Jump to ContentJump to Main Navigation
Show Summary Details
More options …

Journal of Applied Analysis

Editor-in-Chief: Fechner, Włodzimierz / Ciesielski, Krzysztof

Managing Editor: Gajek, Leslaw

CiteScore 2018: 0.45

SCImago Journal Rank (SJR) 2018: 0.181
Source Normalized Impact per Paper (SNIP) 2018: 0.845

Mathematical Citation Quotient (MCQ) 2018: 0.20

See all formats and pricing
More options …
Volume 25, Issue 1


Two-step collocation methods for two-dimensional Volterra integral equations of the second kind

Seyed Mousa Torabi / Abolfazl Tari / Sedaghat Shahmorad
Published Online: 2019-05-21 | DOI: https://doi.org/10.1515/jaa-2019-0001


In this paper, we develop two-step collocation (2-SC) methods to solve two-dimensional nonlinear Volterra integral equations (2D-NVIEs) of the second kind. Here we convert a 2D-NVIE of the second kind to a one-dimensional case, and then we solve the resulting equation numerically by two-step collocation methods. We also study the convergence and stability analysis of the method. At the end, the accuracy and efficiency of the method is verified by solving two test equations which are stiff. In examples, we use the well-known differential transform method to obtain starting values.

Keywords: Two-dimensional nonlinear Volterra integral equations; integral equations of the second kind; two-step collocation methods

MSC 2010: 65R20

1 Introduction

Many problems in applied mathematics, physics and engineering give rise to the nonlinear two-dimensional Volterra integral equation of the form


where g and K are given, sufficiently smooth functions on D:-[0,X]×[0,T] and D×D×R, respectively.

A numerical solution of equations of form (1.1) has been considered in some works. For example, in [13], the block-by-block method has been considered. In [2, 10], collocation and iterated collocation methods have been proposed for two-dimensional nonlinear VIEs (2D-VIEs). In [17], the differential transform method has been developed for linear and nonlinear 2D-VIEs. A new block-by-block method has been presented for these equations in [12]. Also, the Galerkin method has been developed for two-dimensional VIEs in some works. For example, in [11], the extrapolation method based on the asymptotic expansion of iterated Galerkin solutions has been studied for 2D-VIEs of the second kind. In [15], the spectral Galerkin method has been proposed for numerical solution of 2D-VIEs of the second kind. On the other hand, many studies have been made in the numerical solution of one-dimensional Volterra integral equations. For example, in [14], the differential transform method has been considered for VIEs. Recently, many new methods have been presented to solve some types of differential and integral equations [1, 8, 9]. Also, multi-step collocation methods have been proposed for one-dimensional VIEs of the second kind in some interesting works [6, 5, 4]. In this paper, we develop these methods for 2D-VIEs of the second kind. The stability is the main advantage of these methods compared to the majority of available numerical methods. Therefore, the presented method can be applied to stiff equations, which are defined as follows.

Definition 1.1.

Integral equation (1.1) is said to be “stiff” in cases where K(x,t,z,s,y)/y assumes a large negative value [18].

2 Two-step collocation methods

As mentioned above, in this paper, we develop the 2-SC method of [4], to equations of form (1.1). So here, we present the method of [4], for the sake of the reader. Consider the VIE


where g and K are real-valued sufficiently smooth functions. For a given positive integer N, we set tn=t0+nh, n=0,1,,N, with Nh=T-t0. First we rewrite equation (2.1) in the form


with the lag-term


and the increment term


The 2-SC method provides a continuous approximation P(tn+sh), s[0,1], to the solution y(tn+sh) of (1.1) in the interval [tn,tn+1], which uses the information of the equation on the following two consecutive steps:


Here tn,j=tn+cjh are collocation points, and cj are collocation parameters, Fh[n] and Φh[n+1] are approximations to F[n] and Φ[n+1], which are computed by appropriate quadrature rules as


φ0, φ1, χj and ψj, j=1,,m, are polynomials such that P(t) be a continuous approximation to the solution y(t) of (2.1) at each subinterval [tn,tn+1]. The polynomial P(tn+sh) will be determined after solving a system of equations in the values Yi[n+1]:-P(tn,i) and yn+1, at each step. For more details, see [4].

To discuss the order of the method, we recall the following theorem.

Theorem 2.1 ([4]).

Assume that, in (2.1), K and g are sufficiently smooth functions. If the polynomials φ0(s), φ1(s), χj(s) and ψj(s), j=1,,m satisfy the system of equations


for s[0,1] and k=1,2,,p, then method (2.2) has the local discretization error of order p, i.e.,




We choose φ0(s) and φ1(s) as the polynomials of degree at most 2m-1, which satisfy the collocation conditions, that is,


Thus we have


where q0,,qm-1 and p0,,pm-1 are free parameters.

