Abstract
In this paper, two-grid methods (TGMs) are developed for a system of reaction-diffusion equations of bacterial infection with initial and boundary conditions. The backward Euler (B-E) and Crank–Nicolson (C-N) fully discrete schemes are established, and the existence and uniqueness of the solutions of these schemes are proved. Moreover, based on the combination technique of the interpolation and Ritz projection and derivative transfer trick which are important ingredients in the TGMs, the superclose estimates of order O(h2 + H4 + τ) and O(h2 + H4 + τ2) in H1-norm are deduced for the above schemes, respectively, where h is fine mesh size, H is coarse mesh size, and τ is time step size. Then, by the interpolated postprocessing approach, the corresponding global superconvergence results are obtained. Finally, some other popular finite elements are discussed and numerical results are provided to verify the theoretical analysis, which show that the computing cost of TGMs are only half of Galerkin finite element methods (FEMs) for the test problem.
Similar content being viewed by others
Avoid common mistakes on your manuscript.
1 Introduction
We consider the following bacteria equations [1]:
where Ω ⊂ R2 is a rectangle with boundaries parallel to the axis, J = (0, T], T > 0. Here, u and v represent the average concentration of bacteria and the infective human, respectively. Hence, − a11u describes the natural mortality rate of u, a12v is the growth rate of u due to v, and − a12v is the natural damping of v due to the finite duration of infectiousness of humans. g(u) represents the force of infection of u against v, which is twice continuously differentiable functional and satisfies the Lipschitz condition. We assume that the initial values u0(x) and v0(x) are given smooth functions. At the same time, d1, d2, a11, a12, and a22 are positive constants.
The problem (1) established the spatial transmission model of bacteria under the environment pollution of human, which has important research significance and there have been some works about it. For instance, the existence of periodic plane wave solutions was analyzed in [2], and the asymptotic stability of plane wave solutions in the bounded and unbounded domains were studied for 1D case in [3]. Moreover, the nonconforming FEM with \(\text {EQ}_{1}^{\text {rot}}\) element was applied to this problem in [4], and optimal error estimate with order of O(h2) in L2-norm and the superclose estimate with order of O(h2) in H1-norm were deduced for the semi-discrete scheme, respectively. Besides, the superclose estimate of order O(h2 + τ) for the C-N fully discrete scheme was also obtained. The backward and central difference finite element schemes were proposed and their priori error estimates in L2-norm were derived with order of O(h2 + τ) and O(h2 + τ2), respectively in [5]. Meanwhile, the existence, uniqueness and stability of traditional solutions under different boundary conditions were studied in [6,7,8,9]. However, there is no consideration on the estimations of the solutions u and v on the energy norm.
As we know, TGM is a very efficient algorithm for solving nonlinear PDEs [10,11,12], in which the nonlinear problem is first solved on the coarse mesh with size H and then a simple linearized scheme (one Newton like iteration) is dealt with on the fine mesh with size h (h ≪ H). Later on, some developments on this aspect were also achieved, such as unconditional optimal error estimate of order O(h + H3 + τ) in H1-norm and order O(h2 + H3 + τ) in L2-norm were deduced in [13] by introducing an auxiliary time discrete system, and the superclose estimate in H1-norm was improved to order O(h2 + H4 + τ) in [14]. Furthermore, TGM also has been applied to the reaction–diffusion problems [15, 16], Sobolev problems [17], Navier–Stokes problems [18, 19], Maxwell’s problems [20], and so on.
In this paper, we shall take the bilinear finite element, for example, to develop two-grid algorithms for the problem (1), and then to investigate their superconvergent behavior in H1-norm through the combination skill of the interpolation and Ritz projection of [21] and derivative transfer trick.
The remainder of the paper is organized as follows. In Section 2, we give the B-E and C-N schemes and their briefly proofs of the existence and uniqueness of solutions, and then deduce the superclose estimates for the above two schemes in H1-norm, respectively. In Sections 3–4, we present the approximate schemes with TGMs and study their superconvergence properties. In the last section, we provide some numerical results to verify the theoretical analysis.
Through out this paper, without ambiguity, we use the standard Banach space Wm, p(Ω) with norm ∥⋅∥m, p and if p = 2, we can simply denote Wk,2(Ω) by Hk(Ω) with the inner product (⋅,⋅). Then, we equip a function space Lp(J;X) with the norm \(\|g\|_{L^{p}(J;X)}:=\left ({\int \limits }_{J}\|g(\cdot ,t)\|^{p}_{X} \ \text {dt}\right )^{\frac {1}{p}}\) and if \(p=\infty ,\) the integral is replaced by the essential supremum. The generic constant C > 0 is independent of n (time level), H, h, and τ, and may have different values in different places.
