An Efficient Numerical Technique for Solving Time-Fractional Generalized Fisher's Equation

This paper extends the existing Fisher's equation by adding the source term and generalizing the degree β of the non-linear part. A numerical solution of a modified Fisher's equation for different values of β using the cubic B-spline collocation scheme is also investigated. The fractional derivative in a time dimension is discretized in Caputo's form based on the L1 formula, while cubic B-spline basis functions are used to interpolate the spatial derivative. The non-linear part in the model is linearized by the modified formula. The efficiency of the proposed scheme is examined by simulating four test examples with different initial and boundary conditions. The effect of different parameters is discussed and presented in tables and graphics form. Moreover, by using the Von Neumann stability formula, the proposed scheme is shown to be unconditionally stable. The results of error norms reflect that the present scheme is suitable for non-linear time fractional differential equations.


INTRODUCTION
Fractional calculus-based models have been used in different fields of engineering and science. In the last few years, fractional differential equations have been widely used. The main advantage of using fractional order differential equation is its non-local property in mathematical modeling. During the twentieth century, the authors [1][2][3] added a significant amount of research in the area of fractional calculus. The applications can be seen in different branches of science and engineering, such as finance [4], nano-technology [5], electrodynamics [6], and visco-elasticity. Fisher's equation is commonly used in epidemics and bacteria, branching Brownian motion, neolithic transitions and chemical kinetics [7][8][9]. The spatial and temporal propagation of a virile gene in an infinite medium has been explained by Fisher [10]. Several numerical methods for differential equations with Riemann-Liouville and Caputo sense fractional order derivatives have been applied and analyzed [11][12][13].
The time-fractional Fisher's equation used in Baranwal et al. [14] has been modified in this paper in two different ways: (1) by introducing the source term or (2) by generalizing the non-linear power.
The modified form of time fractional Fisher's equation is: ∂t α − ν ∂ 2 Z(r, t) ∂r 2 − Z(r, t)(1 − Z β (r, t)) = f (r, t), a ≤ r ≤ b, 0 < α ≤ 1, t ≥ 0, (1.1) with the initial condition Z(r, 0) = ψ(r), a ≤ r ≤ b, (1.2) and the boundary conditions Z(a, t) = ψ 1 (t), Z(b, t) = ψ 2 (t), t ≥ 0, (1.3) where ν is a parameter of viscosity. The Caputo and Riemann-Liouville fractional derivatives have a wide range of applications [15][16][17]. The Caputo derivative is used in this work: The Caputo derivative is discretized by the L1 formula [18]: In this paper, we generalized the linearization formula used in [19]: where β is a positive integer. The numerical and analytical solution of fractional order PDEs play an important role in explaining the characteristics of non-linear problems that arise in everyday life. In the literature, researchers applied various techniques for the numerical solutions of Fisher's equation. Baranwal et al. [14] introduced an analytic algorithm for solving non-linear time-fractional reaction diffusion equations based on the variational iteration method (VIM) and Adomian decomposition method (ADM). Wazwaz and Gorguis [20] implemented ADM for the analytic study of Fisher's equation. Homotopy perturbation sumudu transform method has been applied for solving fractional nonlinear dispersive equations by Abedle-Rady et al. [21]. Gupta and Saha Ray [22] implemented two methods. Haar wavelet method and the optimal homotopy asymptotic method (OHAM) for the numerical solutions of arbitrary order PDE, such as Burger-Fisher's and generalized Fisher's equations. Cherif et al. [23] solved space-fractional Fisher's equation using classical HPM. Khader and Saad [24] proposed a numerical solution for solving the space-fractional Fisher's equation using Chebyshev spectral collocation technique. Rawashdeh [25] introduced the fractional natural decomposition method (FNDM) to find the analytical and approximate solutions of the non-linear time-fractional Harry Dym equation and the non-linear time-fractional Fisher's equation. Singh [26] introduced an efficient computational method for the approximate solution of a non-linear Lane-Emden-type equation. The numerical solution of fractional vibration equation of large membrane has been investigated in Singh [27] by Jacobi polynomial. The authors in [28] employed the cubic B-spline method for the numerical simulations of time fractional Burgers' and Fisher's equation. Singh et al. [29] constructed a q-homotopy analysis transform method for solving time and space-fractional coupled Burgers' equation. Najeeb et al. [30] used HPM for the analytical solution of time-fractional reaction-diffusion equation. Majeed et al. [28] used B-spline at non-uniform for the construction of craniofacial fractures.
In this paper, we have presented a cubic B-spline (CBS) algorithm for numerical simulation of the time-fractional generalized Fisher's equation. Caputo's time fractional derivative based on the L1 scheme has been discretized by finite difference formula, whereas spatial derivatives are discretized by CBS functions. The present approach is novel for the numerical results of fractional order PDEs and, to the best of our knowledge, any spline solution of the time-fractional generalized Fisher's equation has never yet been studied. Moreover, this scheme is equally effective for homogeneous and non-homogeneous boundary conditions. This article has been presented in the following manner. Section 2 evolves a brief description of temporal discretization, cubic B-spline functions and spatial discretization. In section 4, the stability of the proposed algorithm has been discussed. The discussion on numerical results of four test problems has been reported in section 5. Concluding remarks of this work are given in section 6.