Here we give two lemmas which we need in the following.

Lemma 2.2 (Gronwall inequality [16]).

Let {yn} and {gn} be nonnegative sequences and C a nonnegative constant. If




Lemma 2.3 ([3]).

The determinant of the Vandermonde matrix of the form


is det(V)=0j<in(xi-xj).

In the following theorem, we prove that system (2.5) has a unique solution.

Theorem 2.4.

Assume that cicj, cicj-1 and ci-1,0,1. Then, choosing φ0(s) and φ1(s) as (2.7) and (2.8), respectively, the system (2.5) has a unique solution, and



Setting s=ci in (2.5) for i=1,,m and from collocation conditions (2.6), we obtain


Therefore, the coefficient matrix is of the form


which is of Vandermonde type.

By Lemma 2.3, det(A)=1j<im(ci-cj)2(1-(ci-cj)2) for m2 (for m=1, det(A)=1). Thus, by the assumptions of the theorem, det(A)0. On the other hand, it is obvious from (2.9) that


So the theorem is proved. ∎

The next theorem investigates the order of convergence for the method (2.2).

Theorem 2.5 ([4]).

Let eh(t):-y(t)-P(t) be the error of method (2.2). Suppose that the hypothesis of Theorem 2.1 are satisfied for p=2m-1 with φ0 and φ1 as (2.7) and (2.8), respectively. Moreover, assume that

  • (1)

    Ky(t,η,) exists and is bounded for t0ηtT,

  • (2)

    the quadrature formulas ( 2.3 ) and ( 2.4 ) are of order O(hq),

  • (3)

    the starting error is eh,[t0,t1]=O(hd).

Then the two-step collocation method (2.2) has the uniform order of convergence p*=min{2m,q,d+1} for any choice of 0<c1<c2<<cm<1, that is,


To analyze the stability of the presented method, we apply the method to the test equation


This leads to the following matrix recurrence relation [4]:


where z=hλ and M(z) is called stability matrix. The stability function of the method is defined as


Denoting w1,w2,,w2m+2 as the roots of (2.11), the region of absolute stability of the method is defined by


Also, we say that the method is A-stable if


3 Main results

In this section, we develop the 2-SC method described in the previous section to 2D-VIEs of form (1.1). To this end, let N and M be positive integers, and consider the uniform grids


We set x=xi in (1.1), and thus we have


Now, substituting the inner integral of (3.1) by an appropriate quadrature rule depending on xj, where j=0,1,,i, we obtain


which is a one-dimensional VIE of the second kind, and we solve it by the two-step collocation method described in the previous section. In equation (3.2), yi(t), gi(t) and Kij(t,s,) denote y(xi,t), g(xi,t) and K(xi,t,xj,s,), respectively, and wij are quadrature weights. In this procedure, we use the values obtained from the previous steps.

First, setting x=x1 and using the trapezoidal rule, we have


where it is obvious that y(0,s)=g(0,s).

Therefore, applying the two-step collocation method to this equation, we obtain an approximate polynomial to y1(t)=y(x1,t), namely, P1(t).

For x=x2, we use Simpson’s rule for the interior integral in (3.1). Thus we obtain


where y(0,s) and y(x1,s) are known, and therefore we obtain P2(t), the approximate polynomial of y2(t) using the two-step collocation method.

For x=xi, i=3,4,,N, we use Simpson’s rule for even indices and Simpson’s rule with the trapezoidal rule for the last subinterval with odd indices. From [7], this method (Simpson and trapezoidal) is stable.

To simplify the notation, we set


To analyze the error of the presented method, we assume that the maximum error occurs at the ith stage, that is, for x=xi. At the ith stage, we have equation (3.1), and substituting the inner integral by a quadrature rule of order, for example, r, we have


with AiC1 independent of k.

The next theorem investigates the error of the presented method.

Theorem 3.1.

Let ei,h:-yi(t)-Pi(t) be the error of the new method at stage i. Suppose that the hypotheses of Theorem 2.1 are satisfied for p=2m-1 for the ith stage, with φ0(s) and φ1(s) chosen according to (2.7) and (2.8), respectively. Moreover, assume that

  • (i)

    yKi(t,s,) exists and is bounded for 0stT,

  • (ii)

    the quadrature formulas ( 2.3 ) and ( 2.4 ) at the i th stage are of order O(hq),

  • (iii)

    the quadrature formula ( 3.2 ) used for the i th stage is of order O(kr),

  • (iv)

    the starting error is ei,h,[0,t1]=O(hd).

Then the order of convergence of the method is O(hp*+kr), where p*=min{d+1,q,2m}.


