Applied Mathematical Sciences, Vol. 9, 2015, no.

120, 5987 - 5993

HIKARI Ltd, www.m-hikari.com
http://dx.doi.org/10.12988/ams.2015.48674

Block-Centered Difference Method

for RLW Equation
Juxia Jiang

School of Mathematics and Information Science

Xinxiang University, Henan Province, 453003, China

Weiyin Ren

College of Computer and Information Engineering

Xinxiang University, Henan Province, 453003, China

Copyright c 2014 Juxia Jiang and Weiyin Ren. This article is distributed under the
Creative Commons Attribution License, which permits unrestricted use, distribution, and
reproduction in any medium, provided the original work is properly cited.

Abstract

In this paper, a block-centered difference method is presented for RLW equation

in the bounded domain. The theoretical analysis shows that the scheme is conver-
gent and we get the discrete L2 -norm errors in both the approximate solution and
its first derivatives for all nonuniform grids.

Keywords: RLW equation; block-centered difference method; error esti-

mates

1. Introduction
Regularized long wave equation which is also referred to as RLW equation
occupies an important position in the research of soliton theory and the phys-
ical problems of nonlinear wave. There is some research about the regularized
long wave equation, such as the finite element and finite difference - stream-
line method. However, these methods can only obtain approximate solution
of equation. While the block-centered difference method it can get both the
approximate solution and approximate solution of the first derivative of the
5988 Juxia Jiang and Weiyin Ren

equation. This paper adopts the block-centered difference method to construct

the format which can help us obtain the error estimation O(τ + h2 ) and the
solution of the equation has good convergence and steadiness.

Consider the following initial-boundary value problem of RLW equation:

ut + α(x, t)ux − δuxxt = f (x, t), (x, t) ∈ Ω × (0, T ], (1)

u(x, 0) = u0 (x), x ∈ Ω, (2)
u(0, t) = u(1, t) = 0, t ∈ (0, T ]. (3)
where δ is a positive constant, Ω = (0, 1),and α, f, u0 are functions have been
given,J = [0, T ],for any (x, t) ∈ Ω × J,the following conditions is established:
(i)there exist α0 , α1 ,s.t.0 < α0 ≤ α(x, t) ≤ α1 .
(ii)f, u are properly smooth.
Let: p = − ∂u∂x
.
So that (1) is equivalent to:
∂u ∂ 2p
− α(x, t)p + δ = f (x, t), (x, t) ∈ Ω × (0, T ]. (4)
∂t ∂x∂t
2. Preliminaries
We define the partition Ω = (0, 1) by δx ,where:
δx : 0 = x 1 < x 3 < . . . < xN − 1 < xN + 1 = 1.
2 2 2 2

For each i = 1 to N , define:

xi+ 1 + xi− 1
xi = 2 2
, hi = xi+ 1 − xi− 1 , h = max hi ,
2 2 2 i

hi + hi+1
hi+ 1 = xi+1 − xi = , Ωi = (xi− 1 , xi+ 1 ).
2 2 2 2

n T
Note τ as time step, t = n · τ, n = 0, 1, . . . , L; L = [ τ ];

gin = g(xi , tn ), dt g n = τ1 (g n − g n−1 ), gi+ 1 = g(xi+ 1 ),

2 2
gi+ 1 −gi− 1
gi+1 −gi
[dx g]i+ 1 = hi+ 1
, [Dx g]i = 2
hi
2
.
2
2

Define the following two grid function groups

S = {fi |fi = f (xi ), i = 1, . . . , N },
S (1) = {fi− 1 |f 1 = fN + 1 = 0, i = 1, . . . , N + 1}.
2 2 2

and the discrete inner product and norm

(f, g)M = N h f g , kf k2M = (f, f )M ,
P
PNi=1 i i i
(f, g)x = i=2 hi− 1 fi− 1 gi− 1 , kf k2x = (f, f )x .
2 2 2
 Block-centered difference method for RLW equation 5989

Definition 2.1. Let f ∈ C[0, 1],define

N
X hi
k f k2lx2 = hi {max | f (x) |2 :| x − xi |< }.
i=1
2
[3]
Lemma 2.1. Let (f, g) ∈ S × S (1) ,then
(f, Dx g)M = −(dx f, g)x .
Lemma 2.2. [5] Suppose the following conditions hold
(1) u(0, t) = u(1, t) = 0, t ∈ (0, T ],
∂us+2
(2) ∂x 2 ∂ts (s = 0, 1) is Lipschitz continuous in Ω,
s s n
then there exist ∂∂tQs ∈ S; ∂∂tVs ∈ S (1) ,s.t.
∂s (s) 2 ∂s
| (u i − Qi )| ≤ C 0 h , | (p 1 − Vi− 1 )| ≤ Ch2 ,
∂ts ∂ts i− 2 2

Vi− 1 = −(dx Q)i− 1 .