2 Superclose estimates for Galerkin FEMs
2.1 B-E scheme case
Let \({\mathcal{T}}_{h}\) be a regular rectangular partition of Ω with h, and \(\tilde {V}_{h}^{0}\) be the bilinear finite element space which vanishes on ∂Ω. We define the Ritz projection operator \(R_{h}: {H_{0}^{1}}({\Omega })\rightarrow \tilde {V}_{h}^{0}\) as follows:
At the same time, let Ih be the associated interpolation operator over \(\tilde {V}_{h}^{0}\), then, for \(w\in {H_{0}^{1}}({\Omega })\bigcap H^{3}({\Omega })\), we can obtain the following estimates according lemma 2 in [21]:
The weak formulation of the problem (1) is to find \((u,v): J\rightarrow {H_{0}^{1}}({\Omega })\times {H_{0}^{1}}({\Omega }),\) such that
Let 0 = t0 < t1 < … < tN = T be a uniform partition on [0, T] with τ and φn = φ(x, tn), for φ is a generalized function. Then, we may pose the following B-E fully discrete scheme of the problem (1) to find \(\left ({U^{n}_{h}},{V^{n}_{h}}\right ) \in \tilde {V}_{h}^{0}\times \tilde {V}_{h}^{0}\) for n = 1, 2…, N, such that
where \(D_{\tau } \varphi ^{n}=\left (\varphi ^{n}-\varphi ^{n-1}\right )/\tau \).
Now, we define a new finite element space \(\mathbf {W}_{h}=\left \{\mathbf {x}_{h}=(U_{h},V_{h})^{T}: U_{h},\right .\) \(\left .V_{h} \in \tilde {V}_{h}^{0}\right \}\) endowed with the scalar product (x1h, x2h)∗ := (U1h, U2h) + (V 1h, V 2h) where xih := (Uih, V ih) (i = 1, 2) and the norm \(\|\mathbf {x}_{h}\|^{2}_{*}:=\|U_{h}\|_{0}^{2}+\|V_{h}\|_{0}^{2}\). Then, let
Obviously, the function F(⋅) satisfies the Lipschitz condition, and the problem (6) can be rewritten as to find \(\mathbf {x}^{n}_{h}\in \mathbf {W}_{h}\) for n = 1, 2… , N, such that
where D∗ = diag(d1, d2).
Throughout this paper, we assume that the solution (u, v) to problem (1) exists and satisfies the following:
Theorem 1
The problem (7) has a unique solution \(({U_{h}^{n}},{V_{h}^{n}})\).
Proof
It is similar to the proof of [22] (see pages 236–237).
Now, we consider the following theorem which gives the superclose estimate of the problem (6). □
Theorem 2
Let (u, v) and \(\left ({U_{h}^{n}},{V_{h}^{n}}\right )\) be the solutions of (5) and (6), respectively, then we have the following:
Proof
Let t = tn in (5), we get the following:
where
Denote \(u^{n}-{U_{h}^{n}}=u^{n}-R_{h}u^{n}+R_{h}u^{n}-{U_{h}^{n}}:=\eta ^{n}+\xi ^{n},\ \ v^{n}-{V_{h}^{n}}=v^{n}-R_{h}v^{n}+R_{h}v^{n}-{V_{h}^{n}}:=\chi ^{n}+\gamma ^{n}\).
Then, subtracting (6) from (9), we obtain the following:
Noticing that,
Then, let ϕh = Dτξn in (12) and ψh = Dτγn in (13) to have the following:
Here, by use of (2), we have D1 = 0, E1 = 0. And observing that \(\|D_{\tau }\phi ^{n}\|_{0}^{2}\leq \tau ^{-1}{\int \limits }^{t_{n}}_{t_{n-1}}\left \|{\phi _{t}^{n}}\right \|_{0}^{2}\ \text {ds}\) (see [13]) and (3), we can get by Cauchy inequality and Young inequality that,
Considering (10)–(11), we find that
Under the assumption of g(⋅), we have at once the following:
Then, substituting above results into (16)–(17), we can get the following:
where \(c_{0}={\min \limits } \{d_{1},d_{2},2a_{11},2a_{12}\}\).
Multiplying (18) by 2τ and summing from n = 1, … , m (1 ≤ m ≤ N), we have the following:
Thanks to discrete Gronwall’s lemma, when \(c_{0}-4 a_{12}^{2}\tau >0\) and c0 − 4c2τ > 0, we have the following:
which together with (4) completes the proof. □
2.2 C-N scheme case
For the purpose of obtaining higher accuracy in time, we shall pose the following C-N fully discrete scheme of the problem (1) to find \(\left ({U^{n}_{h}},{V^{n}_{h}}\right ) \in \tilde {V}_{h}^{0}\times \tilde {V}_{h}^{0}\) for n = 1, 2…, N, such that
where \(\bar {\varphi }^{n}=(\varphi ^{n}+\varphi ^{n-1})/2\).
Theorem 3
The problem (19) has a unique solution \(\left ({U_{h}^{n}},{V_{h}^{n}}\right )\).
