Applied Numerical Mathematics 51 (2004) 207–219

www.elsevier.com/locate/apnum

A conservative difference scheme for the viscous Cahn–Hilliard

equation with a nonconstant gradient energy coefficient ✩
S.M. Choo a,∗ , S.K. Chung b , Y.J. Lee b
a Department of Mathematics, University of Ulsan, Ulsan 680-749, South Korea
b Department of Mathematics Education, Seoul National University, Seoul 151-742, South Korea

Available online 17 April 2004

Abstract
A finite difference scheme is considered for the viscous Cahn–Hilliard equation with the nonconstant gradient
energy coefficient. The scheme inherits energy dissipation property and mass conservation as for the classical
solution. We obtain the corresponding error estimate using the extended Lax–Richtmyer equivalence theorem.
 2004 IMACS. Published by Elsevier B.V. All rights reserved.

Keywords: Viscous Cahn–Hilliard equation; Dissipation property; Conservation of mass; Lax–Richtmyer equivalence
theorem; Nonlinear difference scheme

1. Introduction

Consider the viscous Cahn–Hilliard equation


∂ 2 ut ∂ 2 δG
ut − α 2 = 2 , (x, t) ∈ Ω × (0, T ], (1.1a)
∂x ∂x δu
∂u ∂ 3u
= 0, = 0, (x, t) ∈ ∂Ω × (0, T ], (1.1b)
∂x ∂x 3
u(x, 0) = u0 (x), x ∈ Ω, (1.1c)

✩
This work was supported by University of Ulsan Research Fund of 2002.
* Corresponding author.
E-mail addresses: [email protected] (S.M. Choo), [email protected] (S.K. Chung).

0168-9274/$30.00  2004 IMACS. Published by Elsevier B.V. All rights reserved.

doi:10.1016/j.apnum.2004.02.006
208 S.M. Choo et al. / Applied Numerical Mathematics 51 (2004) 207–219

where Ω = (0, L), α > 0 and δG

is the variational derivative of total energy
δu

2
1 2 1 4 1 ∂u
G(u) = − pu + ru + q(u) ,
2 4 2 ∂x
with positive constants p and r. That is,
 
2  
δG 1 ∂ ∂u ∂ ∂u
= −pu + ru + 3
q(u) − q(u) .
δu 2 ∂u ∂x ∂x ∂x
We assume there exist constants q0 and q1 such that 0 < q0
q(x)
q1 for all x.
When α = 0 and q(u) is constant, (1.1) becomes the usual Cahn–Hilliard equation. Since the
pioneering work of Cahn and Hilliard [3], the Cahn–Hilliard equation has been extensively studied by
Wang and Shi [26], Jabbari and Peppas [19], Puri and Binder [24] for the study of interfaces. Global
existence and uniqueness of the solution have been shown by Elliott and Zheng [12]. Yin [27] has shown
the existence of the continuous solution for the problem with degenerate mobility. Finite element Galerkin
solutions have been obtained by Elliott and French [8,9] and French and Jensen [13]. Elliott et al. [10]
have obtained optimal order error bounds using a second order splitting method. Mixed finite element
method is applied by Dean et al. [7]. A finite difference scheme has been studied by Furihata et al. [16]
who have examined the boundedness of the solution. The error estimates have been obtained by Choo
and Chung [4,5] and Choo et al. [6] for conservative nonlinear difference schemes. Recently, Furihata
[14] has proposed a difference scheme which inherits the decrease of the total energy.
Eq. (1.1) with α > 0 and a positive constant q(u) arises as a phenomenological continuum model for
phase separation in glass and polymer systems where intermolecular friction forces may be expected
to be of importance. See Novick-Cohen [22] for a derivation of the model and Novick-Cohen and
Pego [23] for more physical motivation. The viscous Cahn–Hilliard equation, which is viewed as a
singular limit of the phase field model of phase transition, has been studied by Bai et al. [2]. They have
studied the similarities and differences between the Cahn–Hilliard equation and Allen–Cahn equation by
using the viscous Cahn–Hilliard equation. Metastable pattern for the viscous Cahn–Hilliard equation has
been studied by Reyna and Ward [25]. Using explicit energy calculations, Grinfeld and Novick-Cohen
[18] have established a Morse decomposition of the stationary solutions of the viscous Cahn–Hilliard
equation. Existence theory of the solution of (1.1) has been shown in Elliott and Stuart [11].
The problem with nonconstant q(u) has been also considered by Cahn and Hilliard [3]. Novick-Cohen
[21] has analytically studied the stability of spatially homogeneous states. However, there is no numerical
study to our knowledge for nonconstant q(u).
In this paper, a difference scheme for (1.1) is considered, which has the property of mass conservation
and decreasing property of total energy as for the classical solution. In Section 2, following the idea of
Furihata [14] we propose a nonlinear difference scheme which inherits energy dissipation property and
mass conservation. In Section 3, we obtain the error estimates for the scheme using the idea of Lopez-
Marcos and Sanz-Serna [20].

