Professional Documents
Culture Documents
International Journal of
Computer Mathematics
Publication details, including instructions for authors and
subscription information:
http://www.tandfonline.com/loi/gcom20
To cite this article: Zhong Chen , CuiHua Wang & YongFang Zhou (2010): A new method for
solving Cauchy type singular integral equations of the second kind, International Journal of
Computer Mathematics, 87:9, 2076-2087
To link to this article: http://dx.doi.org/10.1080/00207160802600877
The publisher does not give any warranty express or implied or make any representation
that the contents will be complete or accurate or up to date. The accuracy of any
instructions, formulae, and drug doses should be independently verified with primary
sources. The publisher shall not be liable for any loss, actions, claims, proceedings,
demand, or costs or damages whatsoever or howsoever caused arising directly
or indirectly in connection with or arising out of the use of this material.
Downloaded by [UQ Library] at 07:28 18 June 2013
International Journal of Computer Mathematics
Vol. 87, No. 9, July 2010, 20762087
(Received 13 July 2008; revised version received 27 September 2008; accepted 08 October 2008 )
The exact solution and the approximate solution of Cauchy type singular integral equations of the
second kind are given. In order to remove the singularity of the solution at the endpoints and the Cauchy
singularity, a transform is used. By improving the traditional reproducing kernel method, which requests
1
the image space of the operator is W2 [1, 1] and the operator is bounded, the exact solution of Cauchy
type singular integral equations of the second kind is given. The advantage of the approach lies in the
fact that, on the one hand, the bounded approximate solution gn (x) is continuous; on the other hand, gn
(x) and gn (x), gn (x) converge uniformly to the bounded exact solution g(x) and its derivatives g (x), g
(x), respectively. Some numerical experiments show the efficiency of our method.
Keywords: Cauchy type singular integral equations; reproducing kernel; exact solution
1. Introduction
Consider the following Cauchy type singular integral equations of the second kind in
reproducing kernel space
b 1 (t ) dt + 1
a(x) + t x
K (x, t )(t ) dt = f (x), 1 < x < 1 (1)
1 1
subject to
1 (t ) dt = L, (2)
1
where a, b and L are real constants. The kernel function K (x, t ) and the known function f (x)
are assumed to be Hlder-continuous as 1 x, t 1, the unknown function (x) belongs to h0,
*Corresponding author. Email: cz0309@sina.com
a class composed of Hlder continuous functions on (1, 1) which have integrable singularity
at the neighbourhoods of endpoints x = 1. The first integral in Equation (1) is evaluated in the
sense of Cauchy principal value.
Equation (1) often arises in certain contact and fracture problems in solid mechanics [6,7].
In [8], the authors pointed out that Equation (1) with Equation (2) has a unique solution.
The numerical solution of Cauchy type singular integral equations were investigated by many
authors. Krenk [10,11] proposed the GaussJacobi method, LobattoJacobi method and Radau
Jacobi method for solving Equation (1). Du [24] gave a unified framework for various collocation
methods and a direct method for Cauchy type singular integral equations of the second kind.
Karczmarek et al. [9] present a method of constructing approximate solutions of Equation (1)
depending on the index of the characteristic equation. Jin et al. [8] presented a convenient and
efficient numerical method for solving Equation (1). Mohankumar and Natarajan [12] put for-ward
a polynomial collocation coupled with the DE quadrature for the numerical solution of Equation
(1). In addition, some papers discussed the solution of Cauchy type singular integral equation of the
first kind, which is obviously a special case of Equation (1) (a = 0, b = 1). For example, Erdogan
Downloaded by [UQ Library] at 07:28 18 June 2013
w(x) = (1 x) (1 + x) .