Proof
Similar to (7), we shall rewrite the problem (19) as to find \(\mathbf {x}^{n}_{h}\in \mathbf {W}_{h}\) for n = 1, 2… , N, such that
Here, we shall first multiply (20) by 2τ and remark it as \(\left (G_{h}\left (\mathbf {x}_{h}^{n}\right ),{\Phi }_{h}\right )_{*}=0,\) where \(G_{h}: \mathbf {W}_{h}\rightarrow \mathbf {W}_{h}\) is a continuous functional. Then, with the Lipschitz continuous property of F(⋅), we have for given \(\mathbf {x}_{h}^{n-1}\) as follows:
\(q^{2}=\frac {1}{2}\left [(1+C\tau )\left \|\mathbf {x}_{h}^{n-1}\right \|_{*}^{2}+C\tau \|F(0)\|_{*}^{2} +\tau \left (\frac {\max \limits _{\lambda \in \sigma (D)}|\lambda |}{4\min \limits _{\lambda \in \sigma (D)}|\lambda |} \left \|\nabla \mathbf {x}_{h}^{n-1}\right \|_{*}\right )\right ]\) to ensure (Gh(Φh),Φh)∗ > 0 for ∥Φh∥∗ = q, when \(\tau \leq \tau _{0}<\frac {2}{3C}\).
Further, by the Brouwer’s fixed point theorem, we see that the equation Gh(x) = 0 has a solution \(\mathbf {x}\in B_{q}= \{{\Phi }_{h}\in \mathbf {W}_{h};\|{\Phi }_{h}\|_{*}\leqslant {q}\}\). In fact, if we assume that Gh(Φh)≠ 0 in Bq, then the mapping \(\tilde {G}_{h}({\Phi }_{h})=-qG_{h}({\Phi }_{h})/{\|G_{h}({\Phi }_{h})\|_{*}}\) is continuous from Bq to itself, and therefore has a fixed point \(\tilde {\Phi }_{h}\in B_{q},\) with \(q^{2}=\|\tilde {\Phi }_{h}\|_{*}^{2}=-q(G_{h}(\tilde {\Phi }_{h},\tilde {\Phi }_{h}))_{*}/{\|G_{h}(\tilde {\Phi }_{h})\|_{*}},\) which contradicts \((G_{h}(\tilde {\Phi }_{h}),\tilde {\Phi }_{h})_{*}>0\). So, there exists a solution \(\mathbf {x}_{h}^{n}\) of (20) in Bq, namely the problem (19) has a solution \(\left ({U_{h}^{n}},{V_{h}^{n}}\right )\).
Now, we give a briefly proof of the uniqueness of the solution \(\left ({U_{h}^{n}},{V_{h}^{n}}\right )\) of the problem (19), when the solution (u, v) of the problem (1) is smooth and τ is sufficiently small. In fact, let x1 = (U1, V1) and x2 = (U2, V2) be two solutions of the problem (19). Then by subtraction, we have as follows:
Choosing ϕh = x1 −x2, we find the following:
When τ ≤ τ0, we can conclude that ∥x1 −x2∥∗ = 0, thus, U1 = U2, V1 = V2, and the problem (19) has a unique solution \(\left ({U_{h}^{n}},{V_{h}^{n}}\right )\). The proof is completed. □
Theorem 4
Let (u, v) and \(\left ({U_{h}^{n}},{V_{h}^{n}}\right )\) be the solutions of (5) and (19), respectively, then we have the following:
Proof
Let \(t=t_{n-\frac {1}{2}}\) in (5), then we can get the following:
where
We write as before \(u^{n}-{U_{h}^{n}}=u^{n}-R_{h}u^{n}+R_{h}u^{n}-{U_{h}^{n}}:=\eta ^{n}+\xi ^{n},\ \ v^{n}-{V_{h}^{n}}=v^{n}-R_{h}v^{n}+R_{h}v^{n}-{V_{h}^{n}}:=\chi ^{n}+\gamma ^{n}\).
Then subtracting (19) from (21) and choosing ϕh = Dτξn, ψh = Dτγn, we can get the following:
Collecting (24)–(25), the left side of the equation (LHS) can be estimated as follows in the same way as (14)–(15):
So, our purpose now is thus to drive error estimates of \({I_{i}^{n}}\) and \({K_{i}^{n}}\ (i=1,\ldots ,8)\).
As in the estimation of Di, Ej (i = 1, … ,4, j = 1, … ,3), we have the following:
and by use of (22)–(23), we can get that
Then, we have the following estimation with \(\|\bar {\phi }^{n}-\phi ^{n-\frac {1}{2}}\|^{2}_{0}\leq \frac {\tau ^{3}}{16}{\int \limits }_{t_{n-1}}^{t_{n}}\|\phi _{\text {tt}}\|_{0}\ \text {ds},\)
Furthermore, by the Lipschitz continuous property of g(⋅), \({K_{4}^{n}}\ \text {and}\ {K_{8}^{n}}\) can be estimated as follow:
Hence, substituting above results into (24)–(25), we can get the following:
Then, multiplying (27) by 2τ and summing from n = 1, … , m (1 ≤ m ≤ N), we have the following:
By use of the derivative transfer trick, we can rewrite \(2\tau \sum \limits _{n=1}^{m}{I_{6}^{n}}\) as follows:
With Taylor expansion, we have the following:
Then, by Cauchy inequality and Young inequality, we can get the following:
So, we have the following:
Similarly, we can get the estimate as follows:
Substituting (34) and (35) into (28) with σi = c0/2,
where \(c_{1}=3a_{12}^{2}+2,\ c_{2}=3c^{2}\tau +2\).
