The exponential cubic B-spline algorithm is presented to find the numerical solutions of the Korteweg-de Vries (KdV) equation. The problem is reduced to a system of algebraic equations, which is solved by using a variant of Thomas algorithm. Numerical experiments are carried out to demonstrate the efficiency of the suggested algorithm.
1. Introduction
The splines consist of piecewise functions defined on the distributed knots on problem domain and have certain continuity inside problem subdomain and at the knots. Until now, some types of splines have been developed and especially polynomial splines. The exponential splines are defined as more general splines by McCartin [1–3]. The basis of the exponential splines known as the exponential B-splines is also given in the studies of McCartin. Existence of the free parameter in the exponential B-splines yields the different shapes of the splines functions. He has also showed a reliable algorithm by using the exponential spline functions to solve the hyperbolic conservation laws, McCartin and Jameson [4]. However, McCartin stated that application of the exponential spline/exponential B-spline functions has been neglected in the numerical analysis. So, use of the exponential spline in the numerical methods for finding solutions of the differential equations is not common and few papers exist in the literature. McCartin has shown that the exponential splines admit a basis known as the exponential B-splines. These B-splines have been started using to form approximates functions recently which are adapted to set up the numerical methods to find solutions of the differential equations recently. An application of the simple exponential splines is considered for setting up the collocation method to solve the numerical solution of singular perturbation problem [5]. Cardinal exponential B-splines are applied in solving singularly perturbed boundary problems [6]. A variant of B-spline exponential collocation method was also built up for computing numerical solutions of the singularly perturbed boundary value problem [7]. Very recently, the exponential B-spline collocation method has been applied to obtain the solutions of the one-dimensional linear convection-diffusion equation [8].
Types of spline functions are utilized to form approximate solutions for Korteweg-de Vries equation (KdVE). The standard Galerkin formulation using the smooth splines on uniform mesh is set up for 1-periodic solutions of KdVE by Baker and his coauthors [9]. The Galerkin finite element method together with the cubic B-splines is used to solve the KdVE in the paper [10]. The quadratic B-spline Galerkin method is built up to find solutions of the KdVE [11]. A collocation solution of the Korteweg-de Vries equation using septic B-splines is proposed by Soliman [12]. A variant of the Galerkin finite element method is designed for solving the KdVE by Aksan and Özdeş [13]. The collocation method using quintic B-splines is developed to solve the KdVE [14]. A numerical method is developed for the KdVE by using splitting finite difference technique and quintic B-spline functions [15]. The spline finite element method using quadratic polynomial spline for the numerical solution of the KdVE is given by G. Micula and M. Micula [16]. A cubic B-spline Taylor-Galerkin method is developed to find numerical solution of the KdVE by Canıvar et al. in [17]. A study based on cubic B-spline finite element method for the solution of the KdVE is suggested by Kapoor et al. [18]. A Bubnov-Galerkin finite element method with quintic B-spline functions taken as element shape and weight functions is presented for the solution of the KdVE [19]. The paper deals with the numerical solution of the KdVE using quartic B-splines Galerkin method as both shape and weight functions over the finite intervals [20]. A blended spline quasi-interpolation scheme is employed to solve the one-dimensional nonlinear KdVE [21]. A multilevel quartic spline quasi-interpolation scheme is fulfilled to exhibit a large number of physical phenomena for KdVE [22].
The aim of the present paper is to develop an approximate solution of KdVE by collocation method. In Section 2, the exponential B-spline collocation algorithm is defined for the KdVE. In Section 3, the three numerical experiments are constructed to demonstrate the efficiency of the proposed method and the results are documented in tables and graphs are depicted.
We will solve the KdVE:
(1)Ut+εUUx+μUxxx=0,a≤x≤b,
where ε, μ are positive parameters and the subscripts x and t denote differentiation. The boundary conditions will be chosen as
(2)U(a,t)=U(b,t)=0,Ux(a,t)=Ux(b,t)=0.
