Mal 514
Mal 514
Mal 514
MATHEMATICS
MAL-514 ORDINARY DIFFERENTIAL EQUATIONS I
150
CONTENTS
Chapter Name Chapter 1 Differential Equations and Existence of Solution Chapter 2 Existence and Uniqueness of Solutions Chapter 3 Approximate methods of solving first order equation Continuation of solutions and matrix method for system of equation Total Differential Equations and Differential Inequations Sturm Theory and Riccati Equations Strum-Liouville Boundary-Value Problems
Author Dr. Sunita Pannu Dr. Sunita Pannu Dr. Sunita Pannu Dr. Sunita Pannu Dr. Sunita Pannu Dr. Sunita Pannu Dr. Sunita Pannu
Vetter Mr. Kapil Kumar Mr. Kapil Kumar Mr. Kapil Kumar Mr. Kapil Kumar Mr. Kapil Kumar Mr. Kapil Kumar Mr. Kapil Kumar
Page N
1 31
51
Chapter 4
75
Chapter 5
100
Chapter 6
127
Chapter 7
149
151
Written by: Sunita Pannu Vetted by: Kapil Kumar
Chapter-1
Objectives The general purpose of this chapter is to provide an understanding of basic aspects of differential equations. A geometrical approach is used to prove the existence of solution of an initial value problem. Introduction We live in a world of interrelated changing entities. The position of earth changes with time, the velocity of a falling body changes with distance, the path of a projectile changes with the velocity and angle at which it is fired. In the language of mathematics, changing entities are called variables and the rate of change of one variable with respect to another a derivative. Equation which expresses a relationship among these variables and their derivatives are called differential equations. For example, from certain facts about the variable position of a particle and its rate of change with respect to time, we want to determine how the position of particle is related to the time, so that we can know where the particle was, is, or will be at any time t. Differential equations thus originate, whenever a universal law is expressed by means of variables and their derivatives. Relation of Differential Equations to various fields Differential equations arise in many areas of science and technology: where a deterministic relationship involving some continuously changing quantities (modeled by functions) and their rates of change (expressed as derivatives) is known or postulated. Newtons Laws allow one to relate the position, velocity, acceleration and various forces acting on the body and state this relation as a differential equation for the unknown position of the body as a function of time. The study of differential equations is a wide field in pure and applied mathematics, physics and engineering. All of these disciplines are concerned with the
152
properties of differential equations of various types. Pure mathematics focuses on the existence and uniqueness of solutions, while applied mathematics emphasizes the rigorous justification of the methods for approximating solutions. Differential equations play an important role in modeling virtually every physical, technical, or biological process, from celestial motion to bridge design, to interactions between neurons. Many famous mathematicians have studied differential equations and contributed to the field, including Newton, Leibniz, the Bernoulli family, Riccati, Clairaut, d Alembert and Euler. Preliminaries In the real (x,y) plane, we denote x for independent variable and y for dependent variable. Then y will be a function of x and its value at x [a, b] will be considered as y(x). Similarly in the real (t, x) or (t, y) plane, t will be treated as independent variable and x or y as dependent variables. We shall consider any such plane and the real or complex valued functions defined on any domain D in such plane, where by a domain we understand an open connected set. Differential Equation An equation involving derivatives of one or more dependent variables with respect to one or more independent variables is called a differential equation. This relation between dependent variable (and its derivatives) and independent variables is non trivial i.e. some equations that appear to satisfy above definition are really identities. For example,
dy dy dy 2 y = 2y + y dt dt dt
(1)
dy dy sin 2 + cos 2 = 1 dt dt This equation (2) is satisfied by every differential function of one variable.
(2)
So, non trivial manner means we do not include these types of equations in the class of differential equations. We exclude such expressions as
d ax (e ) = aeax , dx
153
d dv du (uu) = u +v dx dx dx and so forth.
Physical Meaning
w.r.t. x. In any natural process the variables involved and their rates of change are connected with one another by means of the basic scientific principles (that govern the process). When this connection is expressed in mathematical symbols, the result is often a differential equation. For example, according to Newtons second law of motion, acceleration of a body of mass m is proportional to the total force F acting on it i.e. a F or a = 1 1 F with as constant of proportionality. m m (1)
F = ma
For instance, let a body of mass m falls freely under the influence of gravity alone. In this case only force acting on it is mg, where g is acceleration due to gravity. In most applications g is equal to 32 feet per second (or 980 centimeters per second per second). If y is distance down to the body from some fixed height, than its velocity v= dy d2 y is rate of change of position and acceleration a = is rate of change of dt dt 2
d2 y =g dt 2
If we alter the situation by assuming that air exerts a resisting force dy proportional to the velocity, then total force acting on the body is mg - k and (1) dt becomes
154
m d2 y dy = mg - k . 2 dt dt (3)
Equation (2) and (3) are differential equations that express the physical processes under consideration.
