1. Introduction
The numerical integration of initial value problems (IVPs) of second order ordinary differential equations (ODEs) has attracted the attention of researchers in the field for decades. The importance of such problems is their use in the applied sciences to model different phenomena such as the mass movement under the action of a force, problems of orbital dynamics, molecular dynamics, circuit theory, control theory, or quantum mechanics among others. It turns out that most of these problems do not have closed-form solutions, and consequently it is important to develop numerical methods that can solve them directly. Accordingly, several numerical methods for solving second-order IVPs that do not contain the first derivative have been investigated by Lambert and Watson [
1], Ananthakrishnaiah [
2], Simos [
3], Hairer et al. [
4], Tsitouras et al. ([
5,
6]), Wang et al. ([
7,
8]), Franco ([
9,
10]), Ramos and Patricio [
11], Chen et al. [
12], Shi and Wu [
13], Fang et al. [
14], Senu et al. [
15] among others.
Recently, some researchers have considered the direct integration of the general second order IVPs containing the first derivative. This can be found in the works by Guo and Yan [
16], Vigo-Aguiar and Ramos [
17], Jator et al. ([
18,
19]), Mahmoud and Osman [
20], Awoyemi [
21], Liu and Wu [
22], You et al. [
23], Li et al. [
24], Chen et al. [
25], Li et al. [
26], and You et al. [
27]. The implementation of some of these methods is based on a step-by-step fashion, while others are implemented in predictor-corrector modes. In either case, the cost of execution increases, especially, for higher-order methods. It turns out that some of these methods do not take advantage of the oscillatory or even periodic behavior of the solutions. If the period is known or can be estimated in advance this could be considered in the development of the method in order to improve its performance.
One of the numerical integrators for the general second order IVP in which the first derivative appears explicitly is an explicit method due to Falkner [
28], while the implicit form is due to Collatz [
29]. Some modifications of the Falkner methods have appeared in the literature (see [
30,
31,
32,
33]). The adapted Falkner methods take advantage of the special periodic feature of the solution of the IVP, and can be found in the works by Li and Wu [
34], Li [
35], and Ehigie and Okunuga [
36]. The use of adapted methods started with the elegant work by Gautchi [
37] and later by Lyche [
38]. Many extensions of these have been investigated by Franco ([
39,
40]), Ixaru et al. [
41], Vanden Berghe and Van Daele [
42], Jator et al. [
43], Jator ([
18,
44]), Ramos and Vigo-Aguiar [
45], Vigo-Aguiar and Ramos ([
46,
47]), Coleman and Duxbury [
48], Coleman and Ixaru [
49], Nguyen et al. [
50], Ozawa [
51], Fang et al. [
14], Franco and Gomez [
52], Wua and Tian [
53]. Nonetheless, most of these methods have been implemented in a step-by-step procedure. Also, in all these extensions, the basis function considered is either the set
or
.
In the current article, we propose a class of Functionally-Fitted third derivative Block Falkner Methods (BFFM) for the direct integration of the general second order initial value problem whose solution is oscillatory or periodic, in the latter case with the frequency known, or that can be estimated in advance. This class, which is an adapted formulation of the methods in Ramos and Rufai [
33], uses a basis function different from what can be seen in the reviewed literature. We emphasize that this method is different from the methods by Jator ([
18,
44]); whereas our methods are implicit Falkner methods whose coefficients are trigonometric and hyperbolic functions depending on the fitting frequency,
, and the step size,
h, the methods by Jator ([
18,
44]) present trigonometric coefficients. It is important to note here that the accuracy in estimating this frequency is crucial in adapted numerical methods, as shown in [
45].
The rest of this paper is organized as follows: the derivation of BFFM is presented in
Section 2. The analysis of the characteristics of the BFFM is discussed in
Section 3 while some numerical experiments are presented in
Section 4. Finally, we give some concluding remarks in
Section 5.
2. Development of the BFFM
Consider the general second order IVP of the form
whose solution is oscillatory or periodic with the frequency approximately known in advance and
is a smooth function that satisfies a Lipschitz condition and
s is the dimension of the system. For the development of the method,
is taken as a scalar function although as we will see in the numerical experiments, the method may be applied in a component wise mode to solve differential systems. We now set-out some useful definitions related with the methods in Ramos and Rufai [
33], that will aid the derivation of the BFFM.