KdVE is prototypical example of exactly solvable mathematical model of waves on shallow water surface. It arises for evolution, interaction of waves, and generation in physics. Due to the term Ut, (1) is called the evolution equation, the nonlinear term causes the steepness of the wave, and the dispersive term defines the spreading of the wave. It is known that the effect of the steepness and spreading results in soliton solutions for the KdVE.
2. Exponential B-Spline Collocation Method
The region [a,b] is partitioned into equal subintervals by points xi, i=0,…,N. On these points together with additional points xi, i=-3,-2,-1,N+1,N+2,N+3 outside the domain, the exponential B-splines, Bi(x), can be defined as
(3)Bi(x)=b2xi-2-x-1psinhpxi-2-xxi-2,xi-1,a1+b1xi-x+c1exppxi-x+d1exp-pxi-xxi-1,xi,a1+b1(x-xi)+c1exppx-xi+d1exp-px-xixi,xi+1,b2(x-xi+2)-1psinhpx-xi+2xi+1,xi+2,0otherwise,
where
(4)a1=phcphc-s,b1=p2cc-1+s2phc-s1-c,b2=p2(phc-s),c1=14exp-ph1-c+sexp-ph-1phc-s1-c,d1=14expphc-1+sexpph-1phc-s1-c,s=sinhph,c=coshph,pisafreeparameter,h=b-aN.
{B-1(x),B0(x),…,BN+1(x)} forms a basis for the exponential spline space on the interval [a,b]. On the four consecutive subintervals, an exponential B-spline Bi(x) is defined and it is second-order continuously differentiable functions.
In Table 1, the values of Bi(x), Bi′(x), and Bi′′(x) at the points xi, which can be obtained from (3) are listed where ′ denotes differentiation with respect to space variable x.
Values of Bi(x) and its first and second derivatives at the knot points.
x
xi-2
xi-1
xi
xi+1
xi+2
Bi
0
s-ph2phc-s
1
s-ph2phc-s
0
Bi′
0
p1-c2phc-s
0
pc-12phc-s
0
Bi′′
0
p2s2phc-s
-p2sphc-s
p2s2phc-s
0
The global approximation UN(x,t) to the solution U(x,t) will be searched in terms of the unknown parameters δi and exponential B-spline function defined on the problem domain:
(5)UNx,t=∑i=-1N+1δiBix.
Substitution of the points xi in (5), in its first and its second derivatives, respectively, yields the numerical solution in terms of parameters:
(6)Ui=U(xi,t)=m1δi-1+δi+m1δi+1,Ui′=U′(xi,t)=m2δi-1-m2δi+1,Ui′′=U′′xi,t=m3δi-1-2m3δi+m3δi+1,
where m1=(s-ph)/2(phc-s), m2=p(1-c)/2(phc-s), and m3=p2s/2(phc-s).
Over the subregion [xi,xi+1], the local approximation is given by
(7)UNex,t=δi-1Bi-1x+δiBix+δi+1Bi+1x+δi+2Bi+2x,
where δi-1, δi, and δi+1 act as subregion parameters and Bi-1, Bi, and Bi+1 are known as the subregion shape parameters.
To be able to apply the collocation method formed with the exponential B-splines, KdV equation is space-splitted as
(8)Ut+εUUx+μVxx=0,Ux-V=0.
This system includes the second-order derivatives so that smooth approximation can be done with the exponential B-splines. To integrate system (8) in time, discretize Ut by the usual finite difference scheme and U, Ux, and Vxx by Crank-Nicolson method and we get
(9)Un+1-UnΔt+εUUxn+1+UUxn2+μVxxn+1+Vxxn2=0,Uxn+1+Uxn2-Vn+1+Vn2=0,
where Un+1=U(x,(n+1)Δt) represent the solution at the (n+1)th time level. Here, tn+1=tn+Δt and Δt is the time step; superscripts denote nth time level, tn=nΔt.
