J. Numer. Anal. Approx. Theory, vol. 44 (2015) no. 2, pp. 154–165 ictp.acad.ro/jnaat
EXISTENCE AND APPROXIMATION OF SOLUTIONS TO BOUNDARY VALUE PROBLEMS
FOR DELAY INTEGRO-DIFFERENTIAL EQUATIONS
IGOR CHEREVKO∗and ANDREW DOROSH∗
Abstract. The paper deals with existence, uniqueness and spline approximation of solutions to boundary value problems for delay integro-differential equations.
An iterative approximation scheme based on the use of cubic splines with defect two is presented, and sufficient conditions for its convergence are obtained.
MSC 2010. 34A45, 34K10, 34K28, 65L03, 65L10.
Keywords. Boundary value problems, integro-differential equations, delay, spline functions.
1. INTRODUCTION
Dynamic processes in many applied problems are described by delay dif- ferential and integral equations (Andreeva, Kolmanovsky and Shayhet 1992).
An analytical solution of such equation exists only in the simplest cases, so the construction and study of approximate algorithms for solutions of these equations are important.
In the present note we study an approximate method of solving boundary value problems for delay integro-differential equations based on approximation of the solution by cubic splines with defect two.
Existence and uniqueness of a solution of delay boundary value problems in various function spaces were considered by Grim and Schmitt (1968), Ka- mensky and Myshkis (1972), Biga and Gaber (2007), Athanasiadou (2013).
Applying spline functions for solving differential-difference equations was in- vestigated by Nikolova and Bainov (1981), Cherevko and Yakimov (1989), Nastasyeva and Cherevko (1999).
∗Faculty of Mathematics and Informatics, Yuriy Fedkovych Chernivtsi National Univer- sity, Ukraine, e-mail: [email protected],[email protected].
2. PROBLEM STATEMENT. EXISTENCE OF A SOLUTION
Let us consider the following boundary value problem
y00(x) =f x, y(x), y(x−τ0(x)), y0(x), y0(x−τ1(x))+ (1)
+ Z b
a
g x, s, y(s), y(s−τ0(s)), y0(s), y0(s−τ1(s))ds, y(i)(x) =ϕ(i)(x), i= 0,1, x∈[a∗;a], y(b) =γ,
(2) where
a∗= minn inf
x∈[a;b](x−τ0(x)), inf
x∈[a;b](x−τ1(x))o, γ∈R, τ0(x)≥0, τ1(x)≥0.
Letf(x, u0, u1, v0, v1), g(x, s, u0, u1, v0, v1) be continuous functions inG= [a, b]×G21×G22 and Q = [a, b]×G, whereG1 ={u ∈R: |u|< P1}, G2 = {v ∈R: |v| ≤P2}, P1, P2 are positive constants, ϕ(t) ∈ C1[a∗;a], delays τ0(x) and τ1(x) are continuous functions on [a, b], and additionally, τ1(x) is such that the setE ={xi∈[a, b] : xi−τ1(xi) =a, i= 1, k}is finite.
We introduce the notations:
P = sup
f(x, u, u1, v, v1)+ Z b
a
g(x, s, u, u1, v, v1)ds:
|ui|< P1,|vi|< P2, i= 0,1, x, s∈[a, b]
, J = [a∗;a], I = [a, b], I1= [a, x1], I2= [x1, x2], . . . , Ik= [xk−1, xk], Ik+1 = [xk, b],
B J∪I=ny(x) :y(x)∈C(J∪I)∩ C1(J)∪C1(I)∩k+1S
j=1
C2(Ij),
|y(x)| ≤P1, |y0(x)| ≤P2
o.
A function y = y(x) from the space B(J∪I) is called a solution of the problem (1)-(2) if it satisfies the equation (1) on [a;b] (with the possible ex- ception of the set E) and boundary conditions (2).
From the definition of the space B(J∪I) we conclude that the solution of the problem (1)-(2) is continuously differentiable for each x ∈ [a, b] where y0(a) is the right derivative.
