You are on page 1of 14

# Neural, Parallel and Scientiﬁc Computations 17 (2009) 17–30

Numerical Methods for Systems of Nonlinear
Differential Functional Equations
Anna Szafra´ nska
Gda´ nsk University of Technology,
Department of Applied Physics and Mathematics,
ul. Gabriela Narutowicza 11-12, 80-952 Gda´ nsk, Poland
Abstract: The paper deals with initial boundary value problems for nonlinear differential
functional systems. We are interested in approximation of solutions of considered differen-
tial problems by solutions of suitable difference schemes. A complete convergence analysis
for the methods is presented. The proof of the stability is based on a comparison technique
with nonlinear estimates of the Perron type for given operators.
Key Words Initial boundary value problems, difference functional equations, difference
methods, stability and convergence, interpolating operators, error estimates.
1 INTRODUCTION
For any metric spaces X and Y we denote by C(X, Y ) the class of all continuous func-
tions from X to Y . We denote by M
k×n
the space of real k × n matrices. We will use
vectorial inequalities with the understanding that the same inequalities hold between their
corresponding components.
For each x = (x
1
, . . . , x
n
) ∈ R
n
we write x = (x

, x
′′
) where x

= (x
1
, . . . , x
κ
), x
′′
=
(x
κ+1
, . . . , x
n
), where 0 ≤ κ ≤ n is ﬁxed. If κ = n we have x

= x, if κ = 0 then x
′′
= x.
We deﬁne the sets
E = [0, a] ×[−b

, b

) ×(−b
′′
, b
′′
], D = [−d
0
, 0] ×[0, d

] ×[−d
′′
, 0]
where a > 0, d
0
∈ R
+
, b = (b
1
, . . . , b
n
) ∈ R
n
+
and d = (d
1
, . . . , d
n
) ∈ R
n
+
are given. Let
c = (c
1
, . . . , c
n
) = b +d and
E
0
= [−d
0
, 0] ×[−b

, c

] ×[−c
′′
, b
′′
],

0
E = ((0, a] ×[−b

, c

] ×[−c
′′
, b
′′
]) \ E, E

= E
0
∪ E ∪ ∂
0
E.
For z : E

→ R
k
, z = (z
1
, . . . , z
k
), and (t, x) ∈ [0, a] × [−b, b] we deﬁne the function
z
(t,x)
: D → R
k
as follows
z
(t,x)
(τ, y) = z(t +τ, x +y), (τ, y) ∈ D.
The function z
(t,x)
is the restriction of z to the set [t − d
0
, t] × [x

, x

+ d

] × [x
′′
− d
′′
, x
′′
]
and this restriction is shifted to the set D. Put Ω = E ×C(D, R
k
) ×R
n
and suppose that
f : Ω → R
k
, f = (f
1
, . . . , f
k
),
Received March 10, 2009 1061-5369 c Dynamic Publishers, Inc.
18 Szafranska
ϕ : E
0
∪ ∂
0
E → R
k
, ϕ = (ϕ
1
, . . . , ϕ
k
)
are given functions. We consider the system of differential functional equations

t
z
i
(t, x) = f
i
(t, x, z
(t,x)
, ∂
x
z
i
(t, x)), 1 ≤ i ≤ k, (1.1)
with the initial boundary condition
z(t, x) = ϕ(t, x) on E
0
∪ ∂
0
E, (1.2)
where ∂
x
z
i
(t, x) = (∂
x
1
z
i
(t, x), . . . , ∂
xn
z
i
(t, x)), 1 ≤ i ≤ k. A function v : E

→ R
k
is
called a classical solution of problem (1.1), (1.2) if
(i) v ∈ C(E

, R
k
) and v is of class C
1
on E,
(ii) v = (v
1
, . . . , v
k
) satisﬁes system of equations (1.1) on E and condition (1.2) holds.
Systems of differential equations with deviated variables and differential integral prob-
lems can be derived from (1.1) by specializing the operator f.
2 DIFFERENCE FUNCTIONAL PROBLEMS
We are interested in the construction of a method for the approximation of classical solu-
tions to problem (1.1), (1.2) with solutions of associated difference scheme and in the error
estimate for the constructed method.
Let us denote by F(X, Y ) the class of all functions deﬁned on X and taking values in Y ,
where X and Y are arbitrary sets. Let Nand Z be the sets of natural numbers and integers,
respectively. For x = (x
1
, . . . , x
n
) ∈ R
n
, p = (p
1
, . . . , p
k
) ∈ R
k
and for the matrix
U ∈ M
k×n
, U = [u
ij
]
i=1,...,k,j=1,...,n
we write
x =
n

i=1
|x
i
| and p

= max {|p
i
| : 1 ≤ i ≤ k},
U = max
_
n

j=1
|u
ij
| : 1 ≤ i ≤ k
_
.
For a function w ∈ C(D, R
k
) we put
w
D
= max {w(t, x)