2. Nonlinear finite difference scheme

Let h = M L
be the uniform step size in the spatial direction for a positive integer M and Ωh = {xi =
ih | i = 0, . . . , M}. Let k = T /N denote the uniform step size in the temporal direction for any positive
 S.M. Choo et al. / Applied Numerical Mathematics 51 (2004) 207–219 209

integer N . Denote Vin = V (xi , tn ) for xi = ih, i = 0, 1, . . . , M and tn = nk, n = 0, 1, . . . , N . For a

function V n = (V−2
n n
, V−1 , V0n , . . . , VMn , VM+1
n n
, VM+2 ) defined on Ωh , define the difference operators as
for 0
i
M,
n
Vi+1 − Vin V n − Vi−1
n
∇+ Vin = , ∇− Vin = i ,  in = 1 (∇+ + ∇− )Vin ,
∇V
h h 2
   
∇ 2 Vin = ∇+ ∇− Vin and ∇ 4 Vin = ∇ 2 ∇ 2 Vin .
Further, define operators V n+1/2 and ∂t V n , respectively, as

n+1/2 Vin+1 + Vin V n+1 − Vin

Vi = , ∂t Vin = i .
2 k
Then the approximate solution U n for (1.1a) is defined as a solution of


δGd
∂t Uin − α∇ 2 ∂t Uin = ∇ 2 , (2.1a)
δ(Uin+1 , Uin )
where

r  n+1 3−  n 
3
δGd n+1/2
= −pUi +U Ui
δ(Uin+1 , Uin ) 4 =0 i
 1
1 dq   2   2 n+1/2
+ n+1 n
∇+ Uin+ + ∇− Uin+ − q(Ui )n+1/2 ∇ 2 Ui
8 d(Ui , Ui ) =0
1
 n+1/2 1
 n+1/2
− ∇− q(Ui )n+1/2 ∇− Ui − ∇+ q(Ui )n+1/2 ∇+ Ui , (2.1b)
2 2
with
 q(U )−q(V )
dq i
Ui −Vi
i
, for Ui = Vi ,
=
d(Ui , Vi ) dq
dU
(Ui ), for Ui = Vi ,
initial conditions
Ui0 = u0 (xi ), 0
i
M (2.1c)
and boundary conditions
∇U  M
 0n = ∇U n  2 U0n = ∇∇
= ∇∇  2 UM
n
= 0, for 1
n
N. (2.1d)
We now introduce the discrete L2 inner product and the corresponding discrete L2 -norm, respectively,

M
 1 
M−1
1/2
(V , W )h = h Vi Wi = h (V0 W0 + VM WM ) + Vi Wi , V h = (V , V )h
i=0
2 i=1

for functions V = (V−2 , . . . , VM+2 ) and W = (W−2 , . . . , WM+2 ) defined on Ωh . And V ∞ =

max0
i
M |Vi |. Whenever there is no confusion, (·, ·) and  ·  will denote (·, ·)h and  · h , respectively.
It follows from summation by parts that Lemma 2.1 holds.
210 S.M. Choo et al. / Applied Numerical Mathematics 51 (2004) 207–219

Lemma 2.1. For functions V and W defined on Ωh and satisfying (2.1d), the following identities hold.
 2  
M
 
∇ V , W = −h (∇− Vi )(∇− Wi ) = V , ∇ 2 W .
i=1

Using Lemma 2.1, we obtain the following.

