMFE solutions errors | TGCNMFE solutions errors | SGCNMFE method CPU running-time | TGCNMFE method CPU running-time |
2.0 | 2.2316 \times 10^{-6} | 3.0273 \times 10^{-6} | 215.332 s | 112.056 s |
4.0 | 2.4187 \times 10^{-6} | 3.1438 \times 10^{-6} | 216.662 s | 113.112 s |
6.0 | 2.5665 \times 10^{-6} | 3.2662 \times 10^{-6} | 217.153 s | 114.224 s |
8.0 | 2.7782 \times 10^{-6} | 3.3861 \times 10^{-6} | 218.709 s | 115.451 s |
10.0 | 2.8864 \times 10^{-6} | 3.4861 \times 10^{-6} | 219.618 s | 116.872 s |
t | SGCNMFE solutions errors | TGCNMFE solutions errors | SGCNMFE method CPU running-time | TGCNMFE method CPU running-time |
0.2 | 2.5781 \times 10^{-6} | 3.1563 \times 10^{-6} | 422.562 s | 211.731 s |
0.4 | 2.6436 \times 10^{-6} | 3.2253 \times 10^{-6} | 424.826 s | 212.813 s |
0.6 | 2.7841 \times 10^{-6} | 3.3265 \times 10^{-6} | 426.716 s | 213.832 s |
0.8 | 2.8862 \times 10^{-6} | 3.4453 \times 10^{-6} | 428.261 s | 214.764 s |
1.0 | 2.9631 \times 10^{-6} | 3.5626 \times 10^{-6} | 429.142 s | 215.635 s |