Let us introduce a norm in the space B(J∪I):
kykB= max
8
(b−a)2 max
x∈J∪I |y(x)|,b−a2 maxnmax
x∈J
y0(x),max
x∈I
y0(x)
o . The spaceB(J∪I) with this norm is a Banach space.
The boundary value problem (1)-(2) is equivalent to the following integral equation (Grim and Schmitt 1968; Kamensky and Myshkis 1972)
y(x) = Z b
a∗
"
f s, y(s), y(s−τ0(s)), y0(s), y0(s−τ1(s))+ (3)
+ Z b
a
g s, ξ, y(ξ), y(ξ−τ0(ξ)), y0(ξ), y0(ξ−τ1(ξ))dξ
#
G¯(x, s)ds +l(x), x∈J∪I,
where G¯(x, s) =
(G(x, s), x, s∈I,
0, otherwise, l(x) =
(ϕ(x), x∈J,
γ−ϕ(a)
b−a (x−a) +ϕ(a), x∈I, and G(x, s) is the Green’s function of the following boundary value problem
y00(x) = 0, x∈I, y(a) =y(b) = 0.
We define an operatorT in the spaceB(J∪I) in the following way (T y) (x) =
Z b a∗
"
f s, y(s), y(s−τ0(s)), y0(s), y0(s−τ1(s))+ +
Z b a
g s, ξ, y(ξ), y(ξ−τ0(ξ)), y0(ξ), y0(ξ−τ1(ξ))dξ
#
G¯(x, s)ds +l(x), x∈J∪I.
Hence,
(T y)0(x) = Z b
a∗
"
f s, y(s), y(s−τ0(s)), y0(s), y0(s−τ1(s))+ (4)
+ Zb
a
g s, ξ, y(ξ), y(ξ−τ0(ξ)), y0(ξ), y0(ξ−τ1(ξ))dξ
#
G¯0x(x, s)ds +γ−ϕ(a)b−a , x∈J∪I.
Theorem 1. Let the following conditions hold:
1) max
maxx∈J |ϕ(x)|, (b−a)
2
8 P+ max (|ϕ(a)|,|γ|)
≤P1, 2) max
maxx∈J |ϕ0(x)|, b−a2 P+γ−ϕ(a)b−a
≤P2,
3) the functions f(x, u0, u1, v0, v1), g(x, s, u0, u1, v0, v1) satisfy the Lip- schitz condition for variablesui, vi, i= 0,1 with constants L1j, L2j, j= 1,4 in G and Q,
4) (b−a)8 2
2
P
j=1
L1j + (b−a)L2j+b−a2
4
P
j=3
L1j+ (b−a)L2j<1.
Then there exists a unique solution of the problem (1)-(2) in B(J∪I).
Proof. Based on Green’s function (Hartman 2002)
G(t, s) =
(s−a)(t−b)
b−a , a≤s≤t≤b,
(t−a)(s−b)
b−a , a≤t≤s≤b, we obtain the following estimates
(5)
Z b a
|G(t, s)|ds≤ (b−a)8 2, Z b
a
G0t(t, s)ds≤ b−a2 .
When the conditions 1)-2) and the inequalities (5) are true the operatorT maps the space B(J∪I) on itself.
Let y1, y2 ∈B(J∪I). Considering the condition 3) and the estimates (5), we get:
|(T y1) (t)−(T y2) (t)| ≤
≤ Z b
a∗
L11+L12 max
t∈ J∪I |y1(t)−y2(t)|
+L13+L14max
maxt∈ I
y01(t)−y20 (t), max
t∈ J
y01(t)−y20 (t)
+ (b−a)L21+L22 max
t∈ J∪I |y1(t)−y2(t)|
+ (b−a)L23+L24max
max
t ∈I
y01(t)−y02(t),max
t∈ J
y01(t)−y02(t)
G¯(t, s)ds
≤ (b−a)8 2 (b−a)2
8
L11+L12+ (b−a)L21+L22 +b−a2 L13+L14+ (b−a)L23+L24
ky1−y2kB,
(T y1)0(t)−(T y2)0(t)≤
≤ b−a2 (b−a)2
8
L11+L12+ (b−a)L21+L22 +b−a2 L13+L14+ (b−a)L23+L24
ky1−y2kB.