General Form
A differential equation having y as the dependent variable (unknown function) and t as the independent variable has the form F[t, y, dy , dt d2 y dn y , ., ] = 0 for some positive integer n. if n = 0, the dt 2 dt n
d ) of dt
one as more dependent variables w.r.t. a single independent variable is called an ordinary differential equation, e.g. d2 y dy + xy = 0 and 2 dx dx d4x d2x + 5 + 3x = sin t dt 4 dt 2
2
(1.1) (1.2)
are ordinary differential equations. In (1.1) the variable x is the single independent variable and y is a dependent variable. In (1.2), the independent variable is t, whereas x is dependent variable. Some general examples of ordinary differential equations Legendres equation (1 x 2 ) Bessels equation
x2
d2 y dy 2 x + p( p + 1) y = 0 2 dx dx
(1.3)
d2 y dy + x + ( x2 p2 ) y = 0 2 dx dx
(1.4)
155
2. Partial Differential Equation
A differential equation involving partial derivatives of one or more t dependent variables w.r.t. more than one independent variables is called a Partial Differential Equation. e.g. u u + =u s t In (2.1), s and t are independent variables and u is dependent variable. Also, If u = f(x, y, z, t) is a function of time and three rectangular coordinates of a point in space, then following are partial differential equations of 2nd order. Laplace equation
2u 2u 2u + + =0 x 2 y 2 z 2
(2.1)
(2.2)
Heat equation
2u 2u 2u u a2 2 + 2 + 2 = y z t x
(2.3)
Wave equation
2u 2u 2u 2u a2 2 + 2 + 2 = 2 y z t x
(2.4)
Generally partial differential equations arise in the physics of continuous media in problems involving electric fields, fluid dynamics, diffusion and wave motion.
Order of a Differential Equation
The order of a differential equation is the order of the highest derivative present in the differential equation. Equations (1.1) and (2.2) are of 2nd order, (1.2) is of fourth order and (2.1) is of first order.
Linear Ordinary Differential equation
A differential equation is said to be linear if it is linear in y and all its derivatives. i.e. a linear differential equation of order n, in the dependent variable y and the independent variable x is an equation that can be expressed in the form
156
a0 ( x) dn y d n-1 y dy + a x + ..... + an 1 ( x) + an ( x) y = b( x) , ( ) 1 n n-1 dx dx dx (A)
where a0 is not identically equal to zero. Thus we observe here that (1) The dependent variable y and its various derivatives occur to the first degree only. (2) (3) No products of y and/or any of its derivatives are present. No transcendental (i.e. vague) function of y and/or its derivatives occur. (i) d2 y dy + 7 + 6 y = 0 (equation with constant coefficients) 2 dt dt d2 y dy t + 6 + y = sin t (equation with variable coefficients) 2 dt dt
7
3 d4 y dy 2 d y + x + x3 = xe x (equation with variable coefficients) 4 3 dx dx dx
Examples:
(ii)
(iii)
Here y and its derivatives occur to the first degree only and no product of y and/or any of its derivatives are present.
Non Linear Differential Equation
Equation F(x, y, y1, .. y(n),) (where F is a known function) is called a non linear differential equation of order n, if it cannot be written in the linear form as in (A) d 2 y dy +5 +6y=0 dx 2 dx d2 y dy + 5y + 6y = 0 2 dx dx
3
(1)
(2)
Equation (1) is non-linear due to presence of the term 5(dy/dx)3, which involves third power of first derivative. Equation (2) is non-linear because of the term 5y(dy/dx), which involves the product of dependent variable and its first derivative.
157
Application of Differential Equations
Differential equations occur in numerous problems that are encountered in various branches of science and engineering. A few such problems are: (1) The problem of determining the motion of a projectile, rocket, satellite or planet. (2) (3) (4) (5) (6) The problem of determining the charge or current in an electric circuit. The problem of the conduction of heat in a rod or in a slab. The problem of determining the vibrations of a wire or a membrane. The study of the rate of growth of a population. The problem of the determination of curves that have certain geometrical properties.
Solutions
A functional relation between the dependent variable y and the independent variable x, that, in some interval J, satisfies the given differential equation i.e. reduces it to an identity in x is said to be a solution of the equation. The general solution of an nth order differential equation depends on x and on the n arbitrary real constants c1, c1, c2, cn. e.g. the function y(x) = x2 + cex is the general solution of the differential equation. y' = y x2 + 2x in J = R Also y(x) = x3 + c is a general solution of x3 (B)
xy' + 3y = 6x3
and the function y(x) = x3 is a particular solution of the equation (B), obtained by taking the particular value c = 0 in the general solution of (B).
Singular Solution
The solution which cannot be obtained by assigning particular values to the constants in general solution is called a singular solution.
158
For example in differential equation y'2 - xy' + y = 0, the general solution is y(x)= cx - x2, which represents a family of straight lines and y(x) = 1 2 x is a singular 4
solution which represents a parabola. Thus in General Solution the word general must not be taken in the sense of complete. A totality of all solutions of a differential equation is called a complete solution.