One linearizes terms (UUx)n+1 and (UUx)n in (9) as
(10)UUxn+1=Un+1Uxn+UnUxn+1-UnUxn,UUxn=UnUxn
to obtain the time-integrated KdVE:
(11)Un+1-Un+εΔt2Un+1Uxn+UnUxn+1-μΔt2Vxxn+1+Vxxn=0,Uxn+1+Uxn2-Vn+1+Vn2=0.
We approximate Un and Vn in terms of the element parameters and exponential B-splines separately as
(12)UNx,t=∑i=-1N+1δiBix,VNx,t=∑i=-1N+1ϕiBix.
Putting the approximate solution (12) and its derivatives into (11) and evaluating the resulting equations at the points xi, i=0,…,N yield the following system of equations:
(13)m1δm-1n+1-μΔt2γϕm-1n+1+δmn+1+μΔtγϕmn+1+m1δm+1n+1-μΔt2γϕm+1n+1=m11-εΔt2Lδm-1n-Δt2μγ+εm1Kϕm-1n+1-εΔt2Lδm-1n-Δt2-2μγ+εKϕm-1n+m11-εΔt2Lδm-1n-Δt2μγ+εm1Kϕm-1n,βδm-1n+1-m1ϕm-1n+1-ϕmn+1-βδm+1n+1-m1ϕm+1n+1=-βδm-1n+m1ϕm-1n+ϕmn+βδm+1n+m1ϕm+1n,m=0,…,N,n=0,1,…,
where
(14)K=m1δi-1+δi+m1δi+1,L=m1ϕi-1+ϕi+m1ϕi+1,β=p(1-c)2(phc-s),γ=p2s2(phc-s).
The system consists of 2N+2 linear equation in 2N+6 unknown parameters dn+1=(δ-1n+1,ϕ-1n+1,δ0n+1,ϕ0n+1,…,δN+1n+1,ϕN+1n+1). A unique solution of the system can be obtained by imposing the boundary conditions U(a,t)=0, U(b,t)=0, V(a,t)=0, V(b,t)=0 to have the following the equations:
(15)m1δ-1+δ0+m1δ1=0,m1ϕ-1+ϕ0+m1ϕ1=0,m1δN-1+δN+m1δN+1=0,m1ϕN-1+ϕN+m1ϕN+1=0.
Elimination of the parameters δ-1, ϕ-1, δN+1, ϕN+1, using (15) from system (13), gives a solvable system of 2N+2 linear equation including 2N+2 unknown parameters. Placing solution parameters in (12) when computed from the system via a variant of the Thomas algorithm gives the approximate solution over the subregion [a,b]. We need the initial parameter vectors d1=(δ-1,δ0,…,δN,δN+1), d2=(ϕ-1,ϕ0,…,ϕN,ϕN+1) to start the iteration process for system of (13). To do that, the following requirements help to determine the initial parameters:
(16)UNx(a,0)=0=m2δ-10+m2δ10,UNx(xi,0)=m1δi-10+δi0+m1δi+10=U(xi,0),i=0,…,N,UNx(b,0)=0=m2δN-10+m2δN+10,VNxa,0=0=m2ϕ-10+m2ϕ10,VNx(xi,0)=m1ϕi-10+ϕi0+m1ϕi+10=V(xi,0),i=0,…,N,VNxa,0=m2ϕN-10+m2ϕN+10.
3. Numerical Tests
Since the conservation laws remain constant at all time, first three numerical conservations are calculated using the rectangular rule for integrals:
(17)C1=∫-∞∞Udx≈∫abUdx≈h2∑i=0N-1Ui+Ui+1,C2=∫-∞∞U2dx≈∫abU2dx≈h2∑i=0N-1U2i+U2i+1,C3=∫-∞∞U3-3μεUx2dx≈∫abU3-3μεUx2dx≈h2∑i=0N-1U3i-3μεUx2i+U3i+1-3μεUx2i+1.