Based on the obtained estimates and on the definition of the norm in the space B(J∪I) we have:
k(T y1) (t)−(T y2) (t)kB≤ (6)
≤
"
(b−a)2 8
2
X
i=1
L1i + (b−a)L2i+b−a2
4
X
i=3
L1i + (b−a)L2i
#
ky1−y2kB. The inequality (6) and the condition 4) imply that the operator T is a contraction inB(J∪I) and it has a single fixed point in this space, therefore the boundary value problem (1)-(2) has a unique solution y(t) ∈ B(J∪I).
The proof is complete.
3. CUBIC SPLINES WITH DEFECT TWO
Let us consider an irregular grid ∆ ={a=x0 < x1 < . . . < xn=b}on the segment [a;b], E ⊂ ∆. We denote by S(y, x) an interpolating cubic spline with defect two on ∆ which belongs to the spaceB(J∪I).
We can obtain a formula ofS(y, x) (Nikolova and Bainov 1981; Nastasyeva and Cherevko 1999; Dorosh and Cherevko 2014):
S(y, x) =Mj−1+ (xj−x)
3
6hj +Mj−(x−xj−1)
3
6hj +
yj−1−M
+ j−1h2j
6
xj−x hj
+
yj−M
− j h2j
6
x−xj−1
hj , x∈[xj−1;xj], hj =xj −xj−1, j= 1, n, (7)
where Mj+ = S00(y, xj + 0), j = 0, n−1, Mj− = S00(y, xj−0), j = 1, n satisfy the following system of equations
hj+1yj−1−(hj+hj+1)yj+hjyj+1=
= hjh6j+1hjMj−1+ + 2hjMj−+ 2hj+1Mj++hj+1Mj+1− , j= 1, n−1, (8)
y0 =ϕ(a), yn=γ.
We shall present the equations (8) in a matrix form Ay =BM+d,
(9) where
A=
−(h1+h2) h1 0 0 · · · 0 h3 −(h2+h3) h2 0 · · · 0
· · · · 0 0 · · · 0 hn −(hn−1+hn)
is an (n−1)×(n−1) matrix, d = (−h2y0, 0, . . . , 0, −hn−1yn)T, B is a right side of the relations (8) coefficient matrix with dimensions(n−1)×2n,
M =M0+, M1−, M1+, M2−, M2+, . . . , Mn−1− , Mn−1+ , Mn−T.
Lemma. The following correlations are true:
1) det (A) = (−1)n−1h2h3. . . hn−1(b−a), (10)
2)kA−1k ≤ 8hK23 (b−a), (11)
3) max
1≤i<n−2 n−1
X
j=1
a−1i+1,j−a−1i,j≤ K22h(b−a)2 , (12)
4)kBk ≤H3, (13)
where h= min
i hi, H= max
i hi, K = Hh and a−1ij are elements of a matrixA−1. The proof of the lemma statements is easy to obtain by applying the prin- ciple of mathematical induction and using the structure of the matrices A, B.
4. COMPUTATIONAL SCHEME
A) Choose a cubic splineSy(0), xrandomly so that the boundary condi- tions (2) are enforced, for instance,Sy(0), x= γ−ϕ(a)b−a (x−a)+ϕ(a).
B) Using the original equation (1) and the spline Sy(k), x, find Mj+(k+1)=f
xj, Sy(k), xj+ 0, Sy(k), xj −τ0(xj) + 0, (14)
S0y(k), xj + 0, S0y(k), xj−τ1(xj) + 0
+ +
Z b a
g
xj, s, Sy(k), s, Sy(k), s−τ0(s), S0y(k), s, S0y(k), s−τ1(s)
ds, j= 0, n−1,
Mj−(k+1)=f
xj, Sy(k), xj−0, Sy(k), xj −τ0(xj)−0, (15)
S0y(k), xj −0, S0y(k), xj−τ1(xj)−0
+ +
Z b a
g
xj, s, Sy(k), s, Sy(k), s−τ0(s), S0y(k), s, S0y(k), s−τ1(s)
ds, j= 1, n.