Explicit Solution
A differential equation of first order may be written as F(x, y, y') = 0. The function y = (x) is called an explicit solution of this differential equation provided F(x, (x), '(x)) = 0 in I. A relation of the form (x,y) = 0 is said to be an implicit solution of
F(x, y, y') = 0 provided it determines one or more function y = (x) which satisfy F(x, (x), '(x)) = 0. It is frequently difficult (if not impossible) to solve (x, y) = 0 for y. So, we can test the solution by obtaining y' by implicit differentiation x = yy' = 0 or y' = - x/y and check if F(x, y, - x/y) 0 Thus a relation (x,y) = 0 is called an implicit solution of given differential equation, if this relation defines at least one real function of the variable x on an interval I such that this function is an explicit solution of given differential equation on this interval. Thus y = g(x) is a solution of differential equation F[x, y, y, .. y(n)] =0 on an interval I means that F[x, g(x), g'(x), .., g(n)(x)] = 0, for every choice of x in the interval I. In other words, a solution, when substituted into the differential equation makes the equation identically true for t in I.
Some Definitions from Real Function Theory Definition A
A sequence {cn} of real numbers is said to converge to limit c if, given any > 0, there exists a positive number N such that
159
cn - c < for all n > N. We indicate this by writing lim cn = c.
n
Definition B
Let {fn} be a sequence of real functions, each of which is defined on the real interval a x b. The sequence {fn} is said to converge pointwise to f an a x b, if, given any > 0, for each x such that a x b there exists a positive number N (in general depending both on and on x) such that
fn(x) - f (x)<
Let {fn} be a sequence of real functions, each of which is defined on the real interval a x b. The sequence {fn} is said to converge uniformly to f on a x b, if, given any > 0, there exists a positive number N (which depends only upon ) such that
fn(x) - f (x)<
for all n > N for every x such that a x b Geometrically, this means that give > 0, the graphs of y = fn(x) for all n >N lie between the graphs of y = f(x) + and y = f(x) - for a x b (see figure).
Figure
160
Statement of two important theorems to be used in existence and uniqueness theorem are as follows:
Theorem - A Hypothesis
1.
Let {fn} be a sequence of real functions which converges uniformly to f on the interval a x b.
2.
1.
Let {fn} be a sequence of real functions which converges uniformly to f on the interval a x b.
2.
Suppose each function fn(n=1, 2, 3, .. ) is continuous on a x b. Then for every and such that a < b. lim f n ( x) dx = lim f n (x) dx .
n
Conclusion
Definition D
u
n =1
which is defined on a real interval a x b. Consider the sequence {fn} of partial sums of this series, defined as follows: f1 = u1, f2 = u1 + u2 f3 = u1 + u2 + u3, fn = u1 + u2 + u3 + . .. . + un The infinite series
u
n =1
161
Weierstrass M Test Hypothesis
1.
Let {Mn} be a sequence of positive constants such that the series of constants
M
n =1
converges.
2.
Let
u
n =1
be a series of real functions such that |un(x)| Mn for all x such that
The series
u
n =1
converges uniformly on a x b.
1.
A set of points A of the xy plane will be called connected if any two points of A can be joined by a continuous curve which lies entirely in A.
2.
A set of points A of the xy plane is called open if each point of A is the center of a circle whose interior lies entirely in A.
3. 4.
An open, connected set in the xy plane is called a domain. A point P is called a boundary point of a domain D if every circle about P contains both points in D and points not in D.
5.
Definition F
Let f be real function defined on a domain D of the xy plane, and let (x0, y0) be an interior point of D. The function f is said to be continuous at (x0, y0) if, given any
> 0, there exists a > 0 such that f(x, y) f(x0, y0)<
162
Definition G
Let f be a real function defined on D, where D is either a domain or a closed domain of the xy plane. The function f is said to be bounded on D if there exists a positive number M such that f(x, y) M for all (x, y) in D.
Theorem C Hypothesis
Let f be defined on D, where D is either a domain or a closed domain of the xy plane. The function f is said to satisfy a Lipschitz condition (with respect to y) in D if there exists a constant k> 0 such that
f(x, y1) - f(x, y2) ky1 - y2
for every (x, y1) and (x, y2) which belong to D. The constant k is called the Lipschitz constant.
Initial Value Problem
We now formulate the basic problem Let D be a domain in the xy plane and let (x0, y0) be an interior point of D. Consider the differential equation dy = f(x,y), dx where f is a continuous real function defined on D. We want to determine: 1. 2. an interval x of the real x-axis such that < x0 < , and a differentiable real function defined on this interval [,] and satisfying the following three requirements. (i) [x, (x)] D, and thus f[x, (x)] is defined, for all x [,]. (1)
163
(ii)
d ( x ) = f [ x, ( x)], and thus satisfies the differential equation (1), for all dx
x [,]. (iii)
(x0) = y0
This problem is called the initial-value problem associated with the differential equation (1) and the point (x0, y0). We shall denote it by dy = f(x,y), dx y(x0) = y0 (2)
and a function satisfying the above requirements on an interval [,] is called a solution of the problem on the interval [,]. If is a solution of the problem on [,], then requirement (ii) shows that has a continuous first derivative ' on [,]
Basic Lemma Hypothesis
1. 2.
Let f be a continuous function defined on a domain D of the xy plane. Let be a continuous function defined on a real interval x and such that [x, (x)] D for all x [, ].
3.