: (t, x) ∈ D}.
We deﬁne a mesh on the set E

and D in the following way. Let (h
0
, h

), h

= (h
1
, . . . , h
n
),
stand for steps of the mesh. For h = (h
0
, h

) and (r, m) ∈ Z
1+r
, where m = (m
1
, . . . , m
n
),
we deﬁne nodal points as follows
t
(r)
= rh
0
, x
(m)
= (x
(m
1
)
1
, . . . , x
(mn)
n
) = (m
1
h
1
, . . . , m
n
h
n
).
Numerical Methods for Systems of Nonlinear Differential Functional Equations 19
Let us denote by H the set of all h = (h
0
, h

) such that there are K
0
∈ Z and N =
(N
1
, . . . , N
n
) ∈ N
n
with the properties K
0
h
0
= d
0
and (N
1
h
1
, . . . , N
n
h
n
) = d. Let
K ∈ Nbe deﬁned by the relations Kh
0
≤ a < (K + 1)h
0
. Write
R
1+n
h
= {(t
(r)
, x
(m)
) : (r, m) ∈ Z
1+n
}
and
E
h
= E ∩ R
1+n
h
, E
h.0
= E
0
∩ R
1+n
h
, D
h
= D ∩ R
1+n
h
,

0
E
h
= ∂
0
E ∩ R
1+n
h
, E

h
= E
h.0
∪ E
h
∪ ∂
0
E
h
.
For functions z : E

h
→ R
k
and w : D
h
→ R
k
we write
z
(r,m)
= z(t
(r)
, x
(m)
) on E

h
and w
(r,m)
= w(t
(r)
, x
(m)
) on D
h
.
For the above z and for a point (t
(r)
, x
(m)
) ∈ E
h
we deﬁne the function z
[r,m]
: D
h
→ R
k
by the formula
z
[r,m]
(τ, y) = z(t
(r)
+τ, x
(m)
+y), (τ, y) ∈ D
h
.
The function z
[r,m]
is the restriction of z to the set
([t
(r)
−d
0
] ×[x
(m

)
, x
(m

)
+d

] ×[x
(m
′′
)
−d
′′
, x
(m
′′
)
]) ∩ R
1+n
h
and the restriction is shifted to the set D
h
. For a function w : D
h
→ R
k
we write
w
D
h
= max {w
(r,m)

: (t
(r)
, x
(m)
) ∈ D
h
}.
Let e
j
= (0, . . . , 0, 1, 0, . . . , 0) ∈ R
n
, 1 ≤ j ≤ n, where 1 is the j-th coordinate. We
consider difference operators δ
0
and δ = (δ
1
, . . . , δ
n
) deﬁned in the following way
δ
0
z
(r,m)
=
1
h
0
_
z
(r+1,m)
−z
(r,m)
_
(2.1)
and
δ
j
z
(r,m)
=
1
h
j
_
z
(r,m+e
j
)
−z
(r,m)
_
, 1 ≤ j ≤ κ, (2.2)
δ
j
z
(r,m)
=
1
h
j
_
z
(r,m)
−z
(r,m−e
j
)
_
, κ + 1 ≤ j ≤ n. (2.3)
Note that δz
(r,m)
is given by (2.3) if κ = 0 and δz
(r,m)
is deﬁned by (2.2) for κ = n.
Right-hand sides of equations (1.1) contain the functional variable z
(t,x)
which is the
element of the space C(D, R
k
). Therefore we need an interpolating operator T
h
:
F(D
h
, R
k
) → F(D, R
k
) and the following assumptions on the operator T
h
.
Assumption H[T
h
]. Suppose that the operator T
h
: F(D
h
, R
k
) → F(D, R
k
) satisﬁes the
conditions
1) if w, ¯ w ∈ F(D
h
, R
k
) then T
h
[w], T
h
[ ¯ w] ∈ C(D, R
k
) and
T
h
[w] −T
h
[ ¯ w]
D
≤ w − ¯ w
D
h
, (2.4)
20 Szafranska
2) if w : D → R
k
is of class C
1
then there is γ : H → R
+
such that
T
h
[w
h
] −w
D
≤ γ(h) and lim
h→0
γ(h) = 0, (2.5)
where w
h
is the restriction of w to the set D
h
.
Remark 2.1 The condition 1) of Assumption H[T
h
] states that the operator T
h
satisﬁes the
Lipschitz condition with the coefﬁcient equal to 1.
Assumption (2.5) implies that the function w is approximated by T
h
[w
h
] and the error of
this approximation is estimated by γ(h).
We formulate a difference problem corresponding to (1.1), (1.2). Write
δ
0
z = (δ
0
z
1
, . . . , δ
0
z
k
),
F[z]
(r,m)
= (F
1
[z]
(r,m)
, . . . , F
k
[z]
(r,m)
)
and
F
i
[z]
(r,m)
= f
i
(t
(r)
, x
(m)
, T
h
z
[r,m]
, s
i
δz
(r,m)
i
+ (1 −s
i
)δz
(r+1,m)
i
), 1 ≤ i ≤ k,
where
s
i
δz
(r,m)
i
= (s
i1
δ
1
z
(r,m)
i
, . . . , s
in
δ
n
z
(r,m)
i
),
(1 −s
i
)δz
(r+1,m)
i
= ((1 −s
i1