Lemma 2.2. For functions V defined on Ωh , the following inequalities hold.

(1) max{∇+ V 2 , ∇− V 2 }
2V ∇ 2 V .

(2) −(∇ 2 V , V )  14 {∇− V 2 + ∇+ V 2 }.

The following lemma can be verified by summation by parts and minimum eigenvalue of a symmetric
matrix (see [1]).

Lemma 2.3. Let M be any positive integer. If V0 = 0, then

 2 M M
π
2 sin V
2
(Vi − Vi−1 )2 .
2(2M + 1) i=1 i i=1

Lemma 2.4. For functions V defined on Ωh and satisfying (2.1d), the following inequalities hold:

(1) V 2∞
4V √

∇− V  + V 2 ;
(2) ∇− V ∞
2 14∇− V ∇ 2 V  + ∇− V 2 .
2

Proof. Let Vξ2 = mini {Vi2 }. Then


M
 
M

Vξ2 =h Vξ2
h Vi2 = V 2 .
i=0 i=0

And it follows that for some 0
s, 0



s+ 
s+
Vj2 − Vξ2 = δj ξ (Vi + Vi−1 )(Vi − Vi−1 ) = δj ξ h (Vi + Vi−1 )∇− Vi
4V ∇− V ,
i=s i=s

where if j  ξ , then δj ξ = 1 and if not, δj ξ = −1. This completes the proof of (1).
The proof of (2) follows from (1) and (2.1d). 2
 M
Let M(V ) = h i=0 Vi . Using Lemma 2.3, we obtain the following discrete Poincaré inequality
(see [4]).

Lemma 2.5. Let V be a discrete function defined on Ωh . Then, for 0 < h
1,

π2
V 2
2∇− V 2 + M(V )2 .
2
 S.M. Choo et al. / Applied Numerical Mathematics 51 (2004) 207–219 211

For the classical solution u of (1.1), we obtain

L L
 
L
d ∂ 2 δG ∂ δG
u(x, t) dx = + αut dx = + αut = 0,
dt ∂x 2 δu ∂x δu 0
0 0

and
L L L

2 
d   δG ∂u δG ∂u ∂u ∂u
G u(x, t) dx = dx = +α −α dx
dt δu ∂t δu ∂t ∂t ∂t
0 0 0
L

2 
∂ δG ∂u 2 ∂u
=− +α +α dx
0.
∂x δu ∂t ∂t
0

That is, the mass is conserved and total energy decreases. We can also show that the scheme (2.1) has
properties of conservation of mass and dissipation.

Theorem 2.1. Let U be the solution of (2.1). Then the conservation of mass holds. That is, for any n,

M
 
M

h Uin =h Ui0 .
i=0 i=0

Proof. Using (2.1) and the definition of ∇ 2 , we obtain, for any n,

 M
1  n+1  M

h Ui − h Uin
k i=0 i=0


M

 δGd
=h ∇ 2
+ α∂t Uin
i=0
δ(Uin+1 , Uin )



1 δGd δGd
= + α∂t U1n − + α∂t U0n
h δ(U1n+1 , U1n ) δ(U0n+1 , U0n )



δGd δGd
+ −∇− + α∂t U1 + ∇−
n
+ α∂t UMn
δ(U1n+1 , U1n ) δ(UM n+1
, UMn
)



1 δGd δGd
+ n+1 n
+ α∂t UM−1 −
n
n+1 n
+ α∂t UM
n
= 0.
h δ(UM−1 , UM−1 ) δ(UM , UM )
Thus we obtain the conservation of mass. 2

Remark 2.1. It follows from Theorem 2.1 and Lemma 2.5 that
 n 2   2
U 
2∇− U n 2 + π M(u0 )2 .
2
The boundedness of ∇− U  implies that of U n .
n
212 S.M. Choo et al. / Applied Numerical Mathematics 51 (2004) 207–219

Let Gd be the discretization of the energy function G(u) by

1 1 1 (∇+ Ui )2 + (∇− Ui )2
Gd (Ui ) = − p(Ui )2 + r(Ui )4 + q(Ui ) .
2 4 2 2
Following the idea of Furihata [14], we can also show that the discrete total energy decreases.

