The first and second order of accuracy difference schemes for the approximate solution of the initial boundary value problem for ultra-parabolic equations are presented. Stability of these difference schemes is established. Theoretical results are supported by the result of numerical examples.
1. Introduction
Mathematical models that are formulated in terms of ultraparabolic equations are of great importance in various problems for instance in age-dependent population model, in the mathematical model of Brownian motion, in the theory of boundary layers, and so forth, see [1–5]. We refer also to [6–9] and the references therein for existence and uniqueness results and other properties of these models. On the other hand, Akrivis et al. [10] and Ashyralyev and Yılmaz [11, 12] developed numerical methods for ultraparabolic equations. In this paper, our interest is studying the stability of first- and second-order difference schemes for the approximate solution of the initial boundary value problem for ultraparabolic equations∂u(t,s)∂t+∂u(t,s)∂s+Au(t,s)=f(t,s),0<t,s<T,u(0,s)=ψ(s),0≤s≤T,u(t,0)=φ(t),0≤t≤T,
in an arbitrary Banach space E with a strongly positive operator A. For approximately solving problem (1.1), the first-order of accuracy difference schemeuk,m-uk-1,mτ+uk-1,m-uk-1,m-1τ+Auk,m=fk,m,fk,m=f(tk,sm),tk=kτ,sm=mτ,1≤k,m≤N,Nτ=1,u0,m=ψm,0≤m≤N,uk,0=φk,0≤k≤N
and second-order of accuracy difference schemeuk,m-uk-1,mτ+uk-1,m-uk-1,m-1τ+12A(uk,m+uk-1,m-1)=fk,m,fk,m=f(tk-τ2,sm-τ2),tk=kτ,sm=mτ,1≤k,m≤N,Nτ=1,u0,m=ψm,0≤m≤N,uk,0=φk,0≤k≤N
are presented. The stability estimates for the solution of difference schemes (1.2) and (1.3) are established. In applications, the stability in maximum norm of difference shemes for multidimensional ultraparabolic equations with Dirichlet condition is established. Applying the difference schemes, the numerical methods are proposed for solving one-dimensional ultraparabolic equations.
Theorem 1.1.
For the solution of (1.2), we have the following stability inequality:
max1≤k≤Nmax1≤m≤N‖uk,m‖E≤C(max0≤m≤N‖ψm‖E+max0≤k≤N‖φk‖E+max1≤k≤Nmax1≤m≤N‖fk,m‖E),
where C is independent of τ, ψm, φk, and fk,m.
Proof.
Using (1.2), we get
uk,m-uk-1,m-1τ+Auk,m=fk,m.
From that it follows
uk,m=Ruk-1,m-1+τRfk,m,
where R=(I+τA)-1. By the mathematical induction, we will prove that
uk,m=Rnuk-n,m-n+∑j=1nτRn-j+1fk-n+j,m-n+j
is true for all positive integers n. It is obvious that for n=1,2 formula (1.7) is true. Assume that for n=ruk,m=Rruk-r,m-r+∑j=1rτRr-j+1fk-r+j,m-r+j
is true. In formula (1.6), replacing k and m with k-r and m-r, respectively, we have
uk-r,m-r=Ruk-r-1,m-r-1+τRfk-r,m-r.
Then, using (1.8) and (1.9), we get
uk,m=Rr+1uk-r-1,m-r-1+τRr+1fk-r,m-r+∑j=1rτRr-j+1fk-r+j,m-r+j.
From that it follows
uk,m=Rr+1uk-r-1,m-r-1+∑j=1r+1τRr-j+2fk-r-1+j,m-r-1+j
is true for n=r+1. So, formula (1.7) is proved. For m>k, replacing n with k in formula (1.7), we obtain that
uk,m=Rkψm-k+∑j=1kτRk-j+1fj,m-k+j.
Using estimate (see [13])
‖Rk‖E→E≤M
and triangle inequality, we get
‖uk,m‖E≤‖Rk‖E→E‖ψm-k‖E+∑j=1kτ‖Rk-j+1‖E→E‖fj,m-k+j‖E≤M[max0≤k,m≤N‖ψm-k‖+max1≤j≤Nmax1≤m≤N‖fj,m‖]
for any k and m. For k>m, replacing n with m in formula (1.7), we get
uk,m=Rmφk-m+∑j=1mτRm-j+1fk-m+j,j.
