Linear B-spline finite element method for the generalized diffusion equation with delay

The main aim of this paper is to develop a linear B-spline finite element method for solving generalized diffusion equations with delay. The linear B-spline basis function is used to discretize the space variable. The time discretization process is based on Crank-Nicolson. The benefit of the scheme is that the numerical solution is obtained as a smooth piecewise continuous function which empowers one to find an approximate solution at any desired position in the domain. Sufficient and necessary conditions for the numerical method to be asymptotically stable are derived. The convergence of the numerical method is studied. Some numerical experiments are performed to verify the applicability of the numerical method.


Introduction
In this paper, we consider a class of the generalized delay diffusion equation of the form with a 1 , a 2 ∈ R are real numbers and τ > 0 is a delay constant. The delay diffusion equation has several applications in science and engineering [1][2][3][4][5]. The generalized delay diffusion equation has intrinsic complex nature because its exact solutions are difficult to obtain. Therefore, one has to mostly rely on numerical treatments. Jackiewicz and Zubik-Kowal [6] used spectral collocation and waveform relaxation methods to investigate (1) , t > 0, 0 < x < π, u(x, t) = ψ (x, t), − τ ≤ t ≤ 0, 0 ≤ x ≤ π, u(0, t) = u(π , t) = 0, t > 0, nonlinear partial differential equations with delay. Chen and Wang [7] used the variational iteration method to study a neutral functional differential equation with delays. The numerical treatments of the generalized delay diffusion equations were studied by many authors(see for instance [8][9][10][11]). Test equation of the type Eq. (1) is also considered in [12,13]. In these works, the authors applied the separation of the variables to solve analytically.
The finite element method (FEM) is a well-established numerical method for solving partial differential equations (PDEs). The method approximates the exact solution by using piecewise polynomials or B-spline basis functions. B-splines as finite element basis functions provide the required continuity and smoothness. The use of various degrees of B-spline functions to obtain the numerical solutions of some PDEs has been shown to provide easy and simple algorithms. For instance, B-spline finite elements have been widely applied to solve elliptic equations [14,15], Korteweg-De Vries equation [16][17][18], Burgers' equation [19][20][21][22], regularized long-wave equation [23,24], Fokker-Planck equation [25], advection-diffusion equation [26], and generalized equal width wave equation [27], etc., successfully. However, to the best knowledge of the authors, the B-spline FEM method is not considered for finding the approximate solution of the diffusion equation with delay. In this paper, we have applied a linear B-spline FEM to find numerical solutions to the problem under consideration.
Notations Let H r = H r (�) = W r 2 (�) denotes the Sobolev spaces of order r with respective to norm . r defined as and �ν� = �ν� Applying Green's formula to the second and third terms of equation (1) we obtain Define the space where P 1 is the space of all polynomials degree less or equal to 1.
We can find the approximate solution u h (t) := u h (., t) belonging to S h for each t, so that where ψ h (., t) is an approximation of ψ(., t) in S h .
Let �t = τ/m be a given step size with m ≥ 1 , the grid points t n = n�t(n = 0, 1, . . . ) and U n be the approximation in S h of u(t) at t = t n = n t.

Assumption
Assume , and

Description of the method
Let �t = τ/m be a step size with m ≥ 1 , the grid points t n = n�t(n = 0, 1, . . . ) and be the approximation in S h of u(t) at t = t n = n t . We partition the x -axis into N finite element by choosing a set of equally-spaced knots The linear B-spline basis functions is chosen as follows: which can be rewritten as Define the following matrices: The (N − 1) × (N − 1) matrices A and B are given as follows with γ n = ψ(t n ) an initial approximation and α n := (α 1 , . . . , α N ) T , and B + 1 2 a 1 tA is positive definite and hence, in particular, invertible. Therefore, it has a unique solution.

Stability analysis
then the zero solution of Eq. (5) is called asymptotically stable.
] be an element the finite element, and K := [−1, 1] be the reference element in η -plane. Then The characteristic of Eq. (14) is: where γB −1Ã denotes the corresponding eigenvalue of B −1Ã .

Convergence Analysis
In this section, we present the convergence analysis for the proposed method. The Ritz projection R h : H 1 0 (�) → S h is a mapping for any ν ∈ H 1 0 (�) such that

holds. Then, with R h defined by Eq. (16), we have
The number r is referred to as the order of accuracy of the family {S h } . For the case of piecewise linear B-spline basis function, r = 2.

Theorem 2 Let u and U n be the solution of
where C is a positive constant independent of h and t.

Proof
Define where µ n = U n − D h u(t n ) , σ n = D h u(t n ) − u(t n ) , so that The term σ n (t) = σ (t n ) is easily bounded by lemma 2.

Numerical experiments
The performance of the proposed methods is tested by using numerical experiments. To evaluate errors, L ∞ and L 2 error norms are applied as follows: Order of convergence is obtained by where E h 1 and E h 2 represent the errors at step sizes h 1 and h 2 , respectively.
We apply the proposed method with different step sizes to solve the problem. The graph of numerical results is shown in Fig. 1. This graph shows that the numerical solution is asymptotically stable. And these confirm the theoretical results in Theorem 1. [30] Consider with the initial condition we take the initial function as ψ(x, t) = sin(x) , and the added term h(x, t) where that is the exact solution is u(x, t) = exp(−t)sin(x) . Here, we take the parameters a 1 = 1, a 2 = 0.5, τ = 0.5 and compute the problem on [0, π] × [0, 2] for different space and temporal step sizes (�x = π/N , �t = τ/m). Table 1 shows the numerical errors and the corresponding orders. When the grid size is reduced, both error    The given results suggest that the proposed method has order 2 of accuracy. The calculated error norms are also compared with the result obtained using the central difference method [30]. In Table 2, the comparison between the exact and approximation solution are given.

Conclusion
In this paper, a finite element method is constructed based on linear B-spline basis functions for solving the generalized diffusion equations with delay. The detailed description of results through tables and graphs proves that the proposed numerical method is working efficiently. For all the test cases, simulations at a different set of data points are carried out to check the applicability of the numerical scheme. Based on these observations, our expectation that the given method is well suited to the generalized diffusion with the delay is confirmed.

Limitations
The linear B-spline basis functions yields an order 2 of accuracy. One can use higher polynomial basis functions in order to increase the order of accuracy in space.