DESCRIPTION OF THE METHOD
Let us consider the interval [a, b] is sub divided into N finite elements of equal spacing h determined by the knots r j , j = 0, 1, 2, 3......., N such that a = r 0 < r 1 < r 2 . . . < r N−1 < r N = b. The cubic B-spline basis function at the grid points is defined as (2.1) From the above basis, the approximation solution Z N (r, t) can be written in terms of linear combination of cubic B-spline base function as follows where ϒ j (t) ′ s are the unknowns to be determined. Four consecutive cubic B-splines are used to construct each element [r j , r j+1 ]. The values of cubic B-splines and its derivatives at the nodal points are given in Table 1. The variation of Z N (r, t) over the typical component [r j , r j+1 ] is given by By plugging the approximation values given in Table 1 into Equation (2.3) at (r j , t n ), The Equation (1.1) yields the following set of fractional order ordinary differential equations.
Here, • represents αth order fractional derivative with respect to time. After some simplification, a recurrence relation for Equation (1.1) with β = 3 can be written as . Moreover, the truncation error ρ n+1 t is bounded as where ̟ is a real constant.

Equation (2.5) is modified as
To acquire unique solution of the system, two extra equations are needed. For this purpose, given boundary conditions are used. Thus, the system of linear equations for expression (2.7) becomes (2.9)

INITIAL VECTOR
For the initial vector, the initial and boundary conditions of the problem under consideration will help to compute the initial To determine ϒ 0 , the approximation for the derivatives of the initial and boundary conditions is as follows [32]:

STABILITY ANALYSIS
The von Neumann analysis is frequently used to determine the requirements of stability, as it is usually simple to apply in a simple way. The solution in single Fourier mode is defined as The approximation solution of generalized Fisher's equation (2.7) can be written as By inserting values of α 0 , α 1 and n 1 , n 2 , n 3 in above expression, we have The applied scheme is stable if augment factor |ϒ k+1 | ≤ 1, and, from the above expression, we can observe that value of numerator is lesser than denominator for the values of γ , η, h. The scheme become unstable as the approximations grows in magnitude.
The above result thus reflects that scheme is unconditionally stable.