1
z
(r+1,m)
i
, . . . , (1 −s
in

n
z
(r+1,m)
i
)
and where 0 ≤ s
ij
≤ 1, i = 1, . . . , k, j = 1, . . . , n, are given constants. We consider the
difference functional system
δ
0
z
(r,m)
= F[z]
(r,m)
(2.6)
with initial boundary condition
z
(r,m)
= ϕ
(r,m)
h
on E
h.0
∪ ∂
0
E
h
(2.7)
where ϕ
h
: E
h.0
∪ ∂
0
E
h
→ R
k
, ϕ
h
= (ϕ
h.1
, . . . , ϕ
h.k
), are given function.
Classical difference methods for partial differential or functional differential equations
consist in replacing partial derivatives by difference expressions. Then the original problem
is transformed into difference or difference functional equations.
In recent years, a number of papers concerning numerical methods for functional partial
differential equations have been published.
Difference methods for nonlinear ﬁrst order partial functional equations were studied
in [1], [2], [3], [4], [6]. The main question in this investigations is to construct a differ-
ence functional equation which satisﬁes consistency conditions with respect to the original
problem and to ﬁnd sufﬁcient conditions for the stability of the difference schemes.
Our difference functional problems have the following property: each equation in system
(2.6) contains the parameters s
i
= (s
i1
, . . . , s
in
), 1 ≤ i ≤ k. If s
i
= (0, . . . , 0) ∈ R
n
for
Numerical Methods for Systems of Nonlinear Differential Functional Equations 21
1 ≤ i ≤ k then (2.6), (2.7) reduces to the explicit difference scheme. It is clear that there
exists exactly one solution of problem (2.6), (2.7) in this case. Sufﬁcient conditions for the
convergence of the explicit difference methods for ﬁrst order partial differential equations
can be found in the monograph [5] (Chapter V).
Note that if k = 1 and s = (s
1
, . . . , s
n
) = (1, . . . , 1) ∈ R
n
then (2.6), (2.7) reduces to
the implicit difference scheme considered in [7].
Numerical methods for nonlinear parabolic problems were investigated in [8]-[10]. Dif-
ference schemes considered in the above papers depend on two parameters s, ¯s ∈ [0, 1].
Right hand sides of difference equations corresponding to parabolic equations contain the
expressions
sδz
(r,m)
+ (1 −s)δz
(r+1,m)
and ¯sδ
(2)
z
(r,m)
+ (1 −¯ s)δ
(2)
z
(r+1,m)
,
where δ = (δ
1
, . . . , δ
n
) and δ
(2)
= [δ
ij
]
i,j=1,...,n
are difference operators corresponding
to the partial derivatives ∂
x
= (∂
x
1
, . . . , ∂
xn
) and ∂
xx
= [∂
x
i
x
j
]
i,j=1...,n
and z is a scalar
unknown function. Our results are motivated by papers [8]-[10].
In the paper we start the investigations of difference schemes for nonlinear mixed prob-
lems. We prove that under natural assumptions on given functions and on the mesh there is
a class of difference schemes for a mixed problem which is convergent.
The paper is organized as follows. In Section 2 we construct a class of difference schemes
for (1.1), (1.2). The convergence theorem and an error estimate for considering difference
methods are presented in Section 3.
3 SOLVABILITY AND CONVERGENCE OF DIFFERENCE METHODS
We ﬁrst prove that there exists exactly one solution u
h
: E

h
→ R
k
of problem (2.6), (2.7).
For each x
(m)
∈ B
h
we put

(m)
= {x
(m+e
j
)
: 1 ≤ j ≤ κ} ∪ {x
(m−e
j
)
: κ + 1 ≤ j ≤ n}.
Assumption H[f]. Suppose that the function f : Ω → R
k
, f = (f
1
, . . . , f
k
), of the
variables (t, x, w, q) is such that
1) for each P = (t, x, w, q) ∈ Ω there exist partial derivatives

q
f(P) = [∂
q
j
f
i
(P)]
i=1,...,k,j=1,...,n
and ∂
q
f
h
(t, x, w, ·) ∈ C(R
n
, M
k×n
),
2) for each P ∈ Ω and for i = 1, . . . , k the estimates