Conclusion
on [, ] and is such that (x0) = y0 if and only if satisfies the integral equation
(x) = y0 +
for all x [,].
x0
f [t, (t)] dt
(2)
164
Proof If satisfies the differential equation dy/dx = f(x, y) on [, ], then
(3)
for all x [,]. So, (x) is a continuous function on [, ], because it is differential on [,]. Since (x) is continuous on [, ] and f is continuous on D, the function F(x) = f [x, (x)] is continuous on [, ], so that it is integrable on [, ]. Integrating (3) from x0 to x.
(x) - (x 0 ) =
x0
f [t, (t)] dt
Since (x0) = y0, so satisfies the integral equation (2) for all x [,]. Conversely, let (x) satisfies the integral equation (2) for all x [, ]. Using the fundamental theorem of integral calculus, differentiation yields
d ( x ) = f [ x, ( x)] dx
for all x [, ] and so satisfies the differential equation Further from (2), we get (x0) = y0.
dy = f(x,y) on [, ]. dx
Thus (x) is a solution of (1) if and only if, it satisfies the integral equation (2).
Note
Equation (2) is called an integral equation because the unknown function (x)
An integral equation is an equation in which an unknown function appears under the integral sign. If the limits of integration are fixed, an integral equation is called a Fredholm Integral Equation. If one limit is variable, it is called a Volterra Integral equation. Volterra Integral equations are divided into two groups referred to as the 1st and 2nd kind. A volterra integral equation of the first kind is f(x)= K(x,t) (t)dt
a x
(1)
165
where (t) is the function to be solved for, f(x) is a specified function i.e. given known function and K(x, t) is the integral Kernel i.e. a known function of two variables. Thus if the unknown function is only under the integral sign, the equation is said to be of 1st kind. If the unknown function is both inside and outside the integral, equation is called of 2nd kind. i.e.
The parameter is an unknown factor (may be = 1). An example of integral equation is f(x) = e-x 1 1 -(x +1) 1 + e + (x + 1)e-xy f(y)dy 2 2 20
1
which has solution f(x) = e-x. An integral equation is called homogeneous if f(x) = 0 i.e. known function f is identically zero. If f is non-zero, it is called inhomogeneous integral equation. Of course, not all integral equations can be written in one of these forms.
Reduction of Initial value problems to Volterra integral equations
Now, we shall illustrate that how an initial value problem associated with a linear differential equation and auxiliary conditions reduce to a Volterra integral equation.
Example 1
Convert the given IVP to an integral equation y"(t) + y(t) = cos(t), y(0) = 0, y'(0) = 0 y"(t) = -y(t) + cos t, integrate from 0 to t (1)
166
y'(t) = y() d + [sin(t) - sin(0)]
0 t
[y'(0) = 0]
(2)
(3)
y()(t - )d + sin t t 0
= -y(t) + cos t
Reduce IVP to integral equation y"(t) 2ty'(t) 3y(t) = 0, y(0) = 1, y'(0) = 0 y"(t) = 2ty'(t) + 3y(t),
167
y'(t) 0 = 2 y() 0 2 y()d + 3 y()d
t
y()d
0
y() d ds
0
nd
This Volterra integral equation satisfies the given initial conditions. We can obtain the original IVP by twice differentiating it. Differentiating once y'(t) =
=t
dt d0 [ (t + ) y() ] =0 dt dt
y'(t)
y()d + 2t y(t)
0
t
0
y()d + y() = t
168
Note Cauchy Method
1.
(x)dx dx =
a a
x x
(x-) ()d
a
or
(y)dy ds
a a
x s
(y)(x-y) dy
a x
2.
0 0
x x 2 x1
3.
0 0 0
x x n x n-1
1 (x ) n-1 () d (n 1)! 0
Example 3
d2 y dy +A(t) + B(t)y(t) = g(t) 2 dt dt y(a) = c1, We write d2 y dy = A(t) B(t)y(t) + g(t) 2 dt dt We now integrate over the interval [a, t] to obtain dy c 2 = - A()y'()d- B()y()d + g()d dt a a a = [-A()y()] + A'()y()d - B()y()d + g()d
t a
(1) (2)
y'(a) = c2.
(3)
169
+ (t s)g(s)ds + c1A(a)(t-a) .
a t
(4)
where the non-homogeneous term f(t) is f(t) = [(t s)g(s)ds + (t-a)[c1 (A(a)) + c2 ] + c1
a t
(5)
Equation (4) is a Volterra integral equation of the second kind of the type y(t) = f(t) + K(t,s)y(s)ds
a t
(6)
in which the kernel K(t,s) is given by K(t, s) = (t-s) [A' (s) B(s)] - A(s). (7)
Integral equation (4) is equivalent to the given initial value problem and it takes care of auxiliary condition in (2).
Exercise
Obtain the Volterra integral equation corresponding to each of the following initial value problems. (a) (b) (c) (d) (e) y'' + y = 0; y'' + y = sin t; y'' - y + t = 0; y'' + y = f(t); y'' + ty = 1; y(0) = 1, y'(0) = 0 y(0) = 1, y'(0) = 1 y(0) = 1, y'(0) = 0 y(0) = 1, y' (0) = 0 y(0) = y'(0) = 0.