Thanks to discrete Gronwall’s lemma, when c0/2 − c1τ > 0 and c0/2 − c2τ > 0, we have the following:
which together with (4) completes the proof. □
3 Superclose estimates for TGMs
3.1 B-E scheme case
We define another bilinear finite element space \(\tilde {V}^{0}_{H}\subset \tilde {V}^{0}_{h}\ (h\ll H\ll 1)\) on the coarse grid. Then, TGM for the B-E scheme can be described as follows.
-
Step 1: On the coarse grid \({\mathcal{T}}_{H},\) for n = 1, … , N, solve \(\left ({U_{H}^{n}} , {V_{H}^{n}}\right )\in \tilde {V}_{H}^{0}\times \tilde {V}_{H}^{0}\) for the following nonlinear system, such that
$$ \left\{ \begin{array}{ll} &\left( D_{\tau} {U_{H}^{n}},\phi_{H}\right)+d_{1}\left( \nabla {U_{H}^{n}},\nabla\phi_{H}\right) +a_{11}\left( {U_{H}^{n}},\phi_{H}\right)-a_{12}\left( {U_{H}^{n}},\phi_{H}\right)=0,\ \ \ \forall\phi_{H}\in \tilde{V}_{H}^{0}, \\ &\left( D_{\tau} {V_{H}^{n}},\psi_{H}\right)+d_{2}\left( \nabla {V_{H}^{n}},\nabla\psi_{H}\right) +a_{22}\left( {V_{H}^{n}},\psi_{H}\right)=\left( g\left( {U_{H}^{n}}\right),\psi_{H}\right),\ \ \ \forall\psi_{H}\in \tilde{V}_{H}^{0},\\ &{U_{H}^{0}}=R_{H}u^{0}, \ {V_{H}^{0}}=R_{H}v^{0}. \end{array} \right. $$(37) -
Step 2: On the fine grid \({\mathcal{T}}_{h},\) for n = 1, … , N, solve \(\left ({U_{h}^{n}} , {V_{h}^{n}}\right )\in \tilde {V}^{0}_{h}\times \tilde {V}^{0}_{h}\) for the following linearized system, such that
$$ \left\{ \begin{array}{ll} & \left( D_{\tau} {U_{h}^{n}},\phi_{h}\right) + d_{1}\left( \nabla {U_{h}^{n}},\nabla\phi_{h}\right) + a_{11}\left( {U_{h}^{n}},\phi_{h}\right)-a_{12}\left( {U_{h}^{n}},\phi_{h}\right) = 0,\ \ \ \forall\phi_{h}\in \tilde{V}^{0}_{h}, \\ & \left( D_{\tau} {V_{h}^{n}},\psi_{h}\right) + d_{2}\left( \nabla {V_{h}^{n}},\nabla\psi_{h}\right) + a_{22}\left( {V_{h}^{n}},\psi_{h}\right) = \left( g\left( {U_{H}^{n}}\right) + g^{\prime}\left( {U_{H}^{n}}\right)\left( {U_{h}^{n}} - {U_{H}^{n}}\right),\psi_{h}\right),\ \ \ \forall\psi_{h} \in \tilde{V}^{0}_{h},\\ & {U_{h}^{0}} = R_{h}u^{0}, \ {V_{h}^{0}} = R_{h}v^{0}. \end{array} \right. $$(38)By the similar arguments to Theorem 2, we can easily prove that (38) has a unique solution.
Now, we present the superclose estimate of the above TGM.
Theorem 5
Let \((u^{n},v^{n}) , ({U_{H}^{n}},{V_{H}^{n}})\), and \(({U_{h}^{n}},{V_{h}^{n}}) \) be the solutions of (9), (37), and (38), respectively. Then, we have the following:
Proof
From Theorem 2, (39) is true obviously. So, we only need to prove (40).
In fact, by Taylor expansion, we have the following:
Set \(u^{n}-{U_{h}^{n}}=u^{n}-R_{h}u^{n}+R_{h}u^{n}-{U_{h}^{n}}:= \tilde {\eta }^{n}+\tilde {\xi }^{n},\ \ v^{n}-{V_{h}^{n}}=v^{n}-R_{h}v^{n}+R_{h}v^{n}-{V_{h}^{n}}:= \tilde {\chi }^{n}+\tilde {\gamma }^{n}\).