Theorem 2.2. Let Gd be the discretization of the energy function G(u). Then we have for all n

M
      M
h Gd Uin+1
h Gd Uin .
i=0 i=0

Proof. Since
 M M  
1   n+1   
M
  n  δGd
h Gd Ui −h Gd Ui =h ∂t Uin
n+1 n
,
k i=0 i=0 i=0
δ(U i , U i )
we obtain
 M
1   n+1  M
  n
h Gd Ui −h Gd Ui
k i=0 i=0
M 

2 
 δGd
=h ∂t Uin + α∂t Uin − α ∂t Uin

i=0
δ(Uin+1 , Uin )
M 

2 M

 2
δGd 
= −h ∇− + α∂t Uin − αh ∂t Uin
0.
i=1
δ(Uin+1 , Uin) i=0

This completes the proof. 2

Remark 2.2. It follows from Lemmas 2.4, 2.5 and Theorem 2.2 that we can see

 j  j  2 
max U n  , ∇− U n  , U n 
∞
j =1,2

 2    2  j/2
6   p

  3π 2

r u −
2
+ q1 ∇+ u0  + ∇− u0 
2 2
+ M(u0 ) 2
.
j =1
q0  0 r  2

3. Convergence of approximate solution

We recall extension of Lax–Richtmyer equivalence theorem which avoids the establishment of a priori
bounds for convergence in Lopez-Marcos and Sanz-Serna [20]. Let u be a solution of the problem and
uh be a discrete evaluation of u on Ωh . Let Uh be an approximate solution of u, which is obtained by
solving a discrete equation
Φh (Uh ) = 0, (3.1)
 S.M. Choo et al. / Applied Numerical Mathematics 51 (2004) 207–219 213

where Φh : Xh → Yh is a continuous mapping and Xh , Yh are normed spaces having the same dimension.
The finite difference scheme (3.1) is said to be convergent if (3.1) has a solution Uh such that
limh→0 Uh − uh Xh = 0. The discretization (3.1) is said to be consistent if limh→0 Φh (uh )Yh = 0.
The scheme (3.1) is said to be stable in the thresholds Rh if there exists a positive constant C such that
for an open ball B(uh , Rh ) ⊂ Xh
 
Vh − Wh X
C Φh (Vh ) − Φh (Wh ) , ∀Vh , Wh ∈ B(uh , Rh ).
h Yh
The following theorem is an extended Lax–Richtmyer equivalence theorem which gives existence and
convergence of approximate solutions [20].

Theorem 3.1. Assume that (3.1) is consistent and stable in the thresholds Rh . If Φh is continuous in
B(uh , Rh ) and Φh (uh )Yh = o(Rh ) as h → 0, then (3.1) has a unique solution Uh in B(uh , Rh ) and
there exists a constant C such that
 
Uh − uh X
C Φh (uh ) .
h Yh

For the error estimation, we rewrite (1.1) as

   
ut − αuxxt + q(u)uxx xx = f (u) xx , (3.2a)
where
δG
f (u) = + q(u)uxx , (3.2b)
δu
and the corresponding difference scheme (2.1a) as
 
2 q(Ui
n+1
) + q(Uin ) 2 n+1/2
∂t Ui − α∇ ∂t Ui + ∇
n 2 n
∇ Ui = ∇ 2 Fin+1 , (3.3a)
2
δGd q(Uin+1 ) + q(Uin ) 2 n+1/2
Fin+1 = + ∇ Ui . (3.3b)
δ(Uin+1 , Uin) 2
According to Theorem 3.1, we have only to show that (3.3) is consistent and stable in the thresholds.
Let Zh be the set of all functions defined on Ωh . We take Xh = Yh = ZhN+1 × ZhN with norms  · Xh and
 · Yh such that
 