(3)
Rewrite the second term of Equation (1) as follows (see [8])
1 (t )
b b (1 t ) 1 + t )
1
(
g(t ) dt
1 t x dt = 1 t x
b1 g(t ) g(x) b 1 (1 t ) (1 + t )
= (1 t ) (1 + t ) dt + g(x) dt ,
1 x 1t xt
(4
)
then the Cauchy singular term is separated into two parts, where the integral in the second part
may be performed in closed-form. According to the known result (see [8]),
1 (1
b t ) (1 + t ) dt = b(1 x) (1 + x) cot( ) + b
(x),
1 tx
where
|x| < 1, > 1, > 1, = 0, 1, 2, . . .
and (5)
=
( 1) ; ; 2
(x) ( + 1) () , 1 1 1 x , (6)
+ + 2+ F
is the Gamma function and F the hypergeometric function. In special cases, (x) may be
further simplified as follows
= 0, + = 1,
+ =
(x) csc( ), 0, (7)
csc( )(2 1 x), += 1.
Downloaded by [UQ Library] at 07:28 18 June 2013
Downloaded by [UQ Library] at 07:28 18 June 2013
0
Proof (a) We first prove that 1 w(t )k(x, t ) dt is convergent.
0
Taking h(x, ) = 1+ w(t )k(x, t ) dt , then we only need to prove h(x, 0+) exists.
2080 Z. Chen et al.
1 + 2 1
+2
1+1
M |w(t )| dt 0 as 1, 2 0+,
1+2
then h(x, 0 +) exists. Similarly, we can obtain that 1
1 0 w(t )k(x, t ) dt is convergent. So
1 w(t )k(x, t ) dt is convergent.
1
(b) Put p(x) = 1 w(t )k(x, t ) dt . From (a), p(x) exists. Now let us show p(x) is continuous
on [1, 1].
Since k(x, t ) C([1, 1] [1, 1]), fix x [1, 1], take any > 0, there exists a > 0 such
that
|k(x, t ) k(x1, t )| <
Downloaded by [UQ Library] at 07:28 18 June 2013
3. Several lemmas
x
(t ) g
1 1
(15)
1
w(t )g1(t ) dt = 0, 1 < x < 1, (16)
1
1
where f (x) = (/b)f (x) c (x) (/b)c w(t )K (x, t ) dt .
1 1
(Ag1)(x) = (x)g1(x) +
1 w(t )
1(t t
dt
x1
g ) g (x)
1 1
+b 1 < x < 1, (17)
w(t )K (x, t )g1 (t ) d t ,
1 1
(Bg1)(x) =
w(t )g1(t ) dt , 1 < x < 1. (18)
1
International Journal of Computer Mathematics 2081
Remark 3.1 Since Equation (1) with Equation (2) has a unique solution, then Equation (11)
with Equation (2) has a unique solution. Hence, Equation (15) with Equation (16), namely,
Ag1 = f1,
(19
Bg1 = 0, )
and let u0 be a special solution of Equation (19), then all solutions of Equation (19) can be
denoted by u0 + N (A) N (B). Note that Equation (19) has a unique solution, we obtain N (A)
N (B) = {0}. From the solution set of equations
Ag1 = 0,
(20
)
2013
Bg1 = 0,
can be denoted by N (A) N (B), Equation (20) has a unique solution.
Downloaded by [UQ Library] at 07:28 18 June
Lemma 3.1 Linear operators A and B are maps from W1[1, 1] to C[1, 1], respectively.
Proof Obviously, B maps W1[1, 1] into C[1, 1]. Next we prove A maps W1[1, 1] into C[1,
1].
Take any g1(x) W1[1, 1], From (g1(t ) g1(x))/(t x), K (x, t )g1(t ) C([1, 1]
[1, 1]) and Theorem 2.3, it follows that the integrals in operator A are all convergent and
belong to C[1, 1]. Thus A maps W1[1, 1] into C[1, 1].
Let {xi }i =1 be a dense subset of interval [1, 1]. Taking
Theorem 3.1 (a) 1i (x) W1[1, 1]; (b) 21(x) W1[1, 1].