Then, subtracting (38) from (9), and choosing \(\phi _{h} = D_{\tau }\tilde {\xi }^{n}, \psi _{h} = D_{\tau }\tilde {\gamma }^{n}\). We get that
We write as before, collecting (41)–(42), and we have the following:
Here, in the same way as the estimates of (16) and (17), we can get the following estimates:
Moreover, due to H1↪L4, interpolation theory, and (39), we have the following:
Altogether, we can see the following:
Multiplying (46) by 2τ and after integration, we have the following:
So, by discrete Gronwall’s lemma, when \(c_{0}-4a_{12}^{2}\tau >0 \) and c0 − 4c2τ > 0, we have the following:
which together with (4) completes the proof. □
3.2 C-N scheme case
We first establish TGM for the C-N scheme of the problem (19) as follows.
-
Step 1: On the coarse grid \({\mathcal{T}}_{H},\) for n = 1, … , N, solve the following nonlinear system for \(\left ({U_{H}^{n}} , {V_{H}^{n}}\right )\in \tilde {V}_{H}^{0}\times \tilde {V}_{H}^{0}\), such that
$$ \left\{ \begin{array}{ll} &\left( D_{\tau} {U_{H}^{n}},\phi_{H}\right)+d_{1}\left( \nabla \bar{U}_{H}^{n},\nabla\phi_{H}\right) +a_{11}\left( \bar{U}_{H}^{n},\phi_{H}\right)-a_{12}\left( \bar{U}_{H}^{n},\phi_{H}\right)=0,\ \ \ \forall\phi_{H}\in \tilde{V}_{H}^{0}, \\ &\left( D_{\tau} {V_{H}^{n}},\psi_{H}\right)+d_{2}\left( \nabla \bar{V}_{H}^{n},\nabla\psi_{H}\right) +a_{22}\left( \bar{V}_{H}^{n},\psi_{H}\right)=\left( g\left( \bar{U}_{H}^{n}\right),\psi_{H}\right),\ \ \ \forall\psi_{H}\in \tilde{V}_{H}^{0},\\ &{U_{H}^{0}}=R_{H}u^{0}, \ {V_{H}^{0}}=R_{H}v^{0}. \end{array} \right. $$(47) -
Step 2: On the fine grid \({\mathcal{T}}_{h},\) for n = 1, … , N, solve the following linear system for \(({U_{h}^{n}} , {V_{h}^{n}})\in \tilde {V}^{0}_{h}\times \tilde {V}^{0}_{h},\) such that
$$ \left\{ \begin{array}{ll} & \left( D_{\tau} {U_{h}^{n}},\phi_{h}\right) +d_{1}\left( \nabla \bar{U}_{h}^{n},\nabla\phi_{h}\right) +a_{11}\left( \bar{U}_{h}^{n},\phi_{h}\right) -a_{12}\left( \bar{U}_{h}^{n},\phi_{h}\right) = 0,\ \ \ \forall\phi_{h}\in \tilde{V}^{0}_{h}, \\ & \left( D_{\tau} {V_{h}^{n}},\psi_{h}\right) +d_{2}\left( \nabla \bar{V}_{h}^{n},\nabla\psi_{h}\right) +a_{22}\left( \bar{V}_{h}^{n},\psi_{h}\right) = \left( g\left( \bar{U}_{H}^{n}\right) + g^{\prime}\left( \bar{U}_{H}^{n}\right)\left( \bar{U}_{h}^{n} - \bar{U}_{H}^{n}\right),\psi_{h}\right),\ \ \forall\psi_{h} \in \tilde{V}^{0}_{h},\\ & {U_{h}^{0}}=R_{h}u^{0}, \ {V_{h}^{0}}=R_{h}v^{0}. \end{array} \right. $$(48)Similar to the proof of Theorem 3, we see that (48) has a unique solution.
Now, we present the superclose estimates of the above TGM of (47)–(48).
Theorem 6
Let \((u^{n},v^{n}), \left ({U_{H}^{n}},{V_{H}^{n}}\right )\), and \(\left ({U_{h}^{n}},{V_{h}^{n}}\right )\) be the solutions of (21), (47), and (48), respectively. Then, we have the following:
Proof
From Theorem 4, (49) is true obviously. So, we only need to prove (50).
In fact, by Taylor expansion, we have the following:
Let \(u^{n}-{U_{h}^{n}}=u^{n}-R_{h}u^{n}+R_{h}u^{n}-{U_{h}^{n}}:= \tilde {\eta }^{n}+\tilde {\xi }^{n},\ \ v^{n}-{V_{h}^{n}}=v^{n}-R_{h}v^{n}+R_{h}v^{n}-{V_{h}^{n}}:= \tilde {\chi }^{n}+\tilde {\gamma }^{n}\).