q
j
f
i
(P) ≥ 0 for 1 ≤ j ≤ κ, ∂
q
j
f
i
(P) ≤ 0 for κ + 1 ≤ j ≤ n
are satisﬁed.
22 Szafranska
Lemma 3.1 If Assumption H[f] is satisﬁed and ϕ
h
: E
h.0
∪ ∂
0
E
h
→ R
k
then there exists
exactly one solution u
h
: E

h
→ R
k
of (2.6), (2.7).
Proof. Is follows from (2.7) that u
h
is deﬁned on E
h.0
. Suppose that 0 ≤ r < K is ﬁxed
and that u
h.i
is deﬁned on E

h
∩([−d
0
, t
r
] ×R
n
) for 1 ≤ i ≤ k. Assume now that i is ﬁxed,
1 ≤ i ≤ k. Consider the problem
z
(r+1,m)
i
= u
(r,m)
h.i
+h
0
f
h.i
(t
(r)
, x
(m)
, (u
h
)
[r,m]
, s
i
δu
(r,m)
i
+ (1 −s
i
)δz
(r+1,m)
i
) (3.1)
u
(r+1,m)
h.i
= ϕ
(r+1,m)
h.i
for x
(m)
∈ ∂
0
B
h
. (3.2)
Suppose now that the numbers u
h.i
(t
(r+1)
, y) where y ∈ ∆
(m)
are known. Write
ψ
i
(τ) = u
(r,m)
h.i
+h
0
f
h.i
(t
(r)
, x
(m)
, (u
h
)
[r,m]
, Q
(r,m)
i
(τ)),
where
Q
(r,m)
i
(τ) =
_
1
h
1
_
s
i1
(u
(r,m+e
1
)
h.i
−u
(r,m)
h.i
) + (1 −s
i1
)(u
(r+1,m+e
1
)
h.i
−τ)
_
, . . . ,
1
h
κ
_
s

(u
(r,m+eκ)
h.i
−u
(r,m)
h.i
) + (1 −s

)(u
(r+1,m+eκ)
h.i
−τ)
_
,
1
h
κ+1
_
s
iκ+1
(u
(r,m)
h.i
−u
(r,m−e
κ+1
)
h.i
) −(1 −s
iκ+1
)(τ −u
(r+1,m−e
κ+1
)
h.i
)
_
, . . . ,
1
h
n
_
s
in
(u
(r,m)
h.i
−u
(r,m−en)
h.i
) −(1 −s
in
)(τ −u
(r+1,m−en)
h.i
)
_
_
,
Then ψ = (ψ
1
, . . . , ψ
k
) : R → R
k
is of class C
1
and
ψ

i
(τ) = −h
0
n

j=1
1
h
j
(1 −s
ij
)|∂
q
j
f
h.i
(t
(r)
, x
(m)
, (u
h
)
[r,m]
, Q
(r,m)
i
(τ))| ≤ 0
for τ ∈ R. Then equation τ = ψ
i
(τ) has exactly one solution and consequently the number
u
(r+1,m)
h.i
can be calculated. Since u
(r+1,m)
h.i
is given for x
(m)
∈ ∂
0
B
h
it follows that there
exists exactly one solution u
(r+1,m)
h.i
of (3.3), (3.4) for x
(m)
∈ B
h
. Then u
h.i
is deﬁned on
E
h.r+1
. Then by induction the solution exists and it is unique on E

h
.
We proof now the convergence of the difference method (2.6), (2.7).
Assumption H[f, σ]. Suppose that
1) the function σ : [0, a] ×R
+
→ R
+
satisﬁes the following conditions:
(i) σ(t, ·) : R
+
→ R
+
is continuous and nondecreasing for each t ∈ [0, a],
(ii) the maximal solution of the Cauchy problem
w

(t) = σ(t, w(t)), w(0) = 0, (3.3)
is ¯ w(t) = 0 for t ∈ [0, a],
Numerical Methods for Systems of Nonlinear Differential Functional Equations 23
2) the estimate
f(t, x, w, q) −f(t, x, ¯ w, q)

≤ σ(t, w − ¯ w
D
) (3.4)
is satisﬁed on Ω.
Theorem 3.2 Suppose that Assumptions H[f], H[f, σ] are satisﬁed and
1) the function v : E

→ R
k
is a solution of (1.1), (1.2) and is of class C
1
,
2) h ∈ H and the function u
h
: E

h
→ R
k
is a solution of (2.6), (2.7) and there is α
0
: H →
R
+
such that
ϕ
(r,m)
−ϕ
(r,m)
h

≤ α
0
(h) on E
h.0
∪ ∂
0
E
h
and lim
h→0
α
0
(h) = 0,
3) the operator T
h
: F(D
h
, R
k
) → C(D, R
k
) satisﬁes the Assumption H[T
h
],
4) for P ∈ Ω we have
1 −h
0
n

j=1
1
h
j
s
ij
|∂
q
j
f
i
(P)| ≥ 0, 1 ≤ i ≤ k. (3.5)
Then there exists a function α : H → R
+
such that
v
(r,m)
h
−u
(r,m)
h