2 2

[5]
Lemma 2.3. Let f ∈ S,then
kf kM ≤ kdx f kx .

3. Block Central Difference Scheme and Error Estimates

The block central difference scheme of problem (1) − (3) is : find (U n , P n ) ∈
S × S (1) s.t.
dt Uin − 12 ((αP )ni+ 1 + (αP )ni− 1 ) + dt [Dx (δP )]ni = fin , (5)
2 2
n n
Pi+ 1 = −[dx U ] , (6)
2
i+ 1 2
Ui0 = u0 (xi ), i = 1, . . . , N, (7)
U0n = 0, n = 0, 1, . . . , L. (8)
Substituting (6) into (5), we get the linear algebraic equations about Uin . It’s
clear that its coefficient matrix is irreducible diagonally dominant tridiagonal
matrix, so the solution vector exists and it is unique.
At the point (xi , tn ), (4) can be written as:
1
dt uni − ((αp)ni+ 1 + (αp)ni− 1 ) + dt [Dx (δp)]ni = fin + Rin + Ani + Bin , (9)
2 2 2

where:
Rin = dt uni − ( ∂u )n ,
∂t i
Ani = − 21 ((αp)ni+ 1 + (αp)ni− 1 ) + (αp)ni ,
2 2
∂ p n 2
Bin = dt [Dx (δp)]ni − δ( ∂x∂t )i .
From Lemma 2.2 we know that there exist Q ∈ S, V ∈ S (1) , and Vi− 1 =
2
−[dx Q]i− 1 .
2
5990 Juxia Jiang and Weiyin Ren

Set:
ξ = P − V, β = V − p, γ = Q − u, η = U − Q.
so:
ξi+ 1 = Pi+ 1 − Vi+ 1 = −[dx U ]i+ 1 + [dx Q]i+ 1 = −[dx η]i+ 1 .
2 2 2 2 2 2

Subtracting (9) from (5), we obtain:

dt ηin − 12 ((αξ)ni+ 1 + (αξ)ni− 1 ) + dt [Dx (δξ)]ni

2 2
= −Rin − Ani − Bin − dt γin − dt [Dx (δβ)]ni + 21 ((αβ)ni+ 1 + (αβ)ni− 1 ). (10)
2 2

Make inner product with hi ηin on both sides of (10), and summing on i from
1 to N , we obtain
(dt η n , η n )M − 21 N n n n n n
P
i=1 hi ((αξ)i+ 1 + (αξ)i− 1 )ηi + (dt [Dx (δξ)] , η )M
2 2
= −(Rn , η n )M − (An , η n )M − (B n , η n )M − (dt γ n , η n )M − (dt [Dx (δβ)]n , η n )M
+ 21 N n n n
P
i=1 hi ((αβ)i+ 1 + (αβ)i− 1 )ηi . (11)
2 2

Now we will estimate one by one.

By Schwarz inequality
(dt η n , η n )M = τ1 (kη n k2M − (η n , η n−1 )M ) ≥ τ1 (kη n k2M − 12 (kη n k2M + kη n−1 k2M ))
= 2τ1 (kη n k2M − kη n−1 k2M ) (12)

2 12 n 2 12
PN
hi (αξ)ni+ 1 ηin ≤ ( N
P n
PN
i=1 i=1 hi ((αξ)i+ 1 ) ) ( i=1 hi (ηi ) )
2 2
2 12
= ( 21 N n
)2 + 12 N n n
P P
h
i=2 i− 1 ((αξ)
i− 1 2 i=2 hi− 1 ((αξ)i− 1 ) ) kη kM
2
2 2
≤ kαn ξ n kx kη n kM ≤ 21 (α1 kξ n k2x + kη n k2M )
Similarly
N
X 1
hi (αξ)ni− 1 ηin ≤ (α1 kξ n k2x + kη n k2M )
i=1
2 2

so
N
1X 1
− hi ((αξ)ni+ 1 + (αξ)ni− 1 )ηin ≥ − (α1 kξ n k2x + kη n k2M ) (13)
2 i=1 2 2 2