APPLICATIONS AND DISCUSSION
This section presents some examples with different initial and boundary conditions. The numerical results are presented graphically and numerically in figures and tables. The error norms L 2 and L ∞ are computed to analyze the precision of the suggested technique as In this manuscript we used, MATLAB 2015b on Intel R CORE TM i5 CPU with 8GB RAM and 64-bit operating system (window 7) for numerical simulations.
and the source term Frontiers in Physics | www.frontiersin.org The approximate solution (2.3) can be written in piecewise form: The exact solution of (5.1) is Z(r, t) = t 2α (1 − r 2 ) exp(2r). Figures 1, 2 explores the comparison of CBS solution with exact solution for Example 5.1 for different parameters. Figure 1A shows the 2-dimensional preview of approximate and exact results for t = 0.25 with α = 0.95, h = 0.01, t = 0.0003 and ν = 1. The graph illustrates that exact and approximate outcomes are indiscriminately similar to each other. Figure 1B cites the action of solution obtained for Equation (5.1) with α = 0.95, h = 0.01, ν = 1 and for various time steps t = 0.5, 0.75, and 1 with t = 0.0003. It is clear from the graph that both solutions are overlapping. Three dimensional preview has been given in Figure 2. While the influence of α has been discussed for distinct Brownian motion, i.e, α = 0.25, 0.5, and 0.98 in Figure 3. It can be observed that as the value of α increases, the solution profile decreases and as α → 1, the numerical solution tends to overlap the exact solution. The comparison of numerical and exact outcomes is expressed in Table 2, which shows that both results are consistent with each other and are accurate up to 5 decimal places. The numerical results for α variation is presented in Table 3. It is clear from tabular data that both results strongly agree with each other, and the accuracy of the scheme is examined by the error norms as shown in Table 4.
source term is The Exact solution of Example 5.2 is Z(r, t) = (1 + t 2 )r 2 exp(2r).    0.95, h = 0.01, t = 0.0003, and ν = 1 is compatible with exact solution. Figure 4B shows the effect of various time steps t = 0.5, 0.75, and 1 on the solution profile. It is clear from the graphics that exact and numerical solutions have identical behavior for fixed value of α = 0.95. The comparison of exact and approximate results is presented in Table 5, which clearly shows that both solutions are very close to each other and have negligible errors. Figure 5 give 3D preview of approximate solution. To examine the accuracy of the present technique, error norms are computed and shown in Table 6. The approximate solution (2.3) can be written in piecewise form: IC : Z(r, 0) = 0, 0 < r < 1,   The source term +4π 2 t 4 sin(2πr) − (t 4 sin(2πr))(1 − t 4 sin(2πr)).
Exact solution for above conditions is Z(r, t) = t 4 sin(2πr).
Frontiers in Physics | www.frontiersin.org   Z(r, t n ) = ϒ j−3 φ 3,j−3 (r) + ϒ j−2 φ 3,j−2 (r) +ϒ j−1 φ 3,j−1 (r) + ϒ j φ 3,j (r), r ∈ [r j , r j+1 ). (5.8)   Figure 6A, displays the numerical and exact solution of Example 5.3 for t = 0.4, α = 0.96, h = 0.01 and t = 0.0001. The graphics illustrate that numerical and exact solutions are obviously shown to be indiscriminately comparable to one another. The effect of time concentrations t = 0.6, 0.8, and 1 is studied and presented in Figure 6B keeping other parameters constant. It can be seen from graphics that both solutions have symmetrical conduct and their corresponding numerical data are presented in Table 7, which demonstrates that both results are accurate and have negligible error. Figure 7 plots three-dimensional solution and results of error norms is given in Table 8. The influence of Brownian motion, i.e, α = 0.25, 0.75, on solution curve is displayed in Figure 8. The identical behavior of solution curves demonstrates that for smaller values of α, the solution profile is away from the exact result and as α → 1, the approximate and exact solution tends to overlap. Example 5.4. Fisher's equation with fractional order for β = 1 with f (r, t) = 0, is ∂ α Z(r, t) ∂t α − ν ∂ 2 Z(r, t) ∂r 2 − Z(r, t)(1 − Z(r, t)) = f (r, t). (5.10) with IC : Z(r, 0) = σ * , 0 ≤ r ≤ 1.
The graphical illustration of exact and numerical solutions for Example 5.4 are shown in Figure 9. Figure 9A shows compatibility of exact and numerical results for h = 0.01, t = 0.02, α = 1, and σ * = 0.25. The multiple curves for exact and numerical solutions for various values of σ * = 0.5, 0.7, and 0.9 are shown in Figure 9B. The comparison of exact and approximate solutions acquired by the proposed scheme is expressed in Table 9. The tabular data demonstrate that both solutions are compatible with each other for various values of σ * . Table 10 demonstrates the error norms.

CONCLUDING REMARKS
In this study, cubic B-spline (CBS) scheme has been successfully implemented to acquire numerical solution of