From estimate (1.13) and triangle inequality, it follows that
‖uk,m‖E≤‖Rm‖E→E‖φk-m‖E+∑j=1mτ‖Rm-j+1‖E→E‖fk-m+j,j‖E≤M[max0≤k,m≤N‖φk-m‖+max1≤j≤Nmax1≤m≤N‖fj,m‖]
for any k and m. Thus, Theorem 1.1 is proved.
Theorem 1.2.
For the solution of (1.3), we have the following stability inequality:
max1≤k≤Nmax1≤m≤N‖uk,m+uk-1,m2‖E+max1≤k≤Nmax1≤m≤N‖uk,m+uk,m-12‖E≤C(max0≤m≤N‖ψm‖E+max0≤k≤N‖φk‖E+max1≤k≤Nmax1≤m≤N‖fk,m‖E),
where C is independent of τ, ψm, φk, and fk,m.
The proof of Theorem 1.2 is based on the following formulas:uk,m=Buk-1,m-1+τBCfk,m,uk,m=Bkψm-k+∑j=1kτBk-jCfj,m-k+j,m>k,uk,m=Bmφk-m+∑j=1mτBm-jCfk-m+j,j,k>m
for the solution of difference scheme (1.3) and the following estimate [14]:‖BkC‖E→E≤M,
where B=(I-τA/2)(I+τA/2)-1 and C=(I+τA/2)-1.
2. Application
Let Ω be the unit open cube in the n-dimensional Euclidean space ℝn(0<xk<1,1≤k≤n) with boundary S,Ω¯=Ω∪S. In [0,1]×[0,1]×Ω¯, we consider the boundary-value problem for the multidimensional parabolic equation∂u(t,s,x)∂t+∂u(t,s,x)∂s-∑r=1nαr(x)∂2u(t,s,x)∂xr2+δu(t,s,x)=f(t,s,x),x=(x1,…,xn)∈Ω,0<t,s<1,u(0,s,x)=ψ(s,x),s∈[0,1],u(t,0,x)=φ(t,x),t∈[0,1],x∈Ω,¯u(t,s,x)=0,t,s∈[0,1],x∈S,
where αr(x)>a>0(x∈Ω) and f(t,s,x)(t,s∈(0,1),x∈Ω) are given smooth functions and δ>0 is a sufficiently large number.
We introduce the Banach spaces C01β(Ω¯)(β=(β1,…,βn), 0<xk<1, k=1,…,n) of all continuous functions satisfying a Hölder condition with the indicator β=(β1,…,βn), βk∈(0,1), 1≤k≤n, and with weight xkβk(1-xk-hk)βk, 0≤xk<xk+hk≤1, 1≤k≤n, which is equipped with the norm‖f‖C01β(Ω¯)=‖f‖C(Ω¯)+sup0≤xk<xk+hk≤11≤k≤n|f(x1,…,xn)-f(x1+h1,…,xn+hn)|×∏k=1nhk-βkxkβk(1-xk-hk)βk,
where C(Ω¯) stands for the Banach space of all continuous functions defined on Ω¯, equipped with the norm‖f‖C(Ω¯)=maxx∈Ω¯|f(x)|.
It is known that the differential expressionAv=-∑r=1nαr(x)∂2v(t,s,x)∂x2+δv(t,s,x)
defines a positive operator A acting on C01β(Ω¯) with domain D(Ax)⊂C012+β(Ω¯) and satisfying the condition v=0 on S.
The discretization of problem (2.1) is carried out in two steps. In the first step, let us define the grid setsΩ̃h={x=xm=(h1m1,…,hnmn),m=(m1,…,mn),0≤mr≤Nr,hrNr=L,r=1,…,n},Ωh=Ω̃h∩Ω,Sh=Ω̃h∩S.