Definition 1. The continuous formulation of the adapted k—step third derivative Falkner method for approximating the solution of Equation (1) is defined bywhere the coefficients , , and are functions of x and , being the frequency of the method (see [45]), and . Definition 2. The primary formulas of the adapted k—step third derivative block Falkner method for the numerical solution of Equation (1) are given by In these formulas, , , and are numerical approximations to the exact values , , and , respectively, with discrete points on being h a fixed stepsize. The coefficients and depend on the parameter u and are obtained after evaluating the fitting function in Theorem 1 and its derivative, respectively, at .
Definition 3. The secondary formulas of the adapted k—step third derivative block Falkner method for the numerical solution of Equation (1) are given bywhere . Again, the coefficients and depend on u, and are obtained after evaluating the fitting function and its derivative at .
Definition 4. The adapted k—step third derivative block Falkner method consists of the primary formulas in (3) and the secondary formulas in (4), which form the BFFM. 2.1. Derivation of the BFFM
Let
be a set of
linearly independent functions. We seek an approximate solution
, called a fitted function associated to the adapted Falkner method, which satisfies the IVP in Equation (
1) at some specified points.
The coefficients of the adapted Falkner method depend on the nature of the fitting function
accordingly on how the set
is chosen, which can be any of the types listed in Nguyen et al. [
50], where for any of the choices we have to take a total of
elements to determine the adapted block Falkner method on the basis that the approximations are of the form in (
2). In order to develop the adapted Falkner methods in this paper we choose
as
To get the coefficients of the fitting function associated to the set
in (
5),
is interpolated at the point
, and the following collocating conditions are considered:
at
,
at the points
, and
at
. This leads to the following system of
equations
Theorem 1. Let be the fitting function associated to the set Ω in (5),and the vector , where T denotes the transpose. Consider the following square matrix of dimension which is the matrix of coefficients of the system in (6),and obtained by replacing the i-th column of Π by the vector Λ. If we impose that satisfies the system of equations in (6), then it can be written as Proof. The proof can be readily obtained, similarly to the one given in Jator [
18] with slight modifications in notations. □
Remark 1. As an illustration of the theoretical result in the above theorem, the explicit form of the matrix Π and are provided in the Appendix A for . 2.2. Specification of the BFFM
We emphasize that for each
k, there are two primary formulas of the form in Equation (
3) and
secondary formulas as those in Equation (
4) (which are obtained by evaluating the fitting function in (
7) at the corresponding points) that combined together form the proposed BFFM. Hence the BFFM has
formulas.
As an illustration, we specified how to obtain the BFFM for and , repectively.
For
, we evaluate the fitting function in (
7) and its first derivative at
to obtain the two primary formulas and the two secondary formulas as
Whereas for
, we evaluate the fitting function in (
7) and its first derivative at
and then at
to obtain the two primary formulas and the four secondary formulas, which result in the following
Remark 2. For small values of u, the coefficients of the BFFM may be subject to heavy cancellations. In that case the Taylor series expansion of the coefficients is preferable (see Lambert, [54]). Specific coefficients of the two primary formulas and their corresponding series expansion up to for are provided in Appendix B. Remark 3. When , the formulas in Equations (8) and (9) reduce to the conventional third derivative Falkner formulas for and , respectively, in Ramos and Rufai [33]. 3. Analysis of the BFFM
We discuss the basic analysis of the proposed BFFM in this section. The analysis includes the Algebraic Order, Local Truncation Error, Consistency, Zero-Stability, Convergence and Linear Stability of the BFFM.
3.1. Algebraic Order, Local Truncation Errors and Consistency of the BFFM
The purpose of this subsection is to establish the uniform algebraic order for each of the formulas that form the BFFM and their corresponding local truncation errors with the aid of the theory of linear operators (Lambert [
54]).
3.1.1. Local Truncation Error of BFFM
Proposition 1. The local truncation error of each formula of the k—step BFFM is of the form , where is the error constant.