Then, subtracting (48) from (19) and choosing \(\phi _{h}=D_{\tau }\tilde {\xi }^{n},\ \psi _{h}=D_{\tau }\tilde {\gamma }^{n},\) we can get error equations at once:
Then, similar to the estimates of \({I^{n}_{i}},{K^{n}_{i}}\ (i=1,\ldots , 5,7,8)\) in Theorem 4, we can easily get the following:
Thanks to the Lipschitz continuous property of \(g(\cdot ),\ {G_{8}^{n}}\) can be bounded,
Moreover, with the help of H1↪L4, interpolation theory, and (49), we can get the following:
Then, substituting (53)–(55) into (51)–(52), we have that
Multiplying (56) by 2τ and summing from n = 1, … , m (1 ≤ m ≤ N), we have the following:
Similar to the estimate of \(2\tau \sum \limits _{n=1}^{m}{I^{n}_{6}}\ \text {and}\ 2\tau \sum \limits _{n=1}^{m} {K^{n}_{6}}\), we can obtain the following:
So, substituting (58) into (57) with σ3 = σ4 = c0/2, we have as follows:
where \(c_{1}=3a_{12}^{2}+2,\ c_{2}=7c^{2}+2\).
Then, by discrete Gronwall’s lemma, when c0/2 − c1τ > 0 and c0/2 − c2τ > 0, we have the following:
which together with (4) completes the proof. □
4 Global superconvergence analysis of TGM
Now, we start to derive the superconvergence results by applying the interpolated postprocessing operator \({\Pi }^{2}_{2h}\) (see [24]), satisfying the following:
where \({S_{2}^{h}}\) is the biquadratic finite element space.
Theorem 7
Under the assumption of (8), we have the following:
Proof
We start to prove (61(a)), and (61(b)) can be treated in the same way. As usual, we shall write the error as a sum of two terms: \(u-{\Pi }^{2}_{2h}U_{h}=u-{\Pi }^{2}_{2h}I_{h}u+{\Pi }^{2}_{2h}I_{h}u-{\Pi }^{2}_{2h}U_{h},\) then by use of (60), we have the following:
Again applying (60), with the help of Theorem 3, we obtain the following:
Together with the estimates above, it is easy to see that
Similarly, we can get desired result of v as follows:
This completes the proof. □
Remark 1
In Theorem 2, if we use Rh alone, how to construct an interpolated postprocessing operator \({\Pi }_{2h}^{2}\) to satisfy \({\Pi }_{2h}^{2}R_{h}u={\Pi }_{2h}^{2}u\) is still an open problem. In addition, if we only use the operator Ih and the estimation (∇(ϕ − Ihϕ),∇ϕh) = O(h2)∥ϕ∥3∥ϕh∥1 proved in [23], it will result in the following:
Obviously, the requirement of ut, vt ∈ L2(J;H3(Ω)) in (62) as well as [25] is higher than that ut, vt ∈ L2(J; H2(Ω)) in Theorem 2. This is the main reason why we use the combination technique in our work.
On the other hand, in the proof of Theorem 4, the derivative transfer trick is crucial to estimate \({J_{6}^{n}}\) and \({K_{6}^{n}}\). Otherwise, how to get the superclose estimate of order O(h2 + H4 + τ2) in H1-norm is also an open problem.
Remark 2
Our analysis presented herein are also valid to some other popular finite elements:
-
(i)
For the conforming linear triangular element space \(\tilde {V}^{0}_{h}\) [23], we have the following:
$$ \begin{array}{@{}rcl@{}} (\nabla(u-I_{h}u),\nabla v)\leq {Ch}^{2}||u||_{3}||v||_{1},\ u\in H^{3}({\Omega})\cap {H_{0}^{1}}({\Omega}),\ v\in\tilde{V}^{0}_{h}. \end{array} $$(63)So, we can get the results of Theorems 2, 4–6 by use of (63) and applying the combination technique as our paper.
-
(ii)
For the nonconforming elements \(Q_{1}^{\text {rot}}\) [26, 27] on square mesh, \(\text {EQ}_{1}^{\text {rot}}\) [28, 29] and \(\text {CNQ}_{1}^{\text {rot}}\) [30] on rectangular mesh, there holds for \(u\in H^{3}({\Omega })\bigcap {H^{1}_{0}}({\Omega })\) as follows:
$$ (\nabla_{h}(u - {\Pi}_{h}u),\nabla_{h} v)\leq \left\{ \begin{array}{ll} &0,\ v\in\tilde{V}^{0}_{h},\ \ \ \text{for}\ Q_{1}^{\text{rot}}\ and\ \text{EQ}_{1}^{\text{rot}}\ \text{elements},\\ &{Ch}^{2}||u||_{3}||v||_{1},\ v\in\tilde{V}^{0}_{h},\ \ \ \text{for}\ \text{CNQ}_{1}^{\text{rot}}\ \text{element}, \end{array} \right. $$(64)$$ \begin{array}{@{}rcl@{}} \left|\sum\limits_{K\in\mathcal{T}_{h}} {\int}_{\partial K}\frac{\partial u}{\partial n}\mathrm{v ds}\right|=O(h^{2})\|u\|_{3}\|v\|_{h}, \ v\in\tilde{V}^{0}_{h}, \end{array} $$(65)where πh is the corresponding interpolator over \(\tilde {V}^{0}_{h}, \nabla _{h}\) denotes the piecewise gradient operator, and \(\|.\|_{h}=\left (\sum \limits _ K|.|_{1,K}^{2}\right )^{\frac {1}{2}}\) is the norm on \(\tilde {V}^{0}_{h}\). So, we can also get the results of Theorems 2, 4–6 through (64)–(65).