We introduce the Banach spaces Ch=Ch(Ω̃h), Chβ=C01β(Ω̃h) of grid functions φh(x)={φ(h1m1,…,hnmn)} defined on Ω̃h, equipped with the norms‖φh‖C(Ω¯h)=maxx∈Ω¯h|φh(x)|,‖φh‖C01β(Ω¯h)=‖φh‖C(Ωh¯)+sup0≤xk<xk+hk≤11≤k≤n|φh(x1,…,xn)-φh(x1+h1,…,xn+hn)|×∏k=1nhk-βkxkβk(1-xk-hk)βk.
To the differential operator A generated by problem (2.1), we assign the difference operator Ahx by the formulaAhxuxh=-∑r=1nar(x)(u-xr̅h)xr,jr
acting in the space of grid functions uh(x), satisfying the condition uh(x)=0 for all x∈Sh.
With the help of Ahx, we arrive at the initial boundary-value problem∂uh(t,s,x)∂t+∂uh(t,s,x)∂s+Ahxuh(t,s,x)=fh(t,s,x),0<t,s<1,x∈Ωh,uh(0,s,x)=ψh(s,x),0≤s≤1,uh(t,0,x)=φh(t,x),0≤t≤1,x∈Ω̃h
for an infinite system of ordinary differential equations.
In the second step, we replace problem (2.8) by difference scheme(1.2)uk,mh-uk-1,mhτ+uk-1,mh-uk-1,m-1hτ+Ahxuk,mh=fk,mh(x),x∈Ωh,fk,mh(x)=fh(tk,sm,x),tk=kτ,sm=mτ,1≤k,m≤N,x∈Ω̃h,u0,mh=ψmh,0≤m≤N,uk,0h=φkh,0≤k≤N
and by difference scheme(1.3)uk,mh-uk-1,mhτ+uk-1,mh-uk-1,m-1hτ+12Ahx(uk,m+uk-1,m-1)=fk,mh(x),x∈Ωh,fk,mh(x)=fh(tk-τ2,sm-τ2,x),tk=kτ,sm=mτ,1≤k,m≤N,x∈Ω̃h,u0,mh=ψmh,0≤m≤N,uk,0h=φkh,0≤k≤N.
It is known that Ahx is a positive operator in C(Ω̃h) and C01β(Ω¯h). Let us give a number of corollaries of Theorems 1.1 and 1.2.
Theorem 2.1.
For the solution of difference scheme (2.9), we have the following stability inequality:
max1≤k≤Nmax1≤m≤N‖uk,mh‖C(Ω̃h)≤C1(max0≤m≤N‖ψmh‖C(Ω̃h)+max0≤k≤N‖φkh‖C(Ω̃h)+max1≤k≤Nmax1≤m≤N‖fk,mh‖C(Ω̃h)),
where C1 is independent of τ, ψmh, φkh, and fk,mh.
Theorem 2.2.
For the solution of difference scheme (2.10), we have the following stability inequality:
max1≤k≤Nmax1≤m≤N‖uk,mh+uk-1,m-1h2‖C(Ω̃h)≤C1(max0≤m≤N‖ψmh‖C(Ω̃h)+max0≤k≤N‖φkh‖C(Ω̃h)+max1≤k≤Nmax1≤m≤N‖fk,mh‖C(Ω̃h)),
where C1 does not depend on τ, ψmh, φkh, and fk,mh.
3. Numerical Analysis
In this section, the initial boundary value problem∂u(s,t,x)∂t+∂u(s,t,x)∂s-∂2u(s,t,x)∂x2+2u(t,s,x)=f(t,s,x),f(t,s,x)=e-(t+s)sinπx,0<s,t<1,0<x<1,u(0,t,x)=e-tsinπx,0<t<1,0<x<1,u(s,0,x)=e-ssinπx,0<s<1,0<x<1,u(s,t,0)=u(s,t,π)=0,0<s,t<1
for one-dimensional ultraparabolic equations is considered.
The exact solution of problem (3.1) isu(t,s,x)=e-(t+s)sinπx.