170
Existence Theorem
Let I denote an open interval on the real line - < t <, that is, the set of all real t satisfying a < t <b for some real constants a and b. The set of all complex-valued functions having k continuous derivatives on I is denoted by Ck(I). If f is a member of this set, one writes f Ck (I), or f Ck on I. The symbol f is to be read is a member of or belongs to. It is convenient to extend the definition of Ck to intervals I which may not be open. The real intervals a < t <b, a t b, a t <b, and a < t b will be denoted by (a, b), [a, b], [a, b), and (a, b], respectively. If f Ck on (a, b) and in addition the right hand kth derivative of f exists at a and is continuous from the right at a, then f is said to be of class Ck on [a, b). Similarly, if the kth derivative is continuous from the left at b, then f Ck on (a, b]. If both these conditions hold, one says f Ck on [a, b]. If D is a domain in the real (t, y) plane, the set of all complex valued functions f defined on D such that all kth order partial derivatives kf/tp yq (p + q = k) exist and are continuous on D is denoted by Ck(D), and one writes f Ck on D or f Ck (D). The sets C0(I) and C0(D), the continuous functions on I and D, will be denoted by C(I) and C(D), respectively.
Definition ( - approximate solution)
Let f be a real valued continuous function on a domain D in the (t, y) plane. An - approximate solution of an ODE of the first order dy = f(t, y) dt on a t-interval I is a function C(I) such that i) (t, (t)) D for all t I
CI(I), except possibly for a finite set of points S on I where '(t)may have
(E)
ii)
simple discontinuities (i.e., at such points of S, the right and left limits of '(t) exist but are not equal), iii)
|'(t) f(t, (t)| < for all t I S
171
Remark
(1)
In geometrical language, (E) prescribes a slope f(t, y) at each point of D. A solution on I is a function whose graph [the set of all points (t, (t)), t I] has the slope f(t, (t)) for each t I.
(2)
When = 0, then it will be understood that the set S is empty, i.e. S = . So (ii) holds for all t I.
(3)
The statement (ii) implies that has a piecewise continuous derivative on I, and we shall denote it by
C1 p (I)
(4) Consider the rectangle R = {(t, y): |t-t0| a, |y-y0| b, a > 0, b > 0} about the point (t0, y0). Let f C on the rectangle R. Since the rectangle R is a closed set, so the continuous function f on R is bounded. Let M = max |f(t, y)| on R
b and = min a, M
(1)
(2) (3)
Let f C on the rectangle R. Given > 0, there exists an - approximate solution of ODE of first order. dy = f(t, y) dt on the interval I = {t: |t-t0| } such that (t0) = y0, being some constant. (1)
172
Proof Let > 0 be given. We shall construct an - approximate solution for the
interval [t0, t0 + ]. A similar construction will hold for [t0- , t0]. This approximate solution will consist of a polygonal path starting at (t0, y0), i.e. a finite number of straight- line segments joined end to end. Since f is continuous on the closed rectangle R = {(t, y): |t-t0| a, |y-y0| b, a > 0, b > 0} So f is bounded and uniformly continuous on R. Let
M = max f(t,y)
R
(2)
(3) (4)
= min a,
b M
= a if M >
b a
(fig 1(a))
(ii)
b b if M M a
(fig 1(b))
In the first case, we get a solution valid in the whole interval |t-t0| a, whereas in the second case, we get a solution valid only on I, a sub-interval of |t-t0| a.
Figure 1
on R, therefore, for given > 0, there exists a real number = () > 0 such that
f(t, y) f( t, y)
(5)
173
whenever
t - t , y - y , where (t, y), ( t, y) R
(6)
Now divide the interval [t0, t0 + ] into n subintervals such that t0 < t1 <. . . <tn = t0 + and
(7)
Figure 2
Draw two lines from C(t0, y0) one having the slope M and the second line having the slope M and then draw third line t = t0 + . These three lines will enclose the region T. Starting from the point C(t0, y0) we draw a straight-line segment with slope f(t0, y0) proceeding to the right of t0 until it intersects the line t = t1 at some point P1 (t1, y1). Here, slope of line CP1 is f(t0, y0). This line segment, CP1, must lie inside the triangular region T, as shown in the figure 2 above because, we have, in this case,
=
b M
(8)
Now, at the point P1 (t1, y1), we construct to the right of t1 a straight-line segment with slope f(t1, y1) upto the intersection with line t = t2, say at the point P2(t2, y2).
174
Continuing in this way, after a finite number of steps, the resultant path (t) will meet the line t = t0 + at the point Pn (tn, yn). Further, this polygon path (fig. 3) will lie entirely within the region T. This is the required -approximate solution.
Pn(tn, yn) C
Figure 3
(9)
for t [tk-1, tk] and k = 1, 2,...., n and (t0) = y0. From the construction of the function , it is clear than C1 p on [t0, t0 + ], and that
(t) - ( t) M t t
where t, t are in [t0, t0 +].
(10)
If t is such that tk-1 t tk, then equations (7) and (10) together imply that
(11)
= M
(12)
175
where tk-1 t tk . This shows that is an - approximate solution as desired. By similar arguments, the theorem can be proved on the interval [t0-, t0]. Hence is an approximate solution on the interval [t0-, t0 + ].