≤ α(h) on E
h
and lim
h→0
α(h) = 0. (3.6)
Proof. Let the function Γ
h
: E

h
→ R
k
be deﬁned by
δ
0
v
(r,m)
h
= F[v
h
]
(r,m)
+ Γ
(r,m)
h
on E

h
. (3.7)
It follows from the assumptions of theorem that there exists a function ¯ γ : H → R
+
such
that
Γ
(r,m)
h

≤ ¯γ(h) on E

h
and lim
h→0
¯ γ(h) = 0.
We write
ε
(r)
h
= max {(v
h
−u
h
)
(i,m)

: (t
(i)
, x
(m)
) ∈ E

h
∩ ([−d
0
, t
(r)
] ×R
n
)}.
It follows from the deﬁnitions of difference operator δ
0
and δ that
(v
h.i
−u
h.i
)
(r+1,m)
_
1 +h
0
n

j=1
1
h
j
(1 −s
ij
)|∂
q
j
f
i
(P
ij
)|
_
= (v
h.i
−u
h.i
)
(r,m)
_
1 −h
0
n

j=1
1
h
j
s
ij
|∂
q
j
f
i
(P
ij
)|
_
+h
0
κ

j=1

q
j
f
i
(P
ij
)
_
s
ij
1
h
j
(v
h.i
−u
h.i
)
(r,m+e
j
)
+ (1 −s
ij
)
1
h
j
(v
h.i
−u
h.i
)
(r+1,m+e
j
)
_
−h
0
n

j=κ+1

q
j
f
i
(P
ij
)
_
s
ij
1
h
j
(v
h.i
−u
h.i
)
(r,m−e
j
)
+ (1 −s
ij
)
1
h
j
(v
h.i
−u
h.i
)
(r+1,m−e
j
)
_
24 Szafranska
+h
0
_
f
i
(t
(r)
, x
(m)
, (v
h
)
(t
(r)
,x
(m)
)
, s
i
δv
(r,m)
h.i
+ (1 −s
i
)δv
(r+1,m)
h.i
)
−f
i
(t
(r)
, x
(m)
, (T
h
u
h
)
[r,m]
, s
i
δv
(r,m)
h.i
+ (1 −s
i
)δv
(r+1,m)
h.i
)
_
+h
0
Γ
(r,m)
h.i
, 1 ≤ i ≤ k,
where P
ij
∈ Ω are intermediate points. According to Assumption H[T
h
] we have the
estimate
(v
h
)
(t
(r)
,x
(m)
)
−(T
h
u
h
)
[r,m]

D
≤ γ(h) +ε
(r)
h
.
Then ε
(r)
h
satisﬁes the difference inequality
ε
(r+1)
h
≤ ε
(r)
h
+h
0
σ(t
(r)
, γ(h) +ε
(r)
h
) +h
0
¯ γ(h), 0 ≤ r ≤ K −1. (3.8)
Let us consider the Cauchy problem
w

(t) = σ(t, γ(h) +w(t)) + ¯ γ(h), (3.9)
w(0) = α
0
(h). (3.10)
It follows from condition 1)-(ii) of Assumption H[f, σ] that there exists the maximal solu-
tion ¯ w
h
of the problem (3.9), (3.10) and ¯ w
h
is deﬁned on [0, a]. Moreover
lim
h→0
¯ w
h
(t) = 0 uniformly on [0, a].
It is easily seen that ¯ w
h
satisﬁes the recurrent inequality
¯ w
(r+1)
h
≥ ¯ w
(r)
h
+h
0
σ(t
(r)
, γ(h) + ¯ w
(r)
h
) +h
0
¯γ(h), 0 ≤ r ≤ K −1
and it follows from (3.10) that the inequality
¯ w
(r)
h
≤ α
0
(h), −K
0
≤ r ≤ 0
holds. By the above relations and (3.8) we have
ε
(r)
h
≤ ¯ w
(r)
h
, 0 ≤ r ≤ K.
Thus we get (3.6) for α(h) = ¯ w
h
(a). This completes the proof.
Now we give an example of the operator T
h
satisfying Assumption H[T
h
] and the error
estimate for the difference method (2.6), (2.7).
Put
S

= {(j, s) : j ∈ {0, 1}, s = (s
1
, . . . , s
n
), s
i
∈ {0, 1} for 1 ≤ i ≤ n}.
Let w ∈ F(D
h
, R
k
) and (t, x) ∈ D. There exists (t
(r)
, x
(m)
) ∈ D
h
such that
t
(r)
≤ t ≤ t
(r+1)
, x
(m)
≤ x ≤ x
(m+1)
, (t
(r+1)
, x
(m+1)
) ∈ D
h
.
We deﬁne
(T
h
w)(t, x) =