Using the first order of accuracy in t and s implicit difference scheme (2.9), we obtain the difference scheme first order of accuracy in t and s and second-order of accuracy in xunk,m-unk-1,mτ-unk-1,m-unk-1,m-1τ-un+1k,m-2unk,m+un-1k,mh2+2unk,m=fk,mh,fk,mh=f(tk,sm,xn)=e-(tk+sm)sinxn,1≤k,m≤N,1≤n≤M-1,un0,m=e-smsinxn,0≤m≤N,0≤n≤M,unk,0=e-tksinxn,0≤k≤N,0≤n≤M,u0k,m=uMk,m=0,0≤k,m≤N,tk=kτ,sm=mτ,1≤k,m≤N,Nτ=1,xn=nh,1≤n≤M,Mh=π
for approximate solutions of initial boundary value problem (3.3). It can be written in the matrix formAun+1+Bun+Cun-1=φn,1≤n≤M-1,u0=0⃗,uM=0⃗.
HereA=[0⋯⋯⋯⋯0000⋯⋯⋯⋯0⋮⋱⋱⋮⋮⋮⋮0⋯0a0⋯000⋯0a0⋯⋮⋱⋱⋱⋱⋱⋮0⋯⋯⋯⋯⋯a](N+1)2×(N+1)2,B=[10⋯⋯⋯⋯000010⋯⋯⋯⋯000010⋯⋯⋯⋯0⋮⋮⋱⋱⋱⋱⋱⋱⋮0b0⋯cd0⋯000b0⋯cd0⋯⋮⋱⋱⋱⋱⋱⋱⋱⋮00⋯⋯b0⋯cd](N+1)2×(N+1)2,C=[0⋯⋯⋯⋯0000⋯⋯⋯⋯0⋮⋱⋱⋮⋮⋮⋮0⋯0a0⋯000⋯0a0⋯⋮⋱⋱⋱⋱⋱⋮0⋯⋯⋯⋯⋯a](N+1)2×(N+1)2,
where
a=-1h2,b=-1τ,c=-1τ,d=1τ+1τ+2h2+2,φn=[φn0,0φn1,0⋮φnN,0φn0,1φn1,1⋮φnN,1⋮φn0,Nφn1,N⋮φnN,N](N+1)2×1,φnk,m=f(tk,sm,xn)=e-(tk+sm)sinxn,un=[un0,0un1,0⋮unN,0un0,1un1,1⋮unN,1⋮un0,Nun1,N⋮unN,N](N+1)2×1.
This type system was used by Samarskii and Nikolaev [15] for difference equations. For the solution of matrix equation (3.4), we will use the modified Gauss elimination method. We seek a solution of the matrix equation by the following form:un=αn+1un+1+βn+1,n=M-1,…,2,1,
where uM=0⃗, αj(j=1,…,M-1) are (N+1)2×(N+1)2 square matrices, βj(j=1,…,M-1) are (N+1)2×1coloumn matrices, α1, β1 are zero matrices, andαn+1=-(B+Cαn)-1A,βn+1=(B+Cαn)-1(φn-Cβn),n=1,2,3,…,(M-1).
Using the second-order of accuracy in t and s implicit difference scheme (2.10), we obtain the difference scheme second-order of accuracy in t and s and second-order of accuracy in xunk,m-unk-1,mτ-unk-1,m-unk-1,m-1τ-12[un+1k,m-2unk,m+un-1k,mh2+2unk,m+un+1k-1,m-1-2unk-1,m-1+un-1k-1,m-1h2+2unk-1,m-1]=fk,mh,fk,mh=f(tk,sm,xn)=e-(tk+sm-τ)sinxn,1≤k,m≤N,1≤n≤M-1,un0,m=e-smsinxn,0≤m≤N,0≤n≤M,unk,0=e-tksinxn,0≤k≤N,0≤n≤M,u0k,m=uMk,m=0,0≤k,m≤N,tk=kτ,sm=mτ,1≤k,m≤N,Nτ=1,xn=nh,1≤n≤M,Mh=π
for approximate solutions of initial boundary value problem (3.9). The matrix form (3.4) can be written. Here,
A=[0⋯⋯⋯⋯0000⋯⋯⋯⋯0⋮⋱⋱⋮⋮⋮⋮a0⋯a0⋯00a0⋯a0⋯⋮⋱⋱⋱⋱⋱⋮0⋯⋯a0⋯a](N+1)2×(N+1)2(N+1)2×(N+1)2,B=[10⋯⋯⋯⋯000010⋯⋯⋯⋯000010⋯⋯⋯⋯0⋮⋮⋱⋱⋱⋱⋱⋱⋮b00⋯1c0⋯00b00⋯1c0⋯⋮⋱⋱⋱⋱⋱⋱⋱⋮00⋯b00⋯1c](N+1)2×(N+1)2,C=[0⋯⋯⋯⋯0000⋯⋯⋯⋯0⋮⋱⋱⋮⋮⋮⋮a0⋯a0⋯00a0⋯a0⋯⋮⋱⋱⋱⋱⋱⋮0⋯⋯a0⋯a](N+1)2×(N+1)2,
where
a=-12h2,b=-1τ+1h2+1,c=1τ+1h2+1,φn=[φn0,0φn1,0⋮φnN,0φn0,1φn1,1⋮φnN,1⋮φn0,Nφn1,N⋮φnN,N](N+1)2×1,φnk,m=f(tk-τ2,sm-τ2,xn)=e-(tk+sm)sinxn,un=[un0,0un1,0⋮unN,0un0,1un1,1⋮unN,1⋮un0,Nun1,N⋮unN,N](N+1)2×1.