-
(iii)
For the quasi-Wilson element [31] on rectangular mesh, the modified quasi-Wilson element [32] on arbitrary quadrilateral mesh and the quasi-Carey element [33] on triangular mesh, since their consistency error estimations can reach order of O(h2) when the exact solution (u, v) belongs to \({H_{0}^{1}}({\Omega })\bigcap H^{3}({\Omega }),\) it can be proved that Theorems 2, 4–6 are also valid to these finite elements.
However, for the rectangular Wilson element [34] and the triangular Carey element [35], how to get the desired results of our work still remains open, for their consistency error estimations only can reach order of O(h).
5 Numerical experiment
In this section, we present numerical example to demonstrate the theoretical analysis. Setting the domain Ω = [0, 1] × [0, 1], and the finial time T = 1. Then, we consider the following problem:
where f1, f2, u0, and v0 are computed from the exact solution as follows:
In order to confirm the superclose and superconvergence orders in Theorems 2, 4–6, we choose H2 = h and use Newton iterations on coarse mesh in our computation. It can be seen from Tables 1, 2, 3, 4, 5, and 6 that \(\left \|I_{h}u^{n}-\tilde {U}_{h}^{n}\right \|_{1}, \left \|u^{n}-I_{2h}^{2}\tilde {U}_{h}^{n}\right \|_{1}, \left \|I_{h}v^{n}-\tilde {V}_{h}^{n}\right \|_{1}\), and \(\left \|v^{n}-I_{2h}^{2}\tilde {V}_{h}^{n}\right \|_{1}\) are convergent at order of O(h2) for B-E and C-N schemes, respectively, which coincide with our theoretical analysis. At the same time, we present the error reduction results at t = 0.1,0.5,and 1 in Figs. 1, 2, and 3, respectively,where errU stands for \(\|u^n-I^2_{2h}\tilde {U}_h^n\|_1\) and errV stands for \(\|v^n-I^2_{2h}\tilde {V}_h^n\|_1\).
On the other hand, we also compare the CPU cost of the Galerkin FEMs to the TGMs for B-E scheme in Table 7 with the same partition (h = 1/16) and for C-N scheme in Table 8 with the same partition (h = 1/36) on a different time level. We can see that the TGMs take almost half as much CPU time as the Galerkin FEMs. Therefore, the proposed TGMs are very efficient algorithms.
References
Capasso, V., Wilson, R.E.: Analysis of a reaction-diffusion system modeling man-environment-man epidemics. SIAM J. Appl. Math. 57, 327–346 (1997)
Liu, B.P., Pao, C.V.: Almost periodic plane wave solutions for reaction diffusion equations. J. Math. Anal. Appl. 105, 231–249 (1985)
Bai, Z.G.: A periodic reaction-diffusion system modelling man-environment-man epidemics. Int. J. Biomath. 10, 327–344 (2017)
Shi, D.Y., Pei, L.F.: Analysis of a nonconforming finite element method for bacterial model. Acta Math. Appl. Sin. 34, 428–439 (2011)
Chang, L.L., Jin, Z.: Efficient numerical methods for spatially extended population and epidemic models with time delay. Appl. Math. Comput. 316, 138–154 (2018)
Capasso, V., Maddalena, L.: Convergence to equilibrium states for a reaction diffusion system modelling the spatial spread of a class of bacterial and viral diseases. J. Math. Biol. 13, 173–184 (1981)
Capasso, V.: Asymptotic stability for an integrodifferential reaction-diffusion system. J. Math. Anal. Appl. 103, 575–588 (1984)
Capasso, V., Kunisch, K.: A reaction-diffusion system arising in modelling man-environment diseases. Q. Appl. Math. 46, 431–450 (1988)
Capasso, V., Anita, S.: A stabilizability problem for a reaction-diffusion system modelling a class of spatially structured epidemic systems. Nonlinear Anal. Real World Appl. 3, 453–464 (2002)
Xu, J.C.: A novel two-grid method for semilinear elliptic equations. SIAM. J. Sci. Comput. 15, 231–237 (1994)
Xu, J.C.: Two-grid discretization techniques for linear and nonlinear PDEs. SIAM J. Numer. Anal. 33, 1759–1777 (1996)
Marion, M., Xu, J.C.: Error estimates on a new nonlinear Galerkin method based on two-grid finite elements. SIAM J. Numer. Anal. 32, 1170–1184 (1992)
Shi, D.Y., Yang, H.J.: Unconditional optimal error estimates of a two-grid method for semilinear parabolic equation. Appl. Math. Comput. 310, 40–47 (2017)