Remark
After finding an - approximate solution of an IVP, one may prove that there exists a sequence of these approximate solutions which tend to a solution. For this, the notion of an equicontinuous set of functions is required.
Equicontinuous Set of Functions
A set of functions F = {f} defined on a real interval I is said to be equicontinuous on I if, for given any > 0, there exists a real number = () > 0, independent of f F, such that
f(t) f( t) <
On a bounded interval I, let F = {f} be an infinite, uniformly bounded, equicontinuous set of functions. Prove that F contains a sequence which is uniformly convergent on I.
Proof Let {rk}, k = 1,2........, be all the rational numbers present in the bounded
interval I listed in some order. The set of numbers {f(r1): f F}is bounded, hence there exists a sequence of distinct functions {fn1}, fn1 F, such that sequence {fn1 (r1)} is convergent (By Bolazano Weierstrass Theorem, every infinite bounded sequence has a subsequence, which is convergent). Similarly, the set of number {fn1 (r2)} has a convergent subsequence {fn2 (r2)}. Continuing in this way, we obtain, an infinite set of functions fnk F, n, k = 1, 2, ..., such that {fnk} converges at r1, r2,...., rk. Define gn = fnn. (1)
176
Now, it will be shown that {gn) is the required sequence which is uniformly convergent on I. Clearly, {gn} converges at each of the rationals r1,r2,....rk on I. Thus, given any > 0, and each rational number rk I, there exists an integer N(rk) such that |gn (rk) gm(rk)| < for n, m > N(rk). (2)
Since the set F is equicontinuous, there exists a real number = = () > 0, which is independent of f F, such that
f(t) f( t) < ,
(3)
whenever
t t < and t, t I .
We divide the interval I into a finite number of subintervals, I1, I2, .. Ip such that the length of the largest subinterval is less than , i.e., max{l(Ik):k = 1, 2, ..... ,p} < . (4)
For each such subinterval Ik, choose a rational number rk I k . If t I, then t Ik for some suitable k. Hence, by (2) and (3), it follows that
g n (t)-g m (t) g n (t)-g n (rk ) + g n (rk )-g m (rk ) + g m (rk ) - g m (t)
(5)
This proves the uniform convergence of the sequence {gn} on I, where gn F for each n N. This completes the proof.
Remark
The existence of a solution to the initial value problem, without any further restriction on the function f(t, y) is guaranteed by the following Cauchy-Peano theorem.
177
Theorem 1.3 Cauchy-Peano Existence Theorem
If f C on the rectangle R, then there exists a solution C1 of the differential equation. dy = f(t,y) dt on the interval t t 0 for which (t 0 ) = y0 , where
b R = {(t, y): t t 0 a, y y0 b,a > 0, b > 0 }, = min a, , M
M = max f(t,y) on R .
Proof. Let {n}, n = 1,2,.........., be a monotonically decreasing sequence of positive
real numbers which tends to 0 as n. By Cauchy-Euler Construction Theorem, for each such n there exists an n- approximate solution, say n of ODE dy = f(t, y) dt on the interval
t - t 0 with n (t 0 ) = y 0
(1)
(2)
M = max |f(t, y)| for (t, y) R R = {(t, y): |t - t0| a, |y - y0| b, a > 0, b > 0} Further, from Cauchy Euler Construction Theorem
n (t) - n ( t) M t - t
for t, t in [t0, t0 + ]. Applying (6) to t = t0 and since, t - t0 it follows that b , M
(6)
(7)
178
|n(t) - y0| < b for all t in |t - t0| < .
(8)
This implies that the sequence {n} is uniformly bounded by |y0| + b. Further, (6) implies that the sequence {n} is an equicontinuous set. Therefore, by Ascoli Lemma there exists a subsequence {nk}, k= 1, 2, ...., of {n}, converging uniformly on the interval [t0- , t0+] to a limit function , which must be continuous, since each n is continuous. Now, we shall show that this limit function is a solution of (1) which satisfies the required specification. For this, we write the relation defining n as an napproximate solution in an integral form, as follows:
n(t) = y0 + [f(s, n ( s )) + n ( s )]ds
t0 t
(9)
where
n ( s ) = 'n ( s ) f ( s, n ( s ))
(10)
at those points where 'n exists and n(s) = 0 otherwise. Because n is an n - approximate solution, so
|n(s)| n.
(11)
Since f is uniformly continuous on R, and nk uniformly on [t0 - , to + ] as k , it follows that f(t, nk(t)) f(t, (t)) uniformly on [t0 - , to + ] as k . Replacing n by nk in (9) and letting k ,
(t) = y0 +
t0
f(s, (s))ds
(12)
(13)
179
d = f (t , (t )) . dt
(14)
It is clear from (13) and (14) that is a solution of ODE (1) through the point (t0, y0) on the interval |t - t0| of class C1. This completes the proof of the theorem.