(j,s)∈S∗
w
(r+j,m+s)
_
Y −Y
(r,m)
h
_
(j,s)
_
1 −
Y −Y
(r,m)
h
_
1−(j,s)
Numerical Methods for Systems of Nonlinear Differential Functional Equations 25
where
_
Y −Y
(r,m)
h
_
(j,s)
=
_
t −t
(r)
h
0
_
j
n

k=1
_
x
k
−x
(m
k
)
k
h
k
_
s
k
and
_
1 −
Y −Y
(r,m)
h
_
1−(j,s)
=
_
1 −
t −t
(r)
h
0
_
1−j
n

k=1
_
1 −
x
k
−x
(m
k
)
k
h
k
_
1−s
k
and we take 0
0
= 1 in the above formulas.
Lemma 3.3 Suppose that
1) the solution v : E

→ R
k
of differential problem (1.1), (1.2) is of class C
2
and the
assumptions of Theorem 3.1 are satisﬁed with σ(t, p) = Lp, L > 0,
2) there exist

M ∈ R
+
and
¯
C ∈ R
+
such that

q
f(t, x, z, q) ≤

M,

x
j
v(t, x)

, ∂
tt
v(t, x)

, ∂
x
j
x
j
v(t, x)

¯
C
where j = 1, . . . , n.
Then
u
(r,m)
h
−v
(r,m)
h

≤ ¯ η
(r)
h
, (3.11)
where
¯ η
(r)
h
= α
0
(h)(1 +h
0
L)
r
+ ¯ γ(h)
(1 +h
0
L)
r
−1
L
and
¯ γ(h) = Lγ(h) + ¯ γ(h), ¯γ(h) =
1
2
¯
Ch
0
+Lγ(h) +
¯
C(2 +h

)

M.
Proof. From the assumptions of Lemma we conclude that the operators δ
0
, δ satisfy the
following conditions
δ
0
v
(r,m)
−∂
t
v
(r,m)

1
2
¯
Ch
0
,
δ
j
v
(r,m)
−∂
x
j
v
(r,m)

1
2
¯
Ch

, j = 1, . . . , n.
It follows from above estimates and from Assumption H[T
h
] that
Γ
(r,m)
h.i
= δ
0
v
(r,m)
i
−∂
t
v
(r,m)
i
+f
i
(t
(r)
, x
(m)
, v
(r,m)
, v
(t
(r)
,x
(m)
)
, ∂
x
v
(r,m)
i
)
−f
i
(t
(r)
, x
(m)
, v
(r,m)
, T
h
v
[r,m]
, s
i
δv
(r,m)
i
+ (1 −s
i
)δv
(r+1,m)
i
).
Then
Γ
(r,m)
h

≤ ¯γ(h).
The function ¯ η
h
is a solution of the problem
η
(r+1)
= η
(r)
(1 +h
0
L) +h
0
[Lγ(h) +¯γ(h)], 1 ≤ i ≤ k.
Then from Theorem 3.1 we get the assertion (3.11).
26 Szafranska
4 NUMERICAL EXAMPLES
Example 4.1 For n = 2, k = 1 we put
E = {(t, x, y) : t ∈ [0, a], x ∈ [−1, 1], y ∈ [−1, 1]}. (4.1)
Consider the differential integral equation

t
z(t, x, y) = ∂
x
z(t, x, y) −∂
y
z(t, x, y) (4.2)
−sin
_

x
z(t, x) −∂
y
z(t, x, y) +t
2
_
1
x
z(t, τ, y)dτ
+t
2
_
y
−1
z(t, x, τ)dτ −te
t(1−y)
−te
t(x+1)
_
−(2t −x +y)e
t(x−y)
.
with the initial boundary condition
z(0, x, y) = 1, x ∈ [−1, 1], y ∈ [−1, 1], (4.3)
z(t, 1, y) = e
t(1−y)
, t ∈ [0, a], y ∈ [−1, 1],
z(t, x, −1) = e
t(x+1)
, t ∈ [0, a], x ∈ [−1, 1].
The exact solution of this problem is known. It is v(t, x) = e
t(x−y)
. Put h = (h
0
, h
1
, h
2
)
stand for the steps of the mesh on E.
Difference method for the problem (4.2), (4.3) has the form
z
(r+1,m
1
,m
2
)
= z
(r,m
1
,m
2
)
+h
0
_

1
z
(r,m
1
,m
2
)
+ (1 −s)δ
1
z
(r+1,m
1
,m
2
)
(4.4)
−sδ
2
z
(r,m
1
,m
2
)
−(1 −s)δ
2
z
(r+1,m
1
,m
2
)
−sin
_