The error norm L∞(18)L∞=U-UN∞=maxjUj-UNjn
is calculated to show the error between analytical and numerical solutions.
(a) The soliton solution of the KdVE is
(19)Ux,t=3csech2Ax-Bt+D,
where A=(1/2)εc/μ and B=εcA. This solution represents propagation of single soliton, having velocity εc and amplitude 3c.
The analytical solution (19) is used as the initial condition when t=0. The Dirichlet boundary conditions U(0,t)=V(0,t)=0 and U(2,t)=V(2,t)=0 are adapted to the system to control numerical solutions at the boundaries. Parameters ε=1, μ=4.84×10-4, c=0.3, D=-6, space step h=0.01, and time step Δt=0.005 on the interval [0,2] from time t=0 to t=3 are chosen. At time t=3, numerical magnitude of the single soliton is calculated as 0.9, so that the numerical amplitude is obtained to be almost the same as the analytical amplitude. Figure 1 illustrates the amplitudes at some times. The distribution of the absolute values of errors can be observed in Figure 2.
A single soliton at some times.
Error distribution.
L∞ error norms and invariants are presented at the selected times in Table 2, as seen from the table that L2 error norm is found small enough and conservation invariants are excellent throughout the simulation. The method gives good results when the free parameter p=1.64×10-5 is used. Invariants C2 and C3 remain constant during the run and C1 remains the constant up to the third decimal digits seen in Table 2.
Error norms and invariants for single soliton.
Time
L∞×103p×105
L∞×103p=1
C1
C2
C3
Method
0.0
0
0
0.144597
0.086759
0.046849
Pre.
0.5
0.521 (p=1.37)
0.51
0.144602
0.086759
0.046849
1.0
0.589 (p=1.64)
0.75
0.144593
0.086759
0.046849
1.5
0.533 (p=1.64)
1.03
0.144592
0.086759
0.046849
2.0
0.595 (p=1.64)
1.26
0.144591
0.086759
0.046849
2.5
0.657 (p=1.64)
1.50
0.144597
0.086759
0.046849
3.0
0.740 (p=1.64)
1.61
0.144597
0.086759
0.046849
3.0
0.08
0.014460
0.08676
0.04685
[23]
3.0
1.04
0.014460
0.08675
0.04685
[24]
3.0
0.04
0.144597
0.086761
0.0468524
[17]
3.0
0.14
0.144601
0.086760
0.046850
[25]
(b) Wave generation is performed by using the Maxwellian initial condition:
(20)U(x,0)=exp(-x2)
and boundary conditions:
(21)U(-15,t)=U(15,t)=0,t>0.
h=0.1 and Δt=0.01 and ε=1.0 are taken. We have verified the case in which μc is some critical parameter and, according to the parameters μ≪μc, initial condition breaks up into a number of solitons and, for values μ≫μc, soliton turns into exhibiting the rapidly oscillating wave packets. When μ≈μc together with parameters ε=1.0, μ=0.04, h=0.1, and Δt=0.01, the solution takes the form of the leading soliton and an oscillating tail. This case is shown in Figure 3.
Soliton at t=12.5.
For μ=0.04, we observe a solitary wave plus an oscillating tail (Figure 3). The actual velocity of the wave Vn has been measured and also computed from the measured amplitude using the formula Va=aε/3. We find that Vn=0.4 and Va=1.1978×1/3=0.39926, so the solitary waves are indeed solitons. In Table 3 invariant for Maxwellian ε=1.0 and μ=0.04.
Invariant for Maxwellian.
ε=1.0, μ=0.04
t
C1
C2
C3
0
1.77245
1.25331
0.87292
2.5
1.77243
1.25332
0.87412
5.0
1.77248
1.25333
0.87437
7.5
1.77170
1.25332
0.87442
10.0
1.77203
1.25333
0.87442
12.5
1.77367
1.25330
0.87442
When μ=0.01, we find three solitons. We have measured the velocity of the largest solitary wave as Vn=0.52 and calculated the expected velocity from the observed amplitude 1.54468 as Va=1.5535×1/3=0.51783. In Figure 4, Maxwellian initial condition is depicted for t=12.5, ε=1.0, μ=0.01, h=0.1, and Δt=0.01. The invariants are given in Tables 4 and 5 for ε=1.0, μ=0.01 and ε=1.0, μ=0.001, respectively.