We seek a solution of the matrix equation by the same algorithm (3.7) and (3.8).
4. Error Analysis
The errors are computed byENK,M=max1≤k,m≤N,1≤n≤M-1|u(tk,sm,xn)-unk,m|
of the numerical solutions, where u(tk,sm,xn) represents the exact solution and unk,m represents the numerical solution at (tk,sm,xn), and the results are given in Table 1.
Difference schemes
N=M=10
N=M=20
N=M=40
N=M=80
Difference scheme (3.3)
0.028100
0.014400
0.006811
0.003225
Difference scheme (3.9)
0.000511
0.000121
0.000028
0.000006
It may be noted from Table 1 that as N, M increase, the value of the errors associated with difference scheme (3.3) decreases by a factor of approximately 1/2 and the errors associated with difference scheme (3.9) decrease by a factor of approximately 1/4. This confirms that difference scheme (3.3) is first order and difference scheme (3.9) is second-order as stated in Section 1. Moreover, the results show that the second-order of accuracy difference scheme (3.9) are more accurate comparing with the first order of accuracy difference scheme (3.3).
DysonJ.SanchezE.Villella-BressanR.WebbG. F.An age and spatially structured model of tumor invasion with haptotaxis200781456010.3934/dcdsb.2007.8.452300321KunischK.SchappacherW.WebbG. F.Nonlinear age-dependent population dynamics with random diffusion1985111–3155173Hyperbolic partial differential equations, II10.1016/0898-1221(85)90144-0787434KolmogorovA. N.Zur Theorie der stetigen zufälligen Prozesse1933108149160KolmogorovA. N.Zufällige Bewegungen193435116117GenčevT. G.On ultraparabolic equations19631512652680151731DengQ.HallamT. G.An age structured population model in a spatially heterogeneous environment: existence and uniqueness theory200665237939410.1016/j.na.2005.06.0192228434di BlasioG.LambertiL.An initial-boundary value problem for age-dependent population diffusion197835359361510.1137/0135049507959di BlasioG.Nonlinear age-dependent population diffusion19798326528410.1007/BF00276312657285TersenovS. A.Boundary value problems for a class of ultraparabolic equations and their applications1987133(175)4529544911807AkrivisG.CrouzeixM.ThoméeV.Numerical methods for ultraparabolic equations1994313-417919010.1007/BF025758771415742AshyralyevA.YılmazS.GüneşM.ÇınarA. K.GürlerI.On the approximate solution of ultra parabolic equationsProceedings of the 2nd International Symposium on Computing in Science & Engineering2011Izmir, TurkeyGediz University533–535AshyralyevA.YılmazS.Second order of accuracy difference schemes for ultra parabolic equations2011138960160410.1063/1.3636802AshyralyevA.SobolevskiiP. E.199469Basel, SwitzerlandBirkh äuserSobolevskiiP. E.On the Crank-Nicolson difference scheme for parabolic equations197898106SamarskiiA. A.NikolaevE. S.1989Basel, SwitzerlandBirkhäuser1004469