Proof. Since the block Falkner formulas in Equations (
3) and (
4) are made up of generalized linear multistep formulas, we associate the Falkner formulas with linear difference operators
,
for the primary formulas and
,
,
, for the secondary formulas, defined respectively by
Consider the Taylor series expansions of the right hand sides of the above formulas in powers of h. It can be shown that the first non zero term is of the form which is the local truncation error of each formula in the k-step BFFM. □
Corollary 1. The Local Truncation Errors of the BFFM for are given by Corollary 2. The Local Truncation Errors of the BFFM for are given by Corollary 3. The order p of the k—step BFFM is . Hence the order of BFFM for and are respectively and .
Theorem 2. When the solution of the problem in Equation (1) is a linear combination of the basis functions , then the local truncation errors vanish. Proof. Solving the differential equation
provides the fundamental set of solutions
which contains the basis function of the BFFM, from which the statement follows immediately. □
3.1.2. Consistency of the BFFM
Remark 4. Since the order of the k—step BFFM is , we therefore conclude that it is consistent (Lambert, [54] and Fatunla, [55]). 3.2. Stability of the BFFM
The BFFM specified by Equations (
3) and (
4) may be written as a difference system given by
where
and
,
,
,
are
matrices containing the coefficients of the formulas. For
and
those matrices are given as follows
3.2.1. Zero Stability of BFFM
Definition 5. Zero stability is concerned with the stability of the difference system in the limit as h tends to 0. Thus as , the difference system in Equation (13) becomeswhere and are constant matrices. Definition 6. (Fatunla [56]) A block method is zero-stable if the roots of the first characteristic polynomial have modulus less than or equal to one and those of modulus one do not have multiplicity greater than 2, i.e. the roots of satisfy and for those roots with , the multiplicity does not exceed 2. Proposition 2. The BFFM is zero-stable.
Proof. We normalize Equation (
14) to obtain the first characteristic equation of BFFM given by
. From our calculations, the roots
of
satisfy
, the roots are simple. Hence for each
and
the BFFM is zero-stable. □
Remark 5. We note that the explicit form of the matrix for and , respectively, are provided in Appendix C. 3.2.2. Convergence of BFFM
The necessary and sufficient condition for a method to be convergent is that it must be zero-stable and consistent (Lambert, [
54] and Fatunla, [
55]). Since BFFM (for each
k) is both zero-stable and consistent, we therefore conclude that it is convergent.
3.2.3. Linear Stability and Region of Stability of BFFM
To analyze the linear stability of BFFM, the block method in Equation (
13) is applied to the Lambert-Watson test equation
. After simple algebraic calculations and letting
, we obtain
, where
The rational function is called the amplification matrix and determines the stability of the method.
Definition 7. (Coleman and Ixaru, [49]): A region of stability is a region in the zu—plane throughout which , where is the spectral radius of . Since the stability matrix depends on two parameters
z and
u, we plot the stability regions in the (
z,
u)—plane for both
and
respectively, in
Figure 1, where the colored regions (blue and green) are the stability regions corresponding to the test problem
.
Since the Lambert-Watson test does not contain the first derivative, another usual test equation to analyze linear stability is the one given by
which has bounded solutions for
that tend to zero when
. We have plotted in
Figure 2 the corresponding stability region for the BFFM for
and
.
4. Implementation and Numerical Experiments
4.1. Implementation of BFFM
The BFFM is implemented using a written code in Maple 2016.1 enhanced by the feature for both linear nonlinear problems respectively. All numerical experiments are conducted on a Laptop with the following features
64 bit Windows 10 Pro Operating System,
Intel (R) Celeron CPU N3060 @ 1.60 GHz processor, and
4.00GB RAM memory.
The summary of how BFFM is applied to solve initial value problems (IVPs) with oscillatory solutions in a block by-block fashion is as follows:
Step 1: Choose N, to form the grid with . Note that N must be a multiple of .
Step 2: Using the difference Equation (
13),
, solve for the values of
and
simultaneously on the block sub-interval
, as
and
are known from the IVP (
1). As an illustration, we outline the procedure with
for the two first block intervals, when
and
, in
Appendix D.
Step 3: Next, for , the values of and are simultaneously obtained over the block sub-interval , as and are known from the previous block.
Step 4: The process is continued for
to obtain the numerical solution to (
1) on the sub-intervals
,
, …,
.
