INTRODUCTION
1.1 Background to the study
A differential equation is an equation involving a relation between an unknown function and one or more of its derivatives. Differential equations are among the most important mathematical tools used in producing models in the physical sciences, biological sciences, social and management sciences and engineering. They occur in connection with the mathematical description of problems that are encountered in various branches of science. Consequently, it constitutes a large and very important aspect of today’s mathematics.
Differential equation is a process by which solutions can be sort to some real life problems. These problems can either be solved by the use of analytical techniques or by numerical methods. Since most ordinary differential equations are not analytically solvable, numerical methods are often better option. Many methods have been proposed and used by different authors with the aim of providing accurate solutions to the various types of differential equations. Differential equation is divided into two parts, ordinary differential equation and partial differential equations; here our work is centred on proposing a technique that can solve problems in ordinary differential equations, although many of such methods already exist. Our focus here is on numerical solutions to ordinary differential equations with particular emphasis on the use of linear multistep methods.
Stiff differential systems including the building energy simulation problems, are difficult and costly to compute. Standard explicit solvers are compact, and time stepping with them is cheap, but many active increments are required. Implicit solvers offer stability for any time increment at the cost of a lot of computation per step. What is needed is a method that can take a long time cheaply. Exponential fitting methods offer this option. Abhulimen (2006).
The rational behind the development of this kind of numerical integrator is that exponentially fitted formulae possess a large region of absolute stability when compared to conventional ones, Hochbruck, Lubich, Selhfer (1998).
In the last decades, several authors such as Enright (1974), Enright and Pryce (1983), Brown (1977), Cash (1981), Jackson and Kenue (1974) Voss (1988), Okunuga (1994), Abhulimen and Okunuga (2008), and Abhulimen and Omeike (2011) developed second derivative integrators for the numerical solutions of stiff differential equations. These integrators however were found to be A-stable, particularly for stiff problems whose solutions have exponential functions.
1.2 Ordinary differential equations (ODEs):
Many problems in science and other areas involving rate of change usually resolve into ODEs. The most general form which ODEs may assume is giving by;
1.2.1)
where is the independent variable, is the dependent variable,
So that
more compactly we represent (1.2.1) in vector form as;
where so that and denotes transpose.
1.3 Initial Value Problems (IVPs) for First Order Ordinary Differential Equation
The first order differential equation may possess an infinite number of solutions. For example the function is, for any value of the constant , a solution of the differential equation , where , is a given constant. We can pick out any particular solution by prescribing an initial value condition, . For the above example, the particular solution satisfying this initial condition is easily found to be we say that the differential equation together with an initial condition constitutes an initial value problem,
1.4 Lipschitz Condition
Theorem 1.1: let be a real function and continuous for all points in the region defined by , containing initial values where are finite. Let there exist a constant such that for any and for any pairs for which are both in
then for any giving number . The initial value problem (1.3.1) has a unique solution . is called the Lipschitz constant. This condition maybe thought of as being intermediate between differentiability and continuity, such that:
is continuously differentiable w.r.t. for all in .
satisfies a lipschitz condition w.r.t. for all in .
is continuous w.r.t. for all in .
In particular, if possesses a continuous derivative with w.r.t. for all in , then by mean value theorem,
where is a point in the interior of interval whose end-point are and and and are both in . Then the lipschitz constant of the system may be taken to be
(1.4.3)
1.5 Numerical Methods for Solving Initial Value Problems in Ordinary Differential Equations
Numerical methods are methods used for solving Ordinary Differential Equations. According to Shepley (1989), numerical methods are employed in the solution of the differential equation with the initial-condition to obtain approximate solution at various selected values of x with the aim of having exact solution. To do this we set as the exact solution of the problem, and let denote a small positive increment in x. Let and consider
( . A numerical method will use the differential equation and the condition to successively approximate these exact values ( . Let be the approximations to respectively, so that finding and finding an approximation to mean the same thing. In finding the approximation we proceed in the following way:
First, we find using the method of interest to solve the differential equation with the initial value . Then is estimated using the estimate , is estimated using the estimate , and so on, so that in general, is estimated using the estimate . A method which proceeds in this manner is called a one-step method. On the other hand, in finding some methods actually use several of the preceding approximations to estimate the differential equation with the given initial condition . Such methods cannot find from with the initial condition . Hence such methods are called multi-step method. To use a multi-step method, the first few must be found by a starting method, until a sufficient number of them are on hand to begin using the continuing method. Most of our attention in this sense will be devoted to starting methods. Shepley continues by saying that given an approximation to , the absolute error, or simply error is defined as ; the error measures how far away the approximation is from the exact value . Naturally we hope that any given numerical method will keep the error small, that is, the method should have some level of accuracy.
1.5 JUSTIFICATION
Numerical method is a part of numerical analysis which studies the methods for finding numerical approximations to the solution of ordinary differential equations. Some of the existing methods for solving IVPs are the one-step methods e.g Runge-kutta, milne method, the linear multistep methods, the hybrid methods.
Of all numerical methods for the numerical solution of this initial value problem the easiest to implement is the Euler’s rule. Lambert (1973).
According to Sheply (1989), numerical methods are employed in the solution of the differential equation with the initial condition to obtain approximate solution at various selected values of .
A good and potential numerical method for the solution of initial value problems in ordinary differential equations must possess good accuracy and some reasonable wide region of absolute stability, Dahlquist (1973).
Stability is the property of a numerical method to keep the errors bounded as the computation advances Abhulimen (2009). A-stability is one of the important stability requirements for a linear multistep method, Enright (1974). Dahlquist proved that the order of an A-stable linear multistep method is which linear multistep method is limited by the requirement of A-stability. This has made researchers to find other classes of numerical methods for higher order to solve differential equations.
Liniger and Willoughby (1970) developed the concept of exponential fittings in the course of higher order A-stable numerical methods, which allows free parameters that are chosen to fit some exponential integration functions that satisfies integration formula exactly. Recently exponential integrators have become an active area of research.
Before designing our integrators we considered many methods and were motivated by the striking proposals made by the following authors:
Jackson and Kenue (1974), Cash (1981), Okunuga (1992), Hochbuck, Lubish et al (1998), Abhulimen and Otunta (2009), Okunuga (1999), Abhulimen and Otunta (2006), Vigo and Martin (2006, 2007), Abhulimen (2006, 2014), Abhulimen and Okunuga (2008).
Most of these developed formulae allow free parameters and act as a motivation in deriving a 4-step 4th
derivative method of order 6 and 6-step 3rd
derivative method of order 8 with one free parameter.It is important to note that for systems for which exponential fitting is appropriate, it is usually found that exponential fitted integration formulas are substantially more efficient than conventional ones. Abhulimen and Omeike (2011). The exponential fitting method also offers favourably properties in the integration of differential equations whose Jacobian has large imaginary eigenvalues (Hochbruck et al.1998).
1.7 OBJECTIVES OF THE STUDY
The overall aim of the study is to derive a four-step fourth derivative exponentially fitted integrator of order six and a six-step third derivative exponentially fitted integrator of order eight which are A-stable for all choices of the fitting parameters.
The specific objectives of the study are as follows:
1.8 RESEARCH METHOD
In the spirit of Cash (1981) and Abhulimen (2014), the general form taken in the derivation of our new integrator is given as:
where equation (1.7) and (1.8) are use as predictor and corrector respectively, and are positive integers, is a step length. The coefficients and are real constants. are approximate to , ,
When deriving exponentially fitted methods, the approach is to allow both (1.8.1) and (1.8.2) to possess free parameters which allow it to be fitted automatically to exponential functions. Abhulimen and Omeike (2011).