Invariant for Maxwellian.
ε=1.0, μ=0.01
t
C1
C2
C3
0
1.77245
1.25331
0.98572
2.5
1.77244
1.25362
0.99277
5.0
1.77250
1.25374
0.99558
7.5
1.77243
1.25376
0.99576
10.0
1.77242
1.25377
0.99582
12.5
1.77222
1.25378
0.99582
Invariant for Maxwellian.
ε=1.0, μ=0.001
t
C1
C2
C3
0
1.77245
1.25331
1.01956
2.5
1.77245
1.25356
1.02461
5.0
1.77245
1.25365
1.02644
7.5
1.77245
1.25365
1.02654
10.0
1.77244
1.25365
1.02656
12.5
1.77244
1.25365
1.02656
Generated waves at t=12.5, ε=1.0, μ=0.01, h=0.1, and Δt=0.01.
For μ=0.001, we observed nine solitons moving to the right in Figure 5. The measured velocity of leading soliton is Vn=0.62 and the corresponding velocities calculated from their measured amplitudes are Va=1.7949×1/3=0.60. The agreement is good. The initial perturbation breaks up into a number of solitons in the course of time depending on the value of μ chosen. So, if we decrease the value of μ, then the number of solitons, amplitude, and the velocity increase.
t=12.5 Maxwellian initial condition; ε=1.0, μ=0.001, h=0.025, and Δt=0.005.
(c) As a final test example, initial condition
(22)U(x,0)=121-tanhx-255
together with boundary conditions
(23)U-50,t=U150,t=0,t>0
cause the production of a train of solitons depending on the value of μ for the KdVE. Computation is done on region [-50,150] up to time t=800 with parameters ε=0.2, μ=0.1, Δt=0.05, and h=0.4. Visual representation of the solution in Figures 6(a)–6(f) is drawn that 10 solitons have been broken up from the given initial condition.
(a) The solution graph for t=0. (b) The solution graph for t=100. (c) The solution graph for t=200. (d) The solution graph for t=400. (e) The solution graph for t=600. (f) The solution graph for t=800.
The first three conservation laws are recorded at some times in Table 6. These are favorably constant. The observed velocity of the leading soliton having the amplitude 1.96342 is Vn=0.128 which was in close agreement with that calculated from its observed amplitude of Va=1.9379×0.2/3=0.129.
Conservation laws ε=0.2, μ=0.1.
t
C1
C2
C3
0
50.0001
45.0004
40.6207
100
49.9999
45.0011
40.4055
200
50.0009
45.0032
40.5267
300
49.9996
45.0044
40.4286
400
49.9989
45.0048
42.6020
500
49.9984
45.0049
40.6564
600
49.9997
45.0051
40.6786
700
50.0009
45.0051
40.4270
800
50.0011
45.0049
40.4629
4. Conclusion
The numerical solution of the KdVE is obtained by the collocation method using the exponential basis functions. Performance of the present method is shown by calculating L∞—the error norm, and conservation laws. The present method gives accurate results and simulations such as the propagation of soliton and generation of waves which are substantiated fairly. Using the exponential cubic B-splines, alternative numerical methods can be set up for finding numerical solutions of the differential equations with high accuracy when an appropriate free parameter is chosen.
Conflict of Interests
The authors declare that there is no conflict of interests regarding the publication of this paper.