1
z
(r,m
1
,m
2
)
+ (1 −s)δ
1
z
(r+1,m
1
,m
2
)
−sδ
2
z
(r,m
1
,m
2
)
−(1 −s)δ
2
z
(r+1,m
1
,m
2
)
+ (t
(r)
)
2
_
1
x
(m
1
)
z(t
(r)
, τ, y
(m
2
)
)dτ
+(t
(r)
)
2
_
y
(m
2
)
−1
z(t
(r)
, x
(m
1
)
, τ)dτ −t
(r)
exp(t
(r)
(1 −y
(m
2
)
)) −t
(r)
exp(t
(r)
(x
(m
1
)
+ 1))
_
−(2t
(r)
−x
(m
1
)
+y
(m
2
)
)z
(r,m
1
,m
2
)
_
.
We put s = s
1
= s
2
. Let us denote by z
h
: E
h
→ R the solution of the explicit difference
problemcorresponding to (4.2), (4.3) which we get from (4.4) with s = 1. By u
h
: E
h
→ R
we denote the solution of the implicit difference problem received form (4.4) with s = 0. If
we put in (4.4) s = 0.5 we get the implicit difference method and by its solution we denote
˜ u
h
: E
h
→ R.
Write
η
(r)
h
=
1
(2N + 1)
2
N

m
1
=−N
N

m
2
=−N
|z
(r,m
1
,m
2
)
h
−v
(r,m
1
,m
2
)
|, (4.5)
Numerical Methods for Systems of Nonlinear Differential Functional Equations 27
¯ η
(r)
h
=
1
(2N + 1)
2
N

m
1
=−N
N

m
2
=−N
|u
(r,m
1
,m
2
)
h
−v
(r,m
1
,m
2
)
|, (4.6)
˜ η
(r)
h
=
1
(2N + 1)
2
N

m
1
=−N
N

m
2
=−N
|˜ u
(r,m
1
,m
2
)
h
−v
(r,m
1
,m
2
)
|, (4.7)
The numbers η
(r)
h
, ¯ η
(r)
h
, ˜ η
(r)
h
are the arithmetical mean of the errors with ﬁxed t
(r)
. The
values of the functions η
(r)
h
, ¯ η
(r)
h
, ˜ η
(r)
h
are listed in the tables. We write “x” for η
h
> 100.
Table of errors for h = (0.01, 0.01, 0.01)
η
h
˜ η
h
t = 0.5 0.001554 0.001664
t = 1.0 0.002772 0.003799
t = 1.5 0.004161 0.009324
t = 2.0 x 0.011336
t = 2.5 x 0.189344
Table of errors for h = (0.05, 0.005, 0.005)
¯ η
h
t = 0.1 0.001631
t = 0.2 0.003236
t = 0.3 0.004798
t = 0.4 0.006284
t = 0.5 0.006986
Our experiments have the following property. The explicit method for steps h =
(0.01, 0.01, 0.01) which are not satisfy the condition (CFL)
1 −h
0
2

j=1
1
h
j
s|∂
q
j
f(P)| ≥ 0, (4.8)
with parameter s = 1, is not stable. The difference method with s = 0.5 for the same
steps gives better results. For the steps h = (0.05, 0.005, 0.005) which are not satisfy the
condition (4.8) with s = 0.5, the difference method is not stable. The implicit difference
method, which we received with s = 0, is stable aside from steps of the mesh.
Example 4.2 For n = 2, k = 1 we put
E = {(t, x, y) : t ∈ [0, a], x ∈ [−1, 1], y ∈ [−1, 1]}. (4.9)
Consider the differential integral equation

t
z(t, x, y) = ∂
x
z(t, x, y) + cos
_

x
z(t, x, y) −tyz(t, x, y)
_
(4.10)
28 Szafranska
−∂
y
z(t, x, y) −sin
_

y
z(t, x, y) −txz(t, x, z)] +z(t,
x −y
2
,
x +y
2
)
+(xy −ty +tx)z(t, x, y) −exp (
1
4
t(x
2
−y
2
)) −1.
with the initial boundary condition
z(0, x, y) = 1, x ∈ [−1, 1], y ∈ [−1, 1], (4.11)
z(t, 1, y) = e
ty
, t ∈ [0, a], y ∈ [−1, 1],
z(t, x, −1) = e
−tx
, t ∈ [0, a], x ∈ [−1, 1].
The exact solution of this problem is known. It is v(t, x) = e
txy
. Put h = (h
0
, h
1
, h
2
) stand
for the steps of the mesh on E.
Difference method for the problem (4.2), (4.3) has the form
z
(r+1,m
1
,m
2
)
= z
(r,m
1
,m
2
)
+h
0
_

1
z
(r,m
1
,m
2
)
+ (1 −s)δ
1
z
(r+1,m
1
,m
2
)
(4.12)
+cos
_

1
z
(r,m
1
,m
2
)
+ (1 −s)δ
1
z
(r+1,m
1
,m
2
)
−t
(r)
y
(m
1
)
z
(r,m
1
,m
2
)
_
−sδ
2
z
(r,m
1
,m
2
)
−(1 −s)δ
2
z
(r+1,m
1
,m
2
)
−sin
_