Remarks
(1)
(2)
It can happen that the choice of a subsequence is unnecessary even though uniqueness is not satisfied. Consider the example. dy = y1/3 dt (1)
There are an infinite number of solutions starting at the point C(0,0) which exist on I = [0,1]. For any c, 0 c 1, the function c defined by
0 c (t) = 2(t c) 3/ 2 3 0 tc c < t 1
(2)
is a solution of (1) on I. If the construction of -approximation theorem is applied to equation (1), one finds that the only polygonal path starting at the point C(0,0) is 1. This shows that this method cannot, in general, gives all solution of (1)
Theorem 1.4 Let f C on a domain D in the (t, y) plane, and suppose (t0, y0) is any
point in D. Then there exists a solution of dy = f(t, y) dt for (t, y) D, y(t 0 ) = y0 (1)
Since domain D is open, there exists an r > 0 such that all points whose
distance from C(t0, y0) is less than r, are contained in D. Let R be any closed rectangle containing C(t0, y0) and let R is contained in this open circle of radius r. Then required result is obtained by applying Cauchy-Peano Existence Theorem on (1).
180
Summary
In this chapter, first of all the reader is made familiar with certain basic concepts of differential equations and real function theory. Existence theorems namely Cauchy-Euler construction theorem and Cauchy-Peano Existence theorem are proved to show the existence of solutions of initial value problems. A relation between initial value problem and Volterra Integral equation is established with suitable illustrations.
Keywords
181
Written by: Sunita Pannu Vetted by: Kapil Kumar
Chapter-2
Objectives
This chapter is concerned with a very important property of ordinary differential equations i.e. existence and uniqueness of solutions. Picards method of successive approximations, which apart from being a mere numerical technique to approximate solutions has far reaching theoretical implications as well, is applied, to obtain approximate solutions of initial value problems.
Introduction
In the previous chapter, we did not device a general method which can assert theoretically the existence and uniqueness of solutions of a wider class of first order ordinary differential equations. This chapter discusses in detail the approximation method of Picard to the solution of the initial value problem of the general first order non-linear differential equations of the type.
dy dx = f ( x, y ),
y (x0) = y0,
(1)
where f(x,y) is some arbitrary function defined and continuous in some neighbourhood of (x0, y0). This method is also useful even when exact solutions are available, especially, if the exact solution is quite involved and we are only interested in the numerical value of the solution function at different points. The Picards theorem gives the unique solution of the above initial value problem (1) by the method of successive approximations, using the integral equation equivalent to the given non-linear differential equation.
182
Theorem 2.1 The Existence and Uniqueness Theorem (Picard-Lindelof
Theorem) Hypothesis
1.
Let D be a domain of the xy plane, and let f be a real function satisfying the following two requirements.
(i) (ii)
f is continuous in D; f satisfies a Lipschitz condition (with respect to y) in D; that is, there exists a constant k > 0 such that
|f(x,y1) f(x, y2) | k| y1 y2|
(1)
for all 2.
Let (x0, y0) be an interior point of D; let a and b be such that the rectangle R: |x - x0| a, | y y0| b, lies in D; let M = max |f(x,y)| for (x, y) R, and let h = min (a, b/M).
Conclusion
(2)
Proof We shall prove this theorem by the method of successive approximations. Let
1 ( x) = y0 + f [t , y0 ] dt ,
x0
2 ( x) = y0 + f [t , 1 (t )] dt ,
x0
3 ( x) = y0 + f [t , 2 (t )] dt ,
x0
(3)
183
. . .
n ( x) = y0 + f [t , n 1 (t )] dt.
x0
b b b smaller interval < a so that h = min a, = M M M |x - x0| h < a associated with the smaller rectangle R1 defined by |x - x0| h < a, | y y0| b.
We shall divide the proof into five main steps. 1. The functions {n} defined by (3) actually exist, have continuous derivatives, and satisfy the inequality |n (x) y0| b on |x x0| h; and thus f[x, n(x)] is defined on this interval. 2. The functions {n} satisfy the inequality |n (x) - n-1(x)| 3. M (kh) n . on x - x 0 h k n!
4.
184
5. This function is the only differentiable function on
x x0 h which
satisfies the differential equation dy/dx=f(x,y) and is such that (x0) = y0. Throughout the entire proof we shall consider the interval [x0, x0 + h]; similar arguments hold for the interval [x0 h, x0]. 1. We shall prove the first step by using mathematical induction. Assume that n-1 exists, has a continuous derivative, and is such that |n-1(x) y0| b for all x such that x0 x x0 + h. Thus [x, n-1(x)] lies in the rectangle R and so f [x, n-1(x)] is defined and continuous and satisfies |f [x, n-1(x)]| M on [x0, x0 + h]. Since n(x) = y0 +
x0
f [t , n 1 (t )]dt ,
we see that n also exists and has a continuous derivative on [x0, x0+h]. Also, |n(x) - y0| = | f [t , n 1 (t )]dt |
x0 x
f [t , n 1 (t )] dt M dt
x0 x0
= M(x x0) Mh b. Thus, [x, n(x)] lies in R and hence f[x, n (x)] is defined and continuous on [x0, x0 + h]. Clearly 1 defined by 1(x) = y0 +
x0
f [t , y0 ]dt
exists and has a continuous derivative on this interval. Also, |1(x) - y0|
x0
f [t,y0 ] dt M(x - x 0 ) b
and so f[x, 1(x)] is defined and continuous on the interval under consideration. Thus, by mathematical induction, each function n of the sequence (3) possesses these desired properties on [x0, x0+ h]. 2. In this step we again employ mathematical induction. We assume that |n-1(x) - n-2(x)|
Mk n 2 ( x x0 ) n 1 on [x0, x0 + h]. (n 1)!