McCartinB. J.Theory computation and application of exponential splines1981DOE/ER/03077-171McCartinB. J.Computation of exponential splines199011224226210.1137/0911015MR1037513McCartinB. J.Theory of exponential splines19916611232-s2.0-0039752562MR111357310.1016/0021-9045(91)90050-KMcCartinB. J.JamesonA.Numerical solution of nonlinear hyperbolic conservation laws using exponential splines19906277912-s2.0-3424995617710.1007/BF00350514SakaiM.UsmaniR. A.A class of simple exponential B-splines and their application to numerical solution to singular perturbation problems198955549350010.1007/BF01398912MR9989062-s2.0-0000924982RadunovicD.Multiresolution exponential B-splines and singularly perturbed boundary problem200847219121010.1007/s11075-008-9171-1MR23832652-s2.0-40149105817RaoS. C.KumarM.Exponential B-spline collocation method for self-adjoint singularly perturbed boundary value problems20085810157215812-s2.0-4974910321510.1016/j.apnum.2007.09.008MR2454966MohammadiR.Exponential B-spline solution of convection-diffusion equations20134693394410.4236/am.2013.46129BakerG. A.DougalisV. A.KarakashianO. A.Convergence of Galerkin approximations for the Korteweg-de Vries equation19834016241943310.2307/2007524MR689464GardnerG. A.GardnerL. R. T.A finite element solution for the Korteweg de vries equation using cubic B-spline shape functions1Proceedings of the International Conference on Modelling and Simulation1988Tassin-la-Demi-Lune, FranceAMSE Conferance PressGardnerL. R.GardnerG. A.AliA. H.Simulations of solitons using quadratic spline finite elements19919222312432-s2.0-002625767710.1016/0045-7825(91)90241-WMR1137557SolimanA. A.Collocation solution of the Korteweg-de Vries equation using septic splines200481332533110.1080/00207160410001660817MR21749942-s2.0-28244457842AksanE. N.ÖzdeşA.Numerical solution of Korteweg-de Vries equation by Galerkin B-spline finite element method200617521256126510.1016/j.amc.2005.08.038MR22203272-s2.0-33645933912ZakiS. I.A quintic B-spline finite element scheme for the KdVB equation2006190532547JainP. C.ShankarR.BhardwajD.Numerical solution of the Korteweg-de Vries (KDV) equation19978694395110.1016/S0960-0779(96)00135-XMR14560082-s2.0-0041627431MiculaG.MiculaM.On the numerical approach of Korteweg-de Vries-Burger equations by spline finite element and collocation methods20023261270MR1929770CanıvarA.SariM.DagI.A Taylor-Galerkin finite element method for the KdV equation using cubic B-splines2010405163376338310.1016/j.physb.2010.05.0082-s2.0-77955305115KapoorS.RawatS.DhawanS.Numerical investigation of separated solitary waves solution for KdV equation through finite element technique20124014273310.5120/5049-7463AmeinN. K.RamadanM. A.A small time solutions for the KdV equation using Bubnov-Galerkin finite element method201119311812510.1016/j.joems.2011.10.005MR2924323SakaB.DagI.Quartic B-spline GALerkin approach to the numerical solution of the KdVB equation2009215274675810.1016/j.amc.2009.05.059MR25615322-s2.0-68949149765YuR.WangR.ZhuC.A numerical method for solving KdV equation with blended b-spline quasi-interpolation201310165093510110.12733/jics201021802-s2.0-84888317385YuR.-G.WangR.-H.ZhuC.-G.A numerical method for solving KdV equation with multilevel B-spline quasi-interpolation20139281682169010.1080/00036811.2012.698267MR31691252-s2.0-84880394618ZakiS. I.A quintic B-spline finite elements scheme for the KdVB equation2000188112113410.1016/s0045-7825(99)00142-52-s2.0-0034227299SakaB.Cosine expansion-based differential quadrature method for numerical solution of the KdV equation20094052181219010.1016/j.chaos.2007.10.004MR25332112-s2.0-65649088728DağDereliY.Numerical solutions of KdV equation using radial basis functions200832453554610.1016/j.apm.2007.02.001MR23886692-s2.0-37248998931