2
z
(r,m
1
,m
2
)
+ (1 −s)δ
2
z
(r+1,m
1
,m
2
)
−t
(r)
x
(m
2
)
z
(r,m
1
,m
2
)
_
+z(t
(r)
, 0.5(x
(m
1
)
−y
(m
2
)
), 0.5(x
(m
1
)
+y
(m
2
)
)) +(x
(m
1
)
y
(m
2
)
t
(r)
y
(m
2
)
+t
(r)
x
(m
1
)
)z
(r,m
1
,m
2
)
−exp
_
1
4
t
(r)
((x
(m
2
)
)
2
−(y
(m
2
)
)
2
)
_
−1
_
We put s = s
1
= s
2
. Let us denote by z
h
: E
h
→ R the solution of the explicit difference
problem corresponding to (4.10), (4.11) which we get from (4.12) with s = 1. By u
h
:
E
h
→ R we denote the solution of the implicit difference problem received form (4.12)
with s = 0. If we put in (4.12) s = 0.5 we get the implicit difference method and by its
solution we denote ˜ u
h
: E
h
→ R.
The numbers η
(r)
h
, ¯ η
(r)
h
, ˜ η
(r)
h
, given by (4.5)-(4.7) respectively, are the arithmetical mean
of the errors with ﬁxed t
(r)
. The values of the functions η
(r)
h
, ¯ η
(r)
h
, ˜ η
(r)
h
are listed in the
tables. We write “x” for η
h
> 100.
Table of errors for h = (0.01, 0.01, 0.01)
η
h
˜ η
h
t = 0.1 0.032645 0.000487
t = 0.2 x 0.000890
t = 0.3 x 0.001229
t = 0.4 x 0.001525
t = 0.5 x 0.001769
Numerical Methods for Systems of Nonlinear Differential Functional Equations 29
Table of errors for h = (0.05, 0.005, 0.005)
¯ η
h
t = 0.1 0.005775
t = 0.2 0.010111
t = 0.3 0.013670
t = 0.4 0.016837
t = 0.5 0.018405
Conclusion from above experiment is that implicit difference method received from (4.12)
with s = 0, is stable for any choose of steps h = (h
0,
, h
1
, h
2
). For stability of the explicit
difference method and implicit difference method received from (4.12) with s = 0.5, we
need satisfying the condition (CFL) given by (4.8).
References
[1] A. Baranowska, Numerical methods for nonlinear ﬁrst order partial differential equa-
tions with deviated variables. Numer. Methods Partial Differential Equations, v. 22,
no. 3, pp. 708–727, 2006.
[2] A. Baranowska, Z. Kamont, Numerical approximation of ﬁrst order partial differential
equations with deviated variables. Comment. Math. Prace. Mat., v. 43, no. 1, pp. 1–31,
2003.
[3] P. Brandi, Z. Kamont, A. Salvadori, Difference methods for nonlinear ﬁrst order par-
tial differential equations with mixed initial boundary conditions. Math. Balkanica, v.
10, no. 2–3, pp. 249–269, 1996.
[4] Z. Kamont, Finite difference approximations for ﬁrst order partial differential func-
tional equations. Ukr. Math. J., v. 46, pp. 985–996, 1994.
[5] Z. Kamont, Hyperbolic Functional Differential Inequalities and Applications. Dor-
[6] Z. Kamont, K. Prza¸dka, Difference methods for ﬁrst order partial differential func-
tional equations with initial - boundary condtions. J. Vycisl. Mat. i Math. Fis., v. 31,
pp. 1476–1488, 1991.
[7] A. Ke¸pczy´ nska, Implicit difference functional inequalities and applications. Math.
Balk., v. 20, no 2, pp. 167–184, 2006.
[8] M. Malec, Sur une famille bi-param´ etrique de sch´ emas des diff´ erences ﬁnies pour
l’´ equation parabolique sans d´ eriv´ ees mixtes. Ann. Polon. Math., v. 31, pp. 47–54,
1975.
30 Szafranska
[9] M. Malec, Sur une famille bi-param´ etrique des sch´ emas des diff´ erences pour les
syst´ emes paraboliques. Bull. Acad. Polon. Sci. Ser. Sci. Math. Astronom. Phys., v.
23, no. 8, pp. 871–875, 1975.
[10] M. Malec, Sur une famille biparam´ etrique de sch´ emas des diff´ erences ﬁnies pour
un syst´ eme d’´ equations paraboliques aux d´ eriv´ ees mixtes et avec des conditions aux
limites du type de Neumann. Ann. Polon. Math., v. 32, pp. 33–42, 1976.