By Lemma 2.1, we get:

n n−1 )
(dt [Dx (δξ)]n , η n )M = ( δ(ξ −ξ τ
, ξ n )x
= τδ (kξ n k2x − (ξ n−1 , ξ n )x ) ≥ δ
2τ
(kξ n k2x − kξ n−1 k2x ) (14)
 Block-centered difference method for RLW equation 5991

R tn 2
kRn k2M = k τ1 tn−1 (s − tn−1 ) ∂∂su2 dsk2M
tn 1 tn 2 1
≤ k τ1 ( tn−1 (s − tn−1 )2 ds) 2 ( tn−1 ( ∂∂su2 )2 ds) 2 k2M
R R
R tn 2
≤ Cτ tn−1 k( ∂∂su2 )2 kM ds ≤ Cτ kutt k2L2 (tn−1 ,tn ;M )
By ε− inequality
|(−Rn , η n )M | ≤ Cτ kutt k2L2 (tn−1 ,tn ;M ) + εkη n k2M (15)
By T aylor theorem we have
R hi 2 (αp)n
Ani = 2hi ( h2i − |z|) ∂ ∂z 2 (xi + z, tn )dz
−2
2 (αp)n R h2i hi
≤ (max{ ∂ ∂x 2 : |x − x i | ≤ hi
2
}) 2
hi ( 2 − |z|) dz
−2
2 (αp)n
≤ Ch 2
(max{ ∂ ∂x 2 : |x − xi | ≤ hi
2
})
so
∂ 2 (αp)n 2
kAn k2M ≤ Ch4 k klx2
∂x2
2 (αp)n
|(−An , η n )M | ≤ Ch4 k ∂ ∂x2
k2lx2 + εkη n k2M (16)
n) 2 (δp) R tn 2
Bin = Dx [ ∂(δp
∂t
]i − ( ∂∂x∂t )ni − τ1 tn−1 (s − tn−1 )Dx ( ∂ ∂s(δp) 2 )ds
hi n
4 1 t ∂ 2 (δp)) 2 1
≤ hCi 2hi ( h2i − |z|)2 ∂∂z(δp)
R n
R
3 ∂t (xi + z, t )dz − τ ( tn−1 (Dx ( ∂s2 ) ds)
2 2
−2
4 n) R h2i hi 1 R tn ∂ 2 (δp) 2 1
≤ hCi (max{ ∂∂x(δp3 ∂t : |x − xi | ≤ 2 })
hi 2
hi ( 2 − |z|) dz + τ ( tn−1 (Dx ( ∂s2 )) ds)
2 2
−2
4 n) R tn R xi+ 1 ∂ 3 (δp) 1
≤ Ch2 (max{ ∂∂x(δp hi
3 ∂t : |x − xi | ≤ 2 }) + ( h
τ
i
( 2
tn−1 x 1 ∂x∂s2
dx)2 ds) 2
i− 2

then
∂ 4 (δpn ) 2 ∂ 3 (δp) 2
kB n k2M ≤ Ch4 k k 2 + τ k k 2 n−1 n 2 .
∂x3 ∂t lx ∂x∂t2 L (t ,t ;lx )
4 n 3
|(−B n , η n )M | ≤ Ch4 k ∂∂x(δp ) 2 ∂ (δp) 2 n 2
3 ∂t kl2 + τ k ∂x∂t2 kL2 (tn−1 ,tn ;l2 ) + εkη kM .
x x
(17)
R tn
|dt γin | = | τ1 (γin − γin−1 )| = | τ1 tn−1 ∂γ
∂s
i
ds|.
C
|(−dt γin , η n )M | ≤ kγt k2L2 (tn−1 ,tn ;M ) +εkη n k2M (18)
τ

By (13) and Lemma 2.2,we get

1
PN
2 i=1 hi ((αβ)ni+ 1 + (αβ)ni− 1 )ηin ≤ 21 (α1 kβ n k2x + kη n k2M )
2 2
≤ 12 (Ch4 + kη n k2M ). (19)
By Lemma 2.2, we have:
(dt [Dx (δβ)]n , η n )M = |( τδ (β n − β n−1 ), ξ n )x |
∗)
≤ Ck ∂β(t ∂t
k2x + εkξ n k2x ≤ Ch4 + εkξ n k2x . (20)
5992 Juxia Jiang and Weiyin Ren

where tn−1 < t∗ < tn .