(4)
185
Then |n(x) - n-1(x)| =
x0
{ f [t , n 1 (t )] f [t , n 2 (t )]}dt
x
f [t , n 1 (t )] f [t , n 2 (t )] dt
x0
Since by step 1, |n(x) y0| b for all n on [x0, x0 +h], using the Lipschitz condition (1) we have | f [t, n-1(t)] f [t, n-2(t)]| k |n-1(t) - n-2(t)| Thus |n(x) - n-1(x)| k n 1 (t ) n 2 (t ) dt
x0 x
Mk n 1 x (t x0 ) n 1 dt = x (n 1)! 0
Mk n 1 (t x0 ) n x Mk n 1 ( x x0 ) n = (n 1)! n x0 n!
Mk n 1 ( x x0 ) n , n!
(5)
which is precisely inequality (4) with (n-1) replaced by n. When n =1, we have as in Step 1: |1(x) y0| M(x - x0). This is inequality (5) when n = 1. Thus by induction the inequality (5) is satisfied on [x0, x0 + h] for all n. Since Mk n 1 Mk n 1 n M (kh) n ( x x0 ) n h = , n! n! k n!
186
we have |n(x) - n-1(x)| M (kh) n k n! (6)
M kh [e 1] . k
[ ( x)
i =1 i
i 1
( x)]
is such that (6) is satisfied for all x on the interval x0 x x0 + h, for each n = 1, 2, 3 Thus by the Weierstrass M test, the series y0 + [i ( x) i 1 ( x)]
i =1 n
converges uniformly on [x0, x0 + h]. Therefore its sequence of partial sums {Sn} converges uniformly to a limit function on [x0, x0 + h]. But Sn(x) = y0 + [i ( x) i 1 ( x)] = n ( x).
i =1 n
In other words, the sequence n converges uniformly to on [x0, x0 + h]. Thus, each n is continuous for [x0, x0 + h]. Theorem A shows that the limit function is also continuous on [x0, x0 + h]. 4. Since each n satisfies |n(x) y0| b on [x0, x0 + h], we also have |(x) y0| b on [x0, x0 + h]. Thus f [x, (x)] is defined on this interval and we can further apply the Lipschitz condition (1) and obtain | f [x, (x)] f [x, n(x)] | k |(x) - n(x)| for x [x0, x0 + h]. By step 3, given > 0, there exists N > 0 such that n(x)| < /k for all n > N and all x on [x0, x0 + h]. Thus (7) |(x) -
187
k|(x) - n(x)| < k = k (8)
for all n > N and all x on the interval under consideration. Thus from (7) and (8) we see that the sequence of functions defined by f [x, n(x)] for n = 1, 2, 3, .. converges uniformly to the function defined by f [x, (x)] on [x0, x0 + h]. Also, each function defined by f [x, n(x)] for n = 1, 2, 3, is continuous on this interval. Thus theorem B applies and (x) = lim n +1 ( x) = y0 + lim f [t , n (t )] dt
n n x0 x
= y0 + lim f [t , n (t )] dt
x0 n
= y0 + f [t , (t )] dt.
x0
x0
f [t , (t )] dt
on [x0, x0 +h]. Thus by the basic lemma, the limit function satisfies the differential equation dy/dx = f(x, y) on [x0, x0 +h] and is such that (x0) = y0. We have thus proved the existence of solution of the basic initial value problem (2) on the interval [x0, x0 +h]. 5. We now prove that the solution is unique. Assume that is another differentiable function defined on [x0, x0 + h] such that d = f [ x, ( x)] dx and (x0) = y0. Then certainly |(x) y0| < b (9)
on some interval [x0, x0 +]. Let x1 be such that |(x) y0| < b for x0 x < x1 and |(x1) y0| = b. Suppose x1 < x0 + h. Then
M1 =
( x1 ) y0
x1 x0
b b > M x1 x0 h
188
But by the mean value theorem there exists , where x0 < < x1, such that M1 = |'()|= |f [, ()]| M, a contradiction. Thus x1 x0 + h and the inequality (9) holds for x0 x < x0 + h, and so |(x) y0| b on the interval x0 x x0 +h. Since is a solution of dy/dx =f(x,y) on [x0, x0+h] such that (x0) = y0, from the basic lemma we see the satisfies the integral equation. (x) = y0 + (10)
x0
f [t , (t )]dt
(11)
on [x0, x0 + h]. We shall now prove by mathematical induction that k nb( x x0 ) n |(x) - n(x)| n! on [x0, x0 + h]. We assume that k n 1b( x x0 ) n 1 |(x) - n-1(x)| ( n 1)! on [x0, x0 + h]. Then from (3) and (11) we have |(x) - n(x)| = (13) (12)