Proof (a) Since
1
Rx (t ) R (x )
1i (x) = [Ay Rx (y)](xi ) = (xi )Rx (xi ) + w(t ) t
x
x i
dt
1 i
+ 1
w(t )K (xi , t )Rx (t ) dt ,
1
= I + II + III.
Obviously, I C[1, 1]. In terms of Theorem 2.3 and K (xi , t ), Rx (t ), (Rx (t ) Rx (xi ))/(t
xi ) C([1, 1] [1, 1]), we obtain II, III C[1, 1]. Hence, i (x) C[1, 1]. It follows
that 1i (x) U [1, 1] from Lemma 3.2. Therefore, by Definition 2.1, 1i (x) W1[1, 1].
2082 Z. Chen et al.
Define
k = 1, 2, . . . . (21)
Downloaded by [UQ Library] at 07:28 18 June 2013
Theorem 4.1 {ri }i =1 is complete in W1[1, 1], where {ri }i =1 given by Equation (21).
Proof By Theorem 3.1, ri (x) W1[1, 1]. Let u(x) W1[1, 1] such that (u, ri ) = 0. So
(u, 1i ) = 0, (u, 21) = 0. Noting that
0 = (u, 1i ) = (u(x), Ay Rx (y)(xi )) = [Ay (u(x), Rx (y))](xi ) = (Ay u(y))(xi ), (22)
then Au(xi ) = 0. From the density of {xi } i=1 in [1, 1] and Lemma 3.1, Au = 0. Similarly, it
follows that Bu = 0 from (u, 21) = 0. From Remark 3.1, we know that
Au = 0,
Bu = 0,
Define the normal orthogonal system {r i } i=1 in W1[1, 1] which derives from Gram-
Schmidt orthogonalization process of {ri } i=1,
i
ri = ik rk , ii > 0, i = 1, 2, . . . ,
k=1
{ri } i=1 is an orthonormal basis of W1[1, 1].
Proof (a) Let g1(x) be the exact solution of Equation (15) with Equation (16). Since
and
(g1, r1) = 11(g1, r1) = 110 = 0 = f 1
and
Downloaded by [UQ Library] at 07:28 18 June 2013
1, ri ) =
i 1i 1 k1 = i
k 1 ik k =k ik
(g (g , ) f (x )f as i > 1.
2
r
= =
=i=1 =i=1
(b) Since i i + L/N i i + L/N .
Hence, g(x) (g, r)r fr
1
2
||Rx (y)||W = (Rx (y), Rx (y)) = Rx (x),
2 x R 2
x Rx (y) x1 (y), x Rx2 (y)
Rx1 (x2) ,
= x
x =
x x
x
W1 1 2 x1 2 1 2 x1 2 x
= = = =
2 4
2
Rx (y) 2
R
2 Rx2 (y) Rx1 (x2)
x2 W1
=
x1
2 x1 (y), x2
2
x1
=
x
2
=
x
=
x1 x2
2 2
x1
=
x
2
=
x
|gn(x) g(x)| = |(gn(y) g(y), Rx (y))| ||gn g|| ||Rx (y)||W2 M ||gn g||,
2 2
| gn (x) g (x) | = (gn(y) g(y), g g R (y) M || gn g || .
x
2
Rx (y)) ||
n
|| x
2
x
W2
5. Numerical examples
In this section, three examples will be given to show that our method is effective for the cases that
the exact solution g(x)
n1
of Equation (11) is a polynomial, a non-polynomial and g(x) is concussive.
Set {x = 1 + (2i/n 1)} .
i i=0
2084 Z. Chen et al.
Example 5.1 Considering the integral equation [8] given by
(t ) dt = L, (24)
1
where is the coefficient of friction, the Poissons ratio and L is related to the external load.
The physics of this problem corresponds to a rigid rough punch with sharp corners sliding on a
half-plane.