Substituting (12) − (20) into (11), we choose properly small positive number
ε, so that
1
τ
(kη n k2M − kη n−1 k2M ) + τδ (kξ n k2x − kξ n−1 k2x )
2 (αp)n 4 n) ∂ 3 (δp) 2
≤ C(τ kutt k2L2 (tn−1 ,tn ;M ) + h4 k ∂ ∂x 2 k2lx2 + h4 k ∂∂x(δp 2
3 ∂t kl2 + τ k ∂x∂t2 kL2 (tn−1 ,tn ;l2 )
x x
+ τ1 kγt k2L2 (tn−1 ,tn ;M ) + h4 + kη n k2M + kξ n k2x ). (21)

Multiply both sides of (21) by τ and summing on n = 1, . . . , L, we obtain:

kη L k2M − kη 0 k2M + kξ L k2x − kξ 0 k2x
3 PL
≤ C(τ 2 kutt k2L2 (0,T ;M ) + τ 2 k ∂∂x∂t
(δp) 2 4 2
2 kL2 (0,T ;l2 ) + h + kγt kL2 (0,T ;M ) + τ
n 2 n 2
n=1 (kη kM + kξ kx )).
x

By Lemma 2.2,
kγt kL2 (0,T ;M ) ≤ Ch2 .

Hence,
L
X
kη L k2M + kξ L k2x ≤ kη 0 k2M + kξ 0 k2x + C(τ 2 + h4 + τ (kη n k2M + kξ n k2x )) (22)
n=1

By discrete Gronwall inequality, we obtain

kη L k2M + kξ L k2x ≤ C(τ 2 + h4 ) + C(kη 0 k2M + kξ 0 k2x ). (23)
By the initial value, we get:
0 0 0
Pi+ 1 = −[dx U ]i+ 1 , V
i+ 1
= −[dx Q0 ]i+ 1 .
2 2 2 2

kη 0 kM = 0, kξ 0 kx = 0.

so
kη L k2M + kξ L k2x ≤ C(τ 2 + h4 ).

By triangle inequality and Lemma 2.2, we obtain

Theorem 3.1. Let u is the solution of (1) − (3), U n and P n are the solutions
of (5) − (8) respectively. Assume that the grid in the solving area is splited
regularly, then there exists a constant C independent h, τ ,s.t. for any 0 ≤ n ≤
L(L ≤ [ Tτ ]) we have:

kP n − pn kx ≤ kξ n kx + kβ n kx ≤ C(τ + h2 ),

kU n − un kM ≤ kη n kM + kγ n kM ≤ C(τ + h2 ).
 Block-centered difference method for RLW equation 5993

References
[1] J.C. Eilbeck, G.R. Meguire, Numerical study of the regularized long wave queation I:
Numerical Methods, Journal of Computational Physics, 19 (1975), 43-57.
http://dx.doi.org/10.1016/0021-9991(75)90115-1
[2] J. Zhu, Nonlinear RLW equation characteristics of the numerical method, Journal of
Applied Mathematics, 1 (1990), 64-73.
[3] A. Weiser and M.F. Wheeler, On convergence of block-centered finite differences for
elliptic problems, SIAM J. Numer. Anal., 25 (1988), 351-375.
http://dx.doi.org/10.1137/0725025
[4] Z.X. Ren and Y. Hao, W.Y. Ren, Economical difference streamline diffusion method
on nonlinear RLW equation, Journal of Engineering Mathematics, 25 (2008), 708-712.
[5] S.L. Wang and S.Y. Sun, Numerical methods for convection-diffusion problems based on
combining the method of chararteristic with block-centered finite difference procedure,
Math. Nume, 4 (1999), 463-474.

Received: August 15, 2014; Published: September 28, 2015