In (14), (15) putS(p)y(k), t=ϕ(p)(t), p= 0,1 ift < a.
C) Computenyjk+1o, j= 0, n from the equations (8).
D) Build a cubic spline Sy(k+1), x according to (7), using the values of nyk+1j o, Mj+(k+1), Mj−(k+1). This spline will be the next approxima- tion.
Let us denote
λ1=L11+L12+ (b−a)L21+L22, λ2=L13+L14+ (b−a)L23+L24, u= K85(b−a)2+H82, v= K25 (b−a) +2H3 .
(16)
Theorem2. Assume that the conditions of Theorem1hold. If the following inequality is true
(17) θ=uλ1+vλ2<1,
then there exists H∗ > 0 such that for each 0 < H < H∗ the sequence of splines nSy(k), xo, k= 0,1. . . converges uniformly on [a;b].
Proof. The equation (10) implies that it is possible to construct an iterative spline sequence Sy(k), x, k = 0,1, . . . using the scheme A)-D). We shall demonstrate that the series
S(p)y(0), x+
∞
X
i=1
hS(p)y(i), x−S(p)y(i−1), xi, p= 0,1
are uniformly convergent on [a;b] and thus the sequences S(p)y(k), x, k = 0,1, . . . , p= 0,1 are also uniformly convergent.
Let us define scalar functionsy(x), M(x) on [a;b] and denote the following vectors
y¯=y(x1), . . . , y(xn−1)T,
M¯ =M(x0+ 0), M(x1−0), M(x1+ 0), . . . , M(xn−1−0), M(xn−1+ 0), M(xn−0)
T
.
We shall write the iterative algorithm A)-D) in a matrix form y¯(k+1)=A−1BM¯k+1+A−1d,
(18)
where the vector ¯M components are defined according to (14)-(15) and the constant vectorddepends only on the boundary conditions (2).
From (18) we obtain the estimate
y(k+1)−y(k)=A−1BMk+1−A−1BMk (19)
≤A−1kBkM¯k+1−M¯k.
From (14)-(15) and the properties of the functions f and g we obtain the following inequalities
Mj+(k+1)−Mj+(k)≤λ1max
x∈[a;b]
Sy(k), x−Sy(k−1), x (20)
+λ2max
x∈[a;b]
S0y(k), x−S0y(k−1), x, j = 0,1, . . . , n−1,
Mj−(k+1)−Mj−(k)≤λ1max
x∈[a;b]
Sy(k), x−Sy(k−1), x +λ2max
x∈[a;b]
S0y(k), x−S0y(k−1), x, j = 1,2, . . . , n.
Therefore, taking into account the above mentioned lemma, (19) can be writ- ten in the following way
y(k+1)−y(k)≤ K85 (b−a)
λ1Sy(k), x−Sy(k−1), x (21)
+λ2
S0y(k), x−S0y(k−1), x
. Letx∈[xj−1;xj]. Considering (7), we have
Sy(k+1), x−Sy(k), x≤x6hj−x
j
(xj−x)2−h2j+ + x−x6hj−1
j
(x−xj−1)2−h2j M¯k+1−M¯k+ (22)
+yk+1j−1 −yjkxjh−x
j
+yjk+1−yjkx−xhj−1
j
. It is easy to show that
max
x∈[xj−1;xj]
xj−x 6hj
h2j −(xj−x)2+x−x6hj−1
j
h2j −(x−xj−1)2≤ H82. (23)
Using (20), (21), (23), from (22) we obtain
Sy(k+1), x−Sy(k), x≤ H82 M¯k+1−M¯k+y(k+1)−y(k) (24)
≤K85(b−a)2+H82
λ1Sy(k), x−Sy(k−1), x+ +λ2
S0y(k), x−S0y(k−1), x
.