4.2. Numerical Examples
In order to examine the effectiveness of the BFFM derived in
Section 2, we apply specifically BFFM for
on some well known oscillatory problems that were solved in the recent literature. The criteria used in the numerical investigations are two-fold, the accuracy and the efficiency. A measure of the accuracy is investigated using the maximum error of the approximate solution defined as
, where
is the exact solution and
is the numerical solution obtained using BFFM while the computational efficiency can be observed through the plots of the maximum errors versus the number of function evaluations, NFE, required by each integrator. We emphasize that the fitting frequencies used in the numerical experiments have been obtained from the problems referenced from the literature.
4.3. Problems Where the First Derivative Appears Explicitly
4.3.1. Example 1
As our first test, we consider the following general second order IVP
with initial conditions
and
, whose analytical solution is
.
We solve this problem in the interval [0,100] with
and compare the result of BFFM with the BNM of order 5 in Jator and Oladejo [
57], BHT of order 5 and BHTRKNM of order 3 in Ngwane and Jator [
19,
58].
Table 1 shows the Maximum errors and the NFE, while the efficiency curves are presented in
Figure 3.
4.3.2. Example 2
Let us consider the following oscillatory system
with
and whose solution in closed form is given as
In our experiment, we choose the parameter value
and the fitting frequency as
. The problem is solved in the interval [0, 100]. The step sizes for the numerical experiment are taken as
. The numerical results of BFFM in comparison with the Block Falkner methods (BFM) of order 5 in Ramos et al. [
32] and Modified Block Falkner methods (MBFM) of order 5 in Ehigie and Okunuga [
36] are displayed in
Table 2 while the efficiency curves are displayed in
Figure 4 respectively.
4.3.3. Example 3
Consider the popular Van der Pol equation given by
with initial values
This is a nonlinear scalar equation. In our numerical experiment, the parameter
is selected as
and the principal frequency is chosen as
. We integrate this problem in the interval [0, 100]. In order to compare error of different methods, we use step lengths
We emphasize that the analytic solution of this problem does not exists, thus, we used a reference numerical solution which is obtained via special perturbation approach (Anderson and Geer [
59] and Verhulst [
60]). The BFFM results in comparison with the Block Falkner methods (BFM) of order 5 in Ramos et al. [
32], Modified Block Falkner methods (MBFM) of order 5 in Ehigie and Okunuga [
36] and The two-stage and three-stage Two-derivative Runge-Kutta-Nystrom Methods (TDRKN2 and TDRKN3) of orders 4 and 5 respectively in Chen et al. [
25] are displayed in
Table 3 while the efficiency curves are displayed in
Figure 4 respectively. It is evident from the results in
Table 3 and
Figure 5 that BFFM performs better than some of the existing methods in the literature.
4.4. Problems Where the First Derivative Does Not Appear Explicitly
4.4.1. Example 4
As our fourth experiment, we consider the periodically forced nonlinear IVP
whose analytic solution is
. For this problem,
is selected as principal frequency with parameter
.
Table 4 shows the performnce of BFFM in comparison with the TFARKN by Fang et al. [
14], the EFRK by Franco [
39] and the EFRKN by Franco [
10] respectively. The efficiency curves of the BFFM and the other methods used for comparisons are displayed in
Figure 6.
4.4.2. Example 5
As our fifth numerical experiment, we consider the following nonlinear system
with
, whose solution in closed form is given as
Table 5 and
Figure 7 show the superiority of the BFFM in the interval [0, 100] over the BNM of order 5 in Jator and oladejo [
57], the BHM of order 11 in Jator and King [
61] and the fourth order ARKN in Franco [
9].
4.4.3. Example 6
We consider the following well known two body problem
where
, and the solution in closed form is given by
,
.
Table 6 reveals the performance of our proposed BFFM in the interval
with
as it compared with the fourth order DIRKNNew of Senu et al. [
15] while
Figure 8 establishes the efficiency of BFFM.
5. Conclusions
In this paper, we have proposed a family of Adapted block Falkner methods using third derivative for solving second order initial value problem with oscillatory solution directly numerically. The methods are applied in block form as simultaneous numerical integrators and thus, do not suffer the disadvantages of predictor-corrector mode. The basic properties of the methods are investigated and discussed. The convergence of the proposed methods was established and the stability regions are presented. The numerical results on well-known second order initial value problems with oscillatory solutions show the effectiveness of the proposed methods compared with some existing methods in the reviewed literature. Although the proposed family of methods can be implemented with variable steps, that aspect was not considered in the current work, but will be considered in our future research.