      

N−1
  
(U, F)2 = max U n 2 + ∇− U n 2 + ∇+ U n 2 + k F n+1 2 + ∇ 2 U n+1/2 2 ,
Xh 0
n
N
n=0

 2  0 2     
N

 n 2  n 2 
 F Yh = 
U,  U  +  ∇− U
0 2 + ∇+ U
0 2 + k   + 
F U  .
n=1

Define a mapping Φh : Xh → Yh by Φh (U, F) = ( U, 

F), where for n = 0, . . . , N − 1
 
 2 q(Ui
n+1
) + q(Uin ) 2 n+1/2
Ui = ∂t Ui − α∇ ∂t Ui + ∇
n+1 n 2 n
∇ Ui − ∇ 2 Fin+1 , (3.4a)
2
in+1 = Fin+1 − δGd q(Ui ) + q(Uin ) 2 n+1/2
n+1
F − ∇ Ui , (3.4b)
δ(Uin+1 , Uin ) 2
0 = U 0 − u0 .
U (3.4c)
214 S.M. Choo et al. / Applied Numerical Mathematics 51 (2004) 207–219

Using Taylor’s Theorem and the Mean Value Theorem, we obtain the following consistency.

Theorem 3.2. Let u be the solution of (1.1) and (uh , fh ) be the discretized solution of (3.2). Then there
exists a constant C such that for all n
   
Φh (uh , fh )
C k + h2 .
Yh

We now obtain the stability of approximate solutions in the thresholds Rh as follows.

Theorem 3.3. Let Φh (U, F) = ( U, 

F) and Φh (V, G) = (  Assume dq is a C 1 (R2 )-function and
V, G). d(x,y)
Rh = O(k 2/3). Then there exists a constant C such that for any (U, F) and (V, G) in a ball B((uh , fh ), Rh )
   
(U − V, F − G)
C Φh (U, F) − Φh (V, G) .
Xh Yh

Proof. Let en = U n − V n , E n = F n − Gn , K1n = U

n − Vn , and K2n = F
n − G
n .
n n n n n n
Replacing U , F , U in (3.4a) by V , G , V , respectively, and then subtracting this result from
(3.4a), we obtain
 
2 q(U
n+1
) + q(U n ) 2 n+1/2
∂t e − α∇ ∂t e + ∇
n 2 n
∇ e
2
 
 2 q(U
n+1
) + q(U n ) 2 n+1/2
=∇ E
2 n+1
+ K1 − ∇
n+1
∇ V
2
 
2 q(V
n+1
) + q(V n ) 2 n+1/2
+∇ ∇ V . (3.5)
2
It follows from the definition of  · Xh that for (V, G) in the ball B((uh , fh ), Rh ),
 2 n+1/2 2      4/3  
∇ V 
2∇ 2 V n+1/2 − un+1/2 2 + 2∇ 2 un+1/2 2
C k + C
C k 1/3 + 1 , (3.6)
h h
k
 n 2    
V  + ∇− V n 2 + ∇+ V n 2
 2  2  2  2  

C k 2/3 + unh  + ∇− unh  + ∇+ unh 
C k 4/3 + 1 . (3.7)
Note that from Lemma 2.4 and (3.6),
  
 q(U n+1 ) + q(U n ) q(V n+1 ) + q(V n ) 2 n+1/2 2
 − ∇ V 
 2 2 
 n+1/2 2  2 n+1/2 2

C e  ∇ V
∞

 1/3  n+1/2  2 n+1/2   n+1/2 2 

C k + 1 e ∇ e  + e 

 2
 1/3 j  n+1/2 2 q0  2 n+1/2 2

C 1+ k e  + ∇ e . (3.8)
j =1
8

Taking an inner product between (3.5) and 2en+1/2 , and applying Lemmas 2.1, 2.2 and (3.8), we obtain
 S.M. Choo et al. / Applied Numerical Mathematics 51 (2004) 207–219 215