(1
2)
We can verify that the analytical solution to the bounded unknown function of Equation (23)
with Equation (24) is
g(x) = sin( ) + x3 + 2( + 1)x + 3 (42 + 6 + 2) . (26)
L
Numerical calculation was performed for a = 0.34, b = 0.66 and L = / sin( ). And in this
computation, 15-point GaussJacobi quadrature rule is used. Write gn(x) for the approximate
solution of g(x). And the absolute error g300 g is given in Figure 1.
Solution Here, = 1/3, = 2/3 and the analytical solution to the bounded unknown func-
tion of Equation (27) with Equation (28) is g(x) = cos x. And 15-point GaussJacobi quadrature
Figure 1. The figure of absolute error g300 g, left side for Example 5.1 and right side for Example 5.2.
International Journal of Computer Mathematics 2085
Figure 2. The figure of absolute error g300 g for Example 5.3 and left side for g(x) = cos 10x, middle for g(x) = cos
20x, right side for g(x) = cos 40x.
rule is employed in this computation. Write gn(x) for the approximate solution of g(x). And the
absolute error g300 g is also given in Figure 1.
Example 5.3 Consider the following equation
1/3 cos
(x) + t x dx = (1 t ) (1 + t )2/3 t x dt (29)
1 (x) mt cos mx
Downloaded by [UQ Library] at 07:28 18 June 2013
3 3
1
1 1
subject to
1 1 1/3 2/3 (30)
(t ) dt = (1 t ) (1 + t ) cos mt dt
1 1
Solution Here, = 1/3, = 2/3. Corresponding to m = 10, 20, 40, the analytical solu-tions to
the bounded unknown function of Equation (29) with Equation (30) are g(x) = cos 10x, cos
20x and cos 40x, respectively. And 20-point GaussJacobi quadrature rule is employed in this
computation. Write gn(x) for the approximate solution of g(x). And the absolute errors g300 g
for m = 10, 20, 40 are given in Figure 2.
Compared with [8]: Our method is not so efficient as [8] when the bounded function g(x) is
a polynomial. But generally speaking, g(x) is not a polynomial. And we know from Figures 1
and 2 that our method is still effective when g(x) is not a polynomial or g(x) is concussive.
6. Conclusion
In this paper, by using a transform, the singularity of Equation (1) is removed. Subsequently,
by improving the traditional reproducing kernel method of solving linear operator equation Au
= f , we give the exact solution, denoted by series, of Equation (1) with Equation (2) in
reproducing kernel space. After truncating the series, the approximate solution is obtained. Our
method is very simple. The approximate solution and its first derivative converge uniformly to
the exact solution of Equation (1) with Equation (2) and its first derivative on interval [1, 1].
The final three examples show the efficiency of our method.
References
[1] J.Y. Du, Singular integral operators and singular quadrature operators associated with singular integral
equations of the first kind and their applications, Acta Math. Sci. 15 (1995), pp. 219234.
[2] J.Y. Du, Singular integral operators and singular quadrature operators associated with singular integral
equations, Acta Math. Sci. 18 (1998), pp. 227240.
2086 Z. Chen et al.
[3] J.Y. Du, The collocation methods for singular integral equations with Cauchy kernels, Acta Math. Sci. 20 (2000),
pp. 289302.
[4] Z.H. Du and J.Y. Du, On direct method of solution for a class of singular integral equations, Acta Math. Sci. 27
(2007), pp. 301307.
[5] F. Erdogan and G.D. Gupta, On the numerical solution of singular integral equations, Q. Appl. Math. 30 (1972),
pp. 525534.
[6] D.A. Hills and D. Nowell, Mechanics of Elastic Contacts, Butterworth-Heinemann, Oxford, 1993.
[7] D.A. Hills, P.A. Kelly, D.N. Dai, and A.M. Korsunsky, Solution of Crack Problems: The Distributed Dislocation
Technique, Kluwer Academic Publishers, Dordrecht, 1996.