From the previous section, the approximate polynomial for yi(t) at the ith stage is


Since the functions φ0(s), φ1(s), χj(s) and ψj(s) satisfy the collocation conditions, setting Yi,j[n+1]:-Pi(tn,j), we have


Hence polynomial (3.4) is of the form


It follows from Theorem 2.1 and equation (3.3) that


with Ri,m,nC2 independent of h. Thus, subtracting (3.6) from (3.5), we obtain


where ei,n+1,j=ei,h(tn,j) and ei,n=ei,h(tn). On the other hand, applying the mean value theorem, hypothesis (i) ensures that


where zi,v-1(s) is between yi(tv-1+sh) and Pi(tv-1+sh).

Now by hypothesis (ii) it follows that


with Ei,m,nC3 independent of h.

Also, from


it follows that


From hypothesis (i),


with VC4 independent of h.

Substituting expressions (3.7) and (3.9) in equation (3.8), we obtain


On the other hand, setting s=1 in (3.7), we have


Therefore, denoting


we obtain


where the matrices Bi[n+1], Bn[v], w[n+1], wn[v] and the vectors ρn[v], ρ[n+1], 𝐒n involve the integrals over [0,cj] or [0,1] of yKi multiplied by φ0, φ1, χj, ψj, Ri,m,n, Vi and Ai,m,n.



Then, from (3.10) and (3.11), it follows that


where D1,D2,γ1,γ2,γ3,γ4 are upper bounds of the norm of vectors and matrices appearing in (3.11).

Hence, using the Gronwall inequality, it follows that


where γ=max{γ1,,γ4} and C is a constant independent of h and k. ∎

To analyze the stability of method, we define the following notations for the ith stage of the method:


where bi,js and wi,j,ls are the weights of quadrature rules of the ith stage for Fi,h[n] and Φi,h[n+1], respectively.

The following theorem can be obtained analogously to the theorem of [4] for each stage of the presented method.

Theorem 3.2.

Applying the presented method to test equation (2.10) for each stage of the method, leads to the matrix recurrence relation


where z=hλ and the stability matrix Mi(z) is Mi(z)=Pi-1(z)Qi(z) with


Therefore, defining


we obtain the matrix recurrence relation of the method in the general case as Vn+1=M(z)Vn, where the stability matrix M(z) is M(z)=diag(M1(z),,MN(z)).

4 Numerical examples

In this section, we give some examples to show the accuracy and stability of the presented method. In the examples, we apply the method with m=2, c1=35, c2=45 and q0=p0=1, q1=p1=-1 (for (2.7), (2.8)). So we have p=2m-1=3 and the polynomials φ0, φ1, χj and ψj, j=1,2, as


Also, the stability polynomial is




From [4], the described method is A-stable by these choices of parameters and polynomials.

As mentioned previously, in this paper, we apply the well-known differential transform (DT) method [17] to obtain the required starting values. This method gives an approximation to the Taylor expansion at (x0,t0) of the solution, which has high accuracy near the point (x0,t0). Therefore, it is suitable to obtain the starting values.

Example 4.1.

Consider the integral equation


which is a stiff equation with the exact solution y(x,t)=x1+t. Applying the two-dimensional DT method of [17] to equation (4.1), we obtain


which is a recurrence relation with Y(m,0)=Y(0,n)=0 for m,n=0,1,2,, and G(m,n) is the differential transform of g(x,t). Therefore, the DT approximate solution of equation (4.1) is given by


We use relation (4.2) to determine the required starting values. Table 1 shows the absolute errors of the presented method and the DT method at some points.

Table 1

Computational results of Example 4.1 at some nodes.

Example 4.2.

As the second example, consider the stiff equation


with g(x,t)=xLn(1+t)+1003x3[12(t2-1)Ln(1+t)+12t-14t2] and the exact solution y(x,t)=xLn(1+t).

Similar to the previous example, using the two-dimensional DT method, we obtain


where G(m,n) is the differential transform of g(x,t) and Y(m,0)=Y(0,n)=0, m,n=0,1,2,. Therefore, the approximate solution of equation (4.3) is given by


which gives us the required starting values. The absolute errors of presented method and DT method are given in Table 2.

Table 2

Computational results of Example 4.2 at some nodes.

5 Conclusion

In this paper, we extended the two-step collocation methods for two-dimensional nonlinear Volterra integral equations (2D-NVIEs) of the second kind. We converted the 2D-NVIE of the second kind to a one-dimensional VIE of the second kind, and then we solved the resulting equations using two-step collocation methods. The numerical results confirm the convergence and stability of the method.


The authors would like to thank the anonymous referees for their valuable comments that helped the authors to improve the paper.