 2 α  2 α  2      2
∂t en  + ∂t ∇− en  + ∂t ∇+ en  +  q U n+1 + q U n ∇ 2 en+1/2 
4 4

2     2
   n+1/2 2  n+1 2

E n+1  + q0 ∇ 2 en+1/2  + C 1 + e  + K
1  .
2 2 j
k 1/3 (3.9)
q0 j =1

n in (3.4b) by V n , Gn , G
Replacing U n , F n , F n , respectively, and subtracting this result from (3.4b), and
dq
taking an inner product, we obtain from C (R2 )-function d(x,y)
1
, (3.7) and Lemma 2.4,


4
E n+1 , E n+1
q0
  1 
 n+1    
1
       n+1 

C K 2  + en+j  p + r U n+j  + V n+j  E 
2 2
∞ ∞
j =0 j =0


1
 n+j  
1
     
+ e  ∇+ U n+j 2 + ∇− U n+j 2 E n+1 
∞ ∞
j =0 j =0
  
 dq  1        

+  ∇+ en+j ∇+ U n+j  + ∇− en+j ∇− U n+j  E n+1 
d(V n+1 , V n )  ∞ j =0
∞ ∞

 1

1
       
+ ∇+ en+j  + ∇− en+j  ∇− U n+j  E n+1 
∞
j =0 j =0
 

    
1
     
+ ∇− q U n+1 + q U n  ∞
∇+ en+j  + ∇− en+j  E n+1 
j =0
 

4
  
1
 n+j 2  2  2 

C 1+ k 1/3
+k 4/3 j e  + ∇− e  + ∇+ e 
n+j n+j

j =1 j =0

1   n+1 2
E n+1 2 + C K
2  .
+ (3.10)
q0
It follows from (3.9) and (3.10) that

 2 α  2 α  2 1 2  2
∂t en  + ∂t ∇− en  + ∂t ∇+ en  + E n+1  + q0 ∇ 2 en+1/2 
4 4 q0
 
 4
 1/3  
1
 n+j 2  2  2 

C 1+ k +k 4/3 j e  + ∇− e  + ∇+ e 
n+j n+j

j =1 j =0

 n+1 2  n+1 2 
1  + K
+ C K 2  .

Summing from 0 to m and applying the discrete Gronwall inequality, we obtain

216 S.M. Choo et al. / Applied Numerical Mathematics 51 (2004) 207–219

 m+1 2     
m

 n+1 2  2 n+1/2 2 
e  + ∇− em+1 2 + ∇+ em+1 2 + k E  + ∇ e 
n=0

 0 2     
N

    

S e  + ∇− e0  + ∇+ e0  + k K1n  + K2n 
2 2 2 2

n=1

 kC{1+4j=1 (k 13 +k 43 )j } 
where S = exp  1 4 T . Since
1−kC{1+ 4j=1 (k 3 +k 3 )j }

0 − V
e0 = U 0 − V 0 = U 0 = K
10 ,
we obtain the desired result. 2

It follows from Theorem 3.2 that for k = O(h ) with 0 <  < 3

2 
Φ(uh , fh )Yh k + h2  2/3 
=O = O h + h 2−2/3
→ 0, as h → 0. (3.11)
Rh k 2/3
Thus applying Theorems 3.2, 3.3 and (3.11) to Theorem 3.1, we obtain the following error estimate
for (3.3).