According to the spline (7), we get
S0y(k+1), x−S0y(k), x≤ (25)
≤
hj
6 −(xj2h−x)2
j
Mj−1+(k+1)−Mj−1+(k) +
(x−xj−1)2 2hj −h6j
Mj−(k+1)−Mj−(k)+h1
j
yk+1j −yj−1k+1−ykj −ykj−1. One can show that
max
x∈[xj−1;xj]
hj
6 − (xj2h−x)2
j
+
(x−xj−1)2 2hj −h6j
≤ 2H3 , (26)
max
1<j<n
yk+1j −yj−1k+1−ykj −yj−1k ≤ K24 (b−a)HM¯k+1−M¯k. (27)
Due to (26)-(27), the inequality (25) implies that
S0y(k+1), x−S0y(k), x≤
≤K25(b−a) +23H
λ1Sy(k), x−Sy(k−1), x (28)
+λ2
S0y(k), x−S0y(k−1), x
.
After iterating (24), (28) and considering the notations (16)-(17) we obtain
Sy(k+1), x−Sy(k), x≤ (29)
≤uθk−1λ1
Sy(1), x−Sy(0), x+λ2
S0y(1), x−S0y(0), x,
S0y(k+1), x−S0y(k), x≤
≤vθk−1λ1Sy(1), x−Sy(0), x+λ2S0y(1), x−S0y(0), x. The correlations (29) with the condition (17) ensure the convergence of the sequencesnS(p)y(k), xo, k= 0,1, . . . , p= 0,1. Theorem 2 is proved.
Let us denote lim
k→∞S(p)y(k), x = S(p)(˜y, x), p = 0,1, . . . Note that the parametersMfj+,Mfj−of the splineS(˜y, x) satisfy the system (8) and equations (14)-(15).
Let S(y, x) be a cubic spline with defect 2 which interpolates the solution y(x) of the boundary value problem (1)-(2). Thus,
S(p)(˜y, x)−y(p)(x)≤S(p)(˜y, x)−S(p)(y, x)
+S(p)(y, x)−y(p)(x), p= 0,1.
(30)
For the second term on the right side of the inequality (30) it is true (Alberg, Nilson and Walsh 1967) that
S(p)(y, x)−y(p)(x)≤KpH2−pω y00(x), H, (31)
p= 0,1,2, K0 = 52, K1=K2= 5, where ω(y00(x), H) = max
1≤r≤k+1ωr(y00(x), H), ωr(y00(x), H) is a modulus of continuity fory00(x) on Ir= [xr−1;xr].
We shall denote max
x∈[a;b]
S(p)(˜y, x)−S(p)(y, x)=αp, p= 0,1.
According to the properties of the functionsf,gand estimates (31), we obtain
Mj+−fxj, y(xj), y(xj−τ0(xj)), y0(xj), y0(xj−τ1(xj))
− Z b
a
gxj, s, y(s), y(s−τ0(s)), y0(s), y0(s−τ1(s))ds (32)
≤51 +12λ1H2+λ2Hω y00(x), H,
Mj−−fxj, y(xj), y(xj−τ0(xj)), y0(xj), y0(xj−τ1(xj))
− Z b
a
gxj, s, y(s), y(s−τ0(s)), y0(s), y0(s−τ1(s))ds (33)
≤51 +12λ1H2+λ2Hω y00(x), H.
Using the formulas of S(˜y, x), S(y, x) and the inequalities (32)-(33), one can get the following system of inequalities
α0 ≤uα0λ1+α1λ2+ 51 +12λ1H2+λ2Hω y00(x), H, α1 ≤vα0λ1+α1λ2+ 51 +12λ1H2+λ2Hω y00(x), H. (34)
Solving the system (34), we find estimates for the first terms on the right side of (30):
α0 ≤ 5(1+12λ1H2+λ2H)u
1−θ ω y00(x), H, α1≤ 5(1+12λ1H2+λ2H)v
1−θ ω y00(x), H. Now the inequalities (30) can be written in the following form
S(p)(˜y, x)−y(p)(x)≤Kpω y00(x), H, p= 0,1, (35)
whereK0 = sup
H≤H∗
uµ
1−θ+ 5H22, K1 = sup
H≤H∗
vµ
1−θ + 5H.