  • [1]

    N. Bildik and S. Deniz, A new efficient method for solving delay differential equations and a comparison with other methods, Europ. Phys. J. Plus. 132 (2017), no. 1, 51–61.  CrossrefGoogle Scholar

  • [2]

    H. Brunner and J.-P. Kauthen, The numerical solution of two-dimensional Volterra integral equations by collocation and iterated collocation, IMA J. Numer. Anal. 9 (1989), no. 1, 47–59.  CrossrefGoogle Scholar

  • [3]

    N. L. Carothers, A Short Course on Banach Space Theory, London Math. Soc. Stud. Texts 64, Cambridge University Press, Cambridge, 2005.  Google Scholar

  • [4]

    D. Conte, Z. Jackiewicz and B. Paternoster, Two-step almost collocation methods for Volterra integral equations, Appl. Math. Comput. 204 (2008), no. 2, 839–853.  Web of ScienceGoogle Scholar

  • [5]

    D. Conte and B. Paternoster, Multistep collocation methods for Volterra integral equations, Appl. Numer. Math. 59 (2009), no. 8, 1721–1736.  Web of ScienceCrossrefGoogle Scholar

  • [6]

    D. Conte and I. D. Prete, Fast collocation methods for Volterra integral equations of convolution type, J. Comput. Appl. Math. 196 (2006), no. 2, 652–663.  CrossrefGoogle Scholar

  • [7]

    L. M. Delves and J. L. Mohamed, Computational Methods for Integral Equations, Cambridge University Press, Cambridge, 1985.  Google Scholar

  • [8]

    S. Deniz, Comparison of solutions of systems of delay differential equations using Taylor collocation method, Lambert W function and variational iteration method, Sci. Iranica. Trans. D Comp. Sci. Engin. Elec. 22 (2015), no. 3, 1052–1058.  Google Scholar

  • [9]

    S. Deniz and N. Bildik, A new analytical technique for solving Lane–Emden type equations arising in astrophysics, Bull. Belg. Math. Soc. Simon Stevin 24 (2017), no. 2, 305–320.  Google Scholar

  • [10]

    H. Guoqiang and Z. Liqing, Asymptotic error expansion of two-dimensional Volterra integral equation by iterated collocation, Appl. Math. Comput. 61 (1994), no. 2–3, 269–285.  Google Scholar

  • [11]

    G. Han and R. Wang, The extrapolation method for two-dimensional Volterra integral equations based on the asymptotic expansion of iterated Galerkin solutions, J. Integral Equations Appl. 13 (2001), no. 1, 15–34.  CrossrefGoogle Scholar

  • [12]

    R. Katani and S. Shahmorad, A new block by block method for solving two-dimensional linear and nonlinear Volterra integral equations of the first and second kinds, Bull. Iranian Math. Soc. 39 (2013), no. 4, 707–724.  Google Scholar

  • [13]

    F. Mirzaee and Z. Rafei, The block by block method for the numerical solution of the nonlinear two-dimensional Volterra integral equations, J. King Saud Uni. Sci. 23 (2011), no. 2, 191–195.  CrossrefGoogle Scholar

  • [14]

    Z. M. Odibat, Differential transform method for solving Volterra integral equation with separable kernels, Math. Comput. Modelling 48 (2008), no. 7–8, 1144–1149.  CrossrefWeb of ScienceGoogle Scholar

  • [15]

    J. Saberi, O. Navid Samadi and E. Tohidi, Numerical solution of two-dimensional Volterra integral equations by spectral Galerkin method, J. Appl. Math. Bioinf. 1 (2011), 159–174.  Google Scholar

  • [16]

    L. Tao and H. Yong, A generalization of discrete Gronwall inequality and its application to weakly singular Volterra integral equation of the second kind, J. Math. Anal. Appl. 282 (2003), no. 1, 56–62.  CrossrefGoogle Scholar

  • [17]

    A. Tari, M. Y. Rahimi, S. Shahmorad and F. Talati, Solving a class of two-dimensional linear and nonlinear Volterra integral equations by the differential transform method, J. Comput. Appl. Math. 228 (2009), no. 1, 70–76.  Web of ScienceCrossrefGoogle Scholar

  • [18]

    P. J. van der Houwen and H. J. J. te Riele, Backward differentiation type formulas for Volterra integral equations of the second kind, Numer. Math. 37 (1981), no. 2, 205–217.  CrossrefGoogle Scholar

About the article

Received: 2018-01-03

Revised: 2018-02-27

Accepted: 2018-04-12

Published Online: 2019-05-21

Published in Print: 2019-06-01

Citation Information: Journal of Applied Analysis, Volume 25, Issue 1, Pages 1–11, ISSN (Online) 1869-6082, ISSN (Print) 1425-6908, DOI: https://doi.org/10.1515/jaa-2019-0001.

Export Citation

© 2019 Walter de Gruyter GmbH, Berlin/Boston.Get Permission

Comments (0)

Please log in or register to comment.
Log in