Theorem 3.4. Suppose that the hypotheses of Theorem 3.3 hold. Let (U, F) be a solution of (3.3). Then
for k = O(h ) with 0 <  < 3,
   
(U − uh , F − fh )
C k 2 + h2 .
Xh

4. Numerical examples

We consider the following viscous Cahn–Hilliard equation


∂ 2 ut ∂ 2 δG
ut − = 2 , (x, t) ∈ Ω × (0, T ], (4.1a)
∂x 2 ∂x δu
∂u ∂ 3u
= 0, = 0, (x, t) ∈ ∂Ω × (0, T ], (4.1b)
∂x ∂x 3
u(x, 0) = u0 (x) = 0.1x 2 (1 − x)2 , (4.1c)
where the total energy is


2
1 2 3 4 1 1 ∂u
G(u) = − u + u + +1 .
2 4 2 u2 + 1 ∂x
Let U n be the approximate solution of


δGd
∂t Ui − ∇ ∂t Ui = ∇
n 2 n 2
, (4.2a)
δ(Uin+1 , Uin)
where initial conditions
Ui0 = u0 (xi ), (4.2b)
 S.M. Choo et al. / Applied Numerical Mathematics 51 (2004) 207–219 217

and boundary conditions

∇− U0n = 0 = ∇+ UM
n
, for 1
n
N. (4.2c)
δGd
Here δ(Uin+1 ,Uin )
is defined in (2.1b) with
 q(U )−q(V )
1 dq i
Ui −Vi
i
, for Ui = Vi ,
q(U ) = 2 + 1, =
U +1 d(Ui , Vi ) dq
(Ui ), for Ui = Vi .
dU

In order to find the computational solution of (4.2), we solve the system of difference equations (4.2) by
the generalized Newton’s method with a relaxation parameter ω = 0.7, h = 0.05 and k = 10−4 . We stop
the numerical computations when the relative errors satisfy the criterion

|Uin,(m+1) − Uin,(m) |

10−9 .
|Uin,(m) |
 20
In Fig. 1,  i=0 Uin for n = 0, . . . , 100 are given, which indicate the mass conservation of the solutions
of (4.2).
 M
Fig. 2 shows h  i=0 Gd (Uin ) for n = 0, . . . , 100 where the Gd is the discretization of the energy
function G(u) defined by

1 3 1 (∇+ Ui )2 + (∇− Ui )2
Gd (Ui ) = − (Ui )2 + (Ui )4 + q(Ui ) .
2 4 2 2
In this figure, we can see that the discrete total energy decreases.

 20 n
Fig. 1. i=0 Ui for n = 0, . . . , 100.
218 S.M. Choo et al. / Applied Numerical Mathematics 51 (2004) 207–219

 M n
Fig. 2. h i=0 Gd (Ui ) for n = 0, . . . , 100.

5. Concluding remarks

Applying the scheme proposed in [14] to the Cahn–Hilliard equation with constant q, Furihata [15]
showed existence and error estimates of the difference scheme. But if q is not constant, it becomes
difficult to show the existence and error analysis of approximate solutions for the difference scheme
because of nonlinear terms.
Adopting the idea of Furihata [14] we proposed a nonlinear difference scheme for the viscous Cahn–
Hilliard equations with nonconstant gradient energy coefficient q and showed the scheme preserves the
energy dissipation property and mass conservation as for the classical solution. Using the idea of Lopez-
Marcos and Sanz-Serna [20], we showed the existence of solution for the scheme and the corresponding
error estimate.
For work on conserving time-integration schemes for Hamiltonian systems, see Gonzalez [17].

References

[1] R.P. Agarwal, Difference Equations and Inequalities, Marcel Dekker, New York, 1992.
[2] F. Bai, C.M. Elliott, A. Gardiner, A. Spence, A.M. Stuart, The viscous Cahn–Hilliard equation. Part I: Computations,
Nonlinearity 8 (1995) 131–160.
[3] J.W. Cahn, J.E. Hilliard, Free energy of a nonuniform system I. Interfacial free energy, J. Chem. Phys. 28 (1958) 258–267.
[4] S.M. Choo, S.K. Chung, A nonlinear difference scheme for the viscous Cahn–Hilliard equation, J. Appl. Math. Comput.,
in press.
[5] S.M. Choo, S.K. Chung, Conservative nonlinear difference scheme for the Cahn–Hilliard equation, Comput. Math.
Appl. 36 (1998) 31–39.
[6] S.M. Choo, S.K. Chung, K.I. Kim, Conservative nonlinear difference scheme for the Cahn–Hilliard equation: II, Comput.
Math. Appl. 39 (2000) 229–243.
 S.M. Choo et al. / Applied Numerical Mathematics 51 (2004) 207–219 219