Shi, D.Y., Mu, P.C., Yang, H.J.: Superconvergence analysis of a two-grid method for semilinear parabolic equations. Appl. Math. Lett. 84, 34–41 (2018)
Chen, L.P., Chen, Y.P.: Two-grid method for nonlinear reaction diffusion equations by mixed finite element methods. J. Sci. Comput. 49, 383–401 (2011)
Chen, Y.P., Chen, L.P., Zhang, X.C.: Two-grid method for nonlinear parabolic equations by expanded mixed finite element methods. Numer. Methods Partial Differential Equations 29, 1238–1256 (2013)
Chen, C.J., Li, K., Chen, Y.P.: Two-grid finite element methods combined with Crank-Nicolson scheme for nonlinear Sobolev equations. Adv. Comput. Math. 45, 1–20 (2018)
He, Y.N.: Two-level method based on finite element and Crank-Nicolson extrapolation for the time-dependent Navier-Stokes equations. SIAM J. Numer. Anal. 41, 1263–1285 (2014)
He, Y.N., Liu, K.M.: A multilevel finite element method in space-time for the Navier-Stokes problem. Numer. Methods Partial Differential Equations 21, 1052–1078 (2005)
Zhong, L.Q., Shu, S., Wang, J.X., Xu, J.: Two-grid methods for time-harmonic Maxwell equations. Numer. Linear Algebra Appl. 20, 93–111 (2013)
Shi, D.Y., Wang, F.L., Fan, M.Z.: A new approach of the lowest-order anisotropic mixed finite element high-accuracy analysis for nonlinear sine-Gordon equations. Math. Numer. Sin. 37, 148–161 (2015)
Thomèe, V.: Galerkin Finite Element Methods for Parabolic Problems, 2nd edn. Springer, Berlin (1984)
Lin, Q., Lin, J.F.: Finite Element Method : Accuracy and Improvement. Science Press, Beijing (2006)
Shi, D.Y., Wang, J.J.: Superconvergence analysis of an H1-Galerkin mixed finite element method for Sobolev equations. Comput. Math. Appl. 72, 1590–1602 (2016)
Thomèe, V., Xu, J.C., Zhang, N.Y.: Superconvergence of the gradient in piecewise linear finite-element approximation to a parabolic problem. SIAM J. Numer. Anal. 26, 553–573 (1989)
Rannacher, R., Turek, S.: Simple nonconforming quadrilateral Stokes element. Numer. Methods Partial Differential Equations 8, 97–111 (1992)
Shi, D.Y., Peng, Y.C., Chen, S.C.: Error estimates for rotated \(Q^{\text {rot}}_{1}\) element approximation of the eigenvalue problem on anisotropic meshes. Appl. Math. Lett. 22, 952–959 (2009)
Lin, Q., Tobiska, L., Zhou, A.H.: Superconvergence and extrapolation of nonconforming low order finite elements applied to the Poisson equation. IMA J. Numer. Anal. 25, 160–181 (2005)
Shi, D.Y., Mao, S.P., Chen, S.C.: An anisotropic nonconforming finite element with some superconvergence results. Comput. Math. 23, 261–274 (2005)
Hu, J., Man, H.Y., Shi, Z.C.: Constrained nonconforming rotated Q1 element for stokes flow and planar elasticity. Math. Number. Sin. 23, 311–324 (2005)
Shi, D.Y., Wang, F.L., Zhao, Y.M.: Superconvergence analysis and extrapolation of quasi-Wilson nonconforming finite element method for nonlinear Sobolev equations. Acta. Math. Appl. Sin. 29, 403–414 (2013)
Shi, D.Y., Pei, L.F.: Nonconforming quadrilateral finite element method for a class of nonlinear sine-Gordon equations. Appl. Math. Comput. 219, 9447–9460 (2013)
Shi, D.Y., Hao, X.B.: Accuracy analysis for quasi-Carey element. J. Syst. Sci. Complex. 21, 456–462 (2008)
Shi, Z.C., Jiang, B., Xue, W.M.: A new superconvergence property of Wilson nonconforming finite element. Numer. Math. 78, 259–268 (1997)
Shi, D.Y., Pei, L.F.: Convergence analysis of the nonconforming triangular Carey element for a kind of nonlinear parabolic integro-differential problems. J. Sys. Sci. Math. Scis. 29, 854–864 (2009)
Funding
This work was supported by the National Natural Science Foundation of China (Nos. 11671369 and 11271340).
Author information
Authors and Affiliations
Corresponding author
Additional information
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., Li, C. Superconvergence analysis of two-grid methods for bacteria equations. Numer Algor 86, 123–152 (2021). https://doi.org/10.1007/s11075-020-00882-0
Received:
Accepted:
Published:
Issue Date:
DOI: https://doi.org/10.1007/s11075-020-00882-0