[8] X.Q. Jin, L.M. Keer, and Q. Wang, A practical method for singular integral equations of the second kind, Eng.
Frac. Mech. 75 (2008), pp. 10051014.
[9] P. Karczmarek, D. Pylak, and M.A. Sheshko, Application of Jacobi polynomials to approximate solution of a
singular integral equation with Cauchy kernel, Appl. Math. Comput. 181 (2006), pp. 694707.
[10] S. Krenk, On quadrature formulas for singular integral equations of the first kind and the second kind, Q. Appl.
Math. 33 (1975), pp. 225232.
[11] S. Krenk, Quadrature formulas of closed type for solution of singular integral equations, J. Inst. Maths. Appl. 22
(1978), pp. 99107.
[12] N. Mohankumar and A. Natarajan, On the numerical solution of Cauchy singular integral equations in neutron
transport, Ann. Nuclear Energy, 35 (2008), pp. 18001804.
[13] N.I. Muskhelishvili, Singular Integral Equations, Noordhoff, Groningen, 1953.
Downloaded by [UQ Library] at 07:28 18 June 2013
[14] P.S. Theocaris and N.L. Ioakimidis, Numerical integration methods for solution of singular integral equations, Q.
Appl. Math. 35 (1977), pp. 173182.
Appendix
Proof of Theorem 2.1 Using formula of integration by parts three times, one has
1 (4) 1 (5) 1 1
1uv | 1 + uv | 1 (6)
(A1)
u v dx = u v(3) |1 uv dx,
further 1 1
(u, v) = 2
u(i) (1)v(i) (1) +
1
1
u v dx
i= 0
(5) (5) (4)
= u(1)[v(1) v (1)] + u(1)v (1) + u (1)[v (1) + v (1)]
(4) (3) (3) 1 (6)
u (1)v (1) + u (1)[v (1) v (1)] + u (1)v (1) uv dx,
thus 1
5 5
4
R(x, 1) R(x, 1) +
(u(y), R(x, y))y = u(1) 5 R(x, 1) + u(1) 5 R(x, 1) + u (1) 4 R(x, 1)
y y y
y
4 2 3 3
u (1) 4 R(x, 1) + u (1) 2 R(x, 1) 3 R(x, 1) + u (1) 3 R(x, 1)
y y y y
1 6
u(y) 6 R(x, y) dy.
1 y
In order to (u(y), R(x, y))y = u(x), it is enough to request the following equalities hold
6 R(x, y) = (y x), (A2)
6
y
R(x, 1) 5 5
5 R(x, 1) = 0, 5 R(x, 1) = 0, (A3)
y y
4 4
1) +
R(x, 4 R(x, 1) = 0, 4 R(x, 1) = 0, (A4)
y y y
2 3
2 R(x, 1) 3 R(x, 1) = 0, (A5)
y y
3
3 R(x, 1) = 0. (A6)
y
International Journal of Computer Mathematics 2087
6 6 6
From Equation (A2), we have ( /y )R(x, y) = 0 as y = x. Its characteristic equation is = 0. Hence, = 0 (six-fold),
we obtain
R(x, y) = 2 3
c0(x) + c1(x)y + c2(x)y + c3(x)y + c4(x)y + c5(x)y ,
4 5 y x, (A7)
2 3 4 5
+ d3 (x)y + d4 (x)y + d5 (x)y , y > x.
d0(x) + d1(x)y + d2(x)y
Integrating on both sides of Equation (A2) from x to x + with respect to y and let 0, one gets
i
i
i R(x, x + 0) = i R(x, x 0), i = 0, 1, 2, 3, 4, (A8)
y y
5 5
5 R(x, x 0) 5 R(x, x + 0) = 1. (A9)
y y
Using software mathematica 5.0, we can solve ci , di from Equations (A3)(A6), (A8), (A9). Substituting the results into
Equation (A7), R(x, y) is obtained.
Downloaded by [UQ Library] at 07:28 18 June 2013