[7] E.J. Dean, R. Glowinski, D.A. Trevas, An approximate factorization/least squares solution method for a mixed finite
element approximation of the Cahn–Hilliard equation, Japan J. Indust. Appl. Math. 13 (1996) 495–517.
[8] C.M. Elliott, D.A. French, Numerical studies of the Cahn–Hilliard equation for phase separation, IMA J. Appl. Math. 38
(1987) 97–128.
[9] C.M. Elliott, D.A. French, A nonconforming finite-element method for the two-dimensional Cahn–Hilliard equation,
SIAM J. Numer. Anal. 26 (1989) 884–903.
[10] C.M. Elliott, D.A. French, F.A. Milner, A second order splitting method for the Cahn–Hilliard equation, Numer. Math. 54
(1989) 575–590.
[11] C.M. Elliott, A.M. Stuart, Viscous Cahn–Hilliard equation II. Analysis, J. Differential Equations 128 (1996) 387–414.
[12] C.M. Elliott, S. Zheng, On the Cahn–Hilliard equation, Arch. Rat. Mech. Anal. 96 (1986) 339–357.
[13] D.A. French, S. Jensen, Long-time behaviour of arbitrary order continuous time Galerkin schemes for some one-
dimensional phase transition problems, IMA J. Numer. Anal. 14 (1994) 421–442.
[14] D. Furihata, Finite difference schemes for ∂u ∂ α δG
∂t = ( ∂x ) δu that inherit energy conservation or dissipation property,
J. Comput. Phys. 156 (1999) 181–205.
[15] D. Furihata, A stable and conservative finite difference scheme for the Cahn–Hilliard equation, Numer. Math. 87 (2001)
675–699.
[16] D. Furihata, T. Onda, M. Mori, A finite difference scheme for the Cahn–Hilliard equation based on a Lyapunov functional,
GAKUTO Internat. Ser. Math. Sci. Appl. 2 (1993) 347–358.
[17] O. Gonzalez, Time integration and discrete Hamiltonian systems, J. Nonlinear Sci. 6 (1996) 449–467.
[18] M. Grinfeld, A. Novick-Cohen, The viscous Cahn–Hilliard equation: Morse decomposition and structure of the global
attractor, Trans. Amer. Math. Sci. 351 (1999) 2375–2406.
[19] E. Jabbari, N.A. Peppas, A model for interdiffusion at interfaces of polymers with dissimilar physical properties,
Polymer 36 (1995) 575–586.
[20] J.C. Lopez-Marcos, J.M. Sanz-Serna, Stability and convergence in numerical analysis III: Linear investigation of nonlinear
stability, IMA J. Numer. Anal. 8 (1988) 71–84.
[21] A. Novick-Cohen, Energy method for the Cahn–Hilliard equation, Quart. Appl. Math. 46 (1988) 681–690.
[22] A. Novick-Cohen, On the viscous Cahn–Hilliard equation, in: J.M. Ball (Ed.), Material Instabilities in Continuum and
Related Mathematical Problems, Oxford Univ. Press, Oxford, 1988, pp. 329–342.
[23] A. Novick-Cohen, R.L. Pego, Stable patterns in a viscous diffusion equation, Trans. Amer. Math. Soc. 324 (1991) 331–351.
[24] S. Puri, K. Binder, Phenomenological theory for the formation of interfaces via the interdiffusion of layers, Phys. Rev.
B 44 (1991) 9735–9738.
[25] L.G. Reyna, M.J. Ward, Metastable internal layer dynamics for the viscous Cahn–Hilliard equation, Methods Appl. Anal. 2
(1995) 285–306.
[26] S.Q. Wang, Q. Shi, Interdiffusion in binary polymer mixtures, Macromolecules 26 (1993) 1091–1096.
[27] J. Yin, On the existence of nonnegative continuous solutions of the Cahn–Hilliard equation, J. Differential Equations 97
(1992) 310–327.