We can summarize the aforementioned arguments concerning accuracy of approximating the solution of the boundary value problem (1)-(2) based on the spline sequence as the following theorem.
Theorem3. Let the solution of the boundary value problem(1)-(2)exist, be unique and belong to the space B[a∗;b]. If the condition(17) holds, then there exists H∗ > 0 such that for any H < H∗ the spline sequence nSy(k), xo is approximating the solution of the boundary value problem (1)-(2) and the correlations (35) are true.
5. EXAMPLE
Let us consider the usage of this calculation scheme for finding an approxi- mate solution of the following boundary value problem
y00(x) =−αy0 x−π2+ Z π
2 0
y(t−π2)dt+ cosx, 0≤x≤ π2, y(x) = sin(x) + 1, −π2 ≤x <0,
y(0) = 1, y(π2) = 2 +α.
In this example L11 =L12 = L13 = 0, L14 = α, L21 = L23 = L24 = 0, L22 = 1, so λ1 = π2, λ2 = α, h = H = 40π, K = 1, u = π322 + H82, v = π4 + 23H, θ = π322 +H82π2 +π4 +23Hα. If we put α = 14, then θ ≈ 0.695 < 1 and therefore the conditions of the Theorems 1 and 2 are satisfied. The precise solutionyp(x) of this boundary value problem, which was found using the step method, is
yp(x) =αsinx−cosx+ π2 −1x22 +π4 1−π2x+ 2.
x ya(x) yp(x) ∆ δ
0 1 1 0 0%
π
8 1.03971 1.03976 0.00005 0.01%
π
4 1.2935 1.2936 0.00010 0.01%
3π
8 1.7162 1.7163 0.00010 0.01%
π
2 2.25 2.25 0 0%
Table 1. Precise and approximate solutions.
The results of the calculation are given in Table 1, whereyp(x) is the pre- cise solution, ya(x) is the approximate solution obtained with h= 40π after 2 iterations, ∆ is the absolute error and δ is the relative error.
REFERENCES
[1] J. Alberg, E. Nilson, J. Walsh, The theory of splines and their applications, Aca- demic, New York, 1967.
[2] E. Andreeva, V. Kolmanovsky, D. Shayhet, Control of systems with aftereffect, Nauka, Moscow, 1992.
[3] E.S. Athanasiadou,On the existence and uniqueness of solutions of boundary problems for second order functional differential equations, Mathematica Moravica,17(2013) 1, pp. 51–57.
[4] A.M. Bica, R. Gabor,Existence, uniqueness and approximation for the solution of a second order neutral differential equation with delay in Banach spaces, Mathematica,49 (2007) 2, pp. 117–130.
[5] A. Dorosh, I. Cherevko,Application of spline functions for approximating solutions of linear boundary value problems with delay, Mathematical and computer modelling.
Series: Physical and mathematical sciences,10(2014), pp. 80–88.
[6] I.M. Cherevko, I.V. Yakimov,Numerical method of solving boundary value problems for integro-differential equations with deviating argument, Ukrainian Mathematical Jour- nal,41(1989) 6, pp. 854–860.
[7] L.J. Grim, K. Schmitt,Boundary value problems for delay differential equations, Bull.
Amer. Math. Soc.,74(1968) 5, pp. 997–1000.
[8] F. Hartman, Ordinary differential equations, The Society for Industrial and Applied Mathematics, Philadelphia, 2002.
[9] G. Kamensky, A. Myshkis,Boundary value problems for nonlinear differential equa- tions with deviating argument of neutral type, Differential equations, 8 (1972) 12, pp. 2171–2179.
[10] N. Nastasyeva, I. Cherevko, Cubic splines with defect two and their applications to boundary value problems, Bulletin of Kyiv University. Physics and mathematics, 1 (1999), pp. 69–73.
[11] T.S. Nikolova, D.D. Bainov, Application of spline-functions for the construction of an approximate solution of boundary problems for a class of functional-differential equations, Yokohama Math. J.,29(1981) 1, pp. 108–122.
Received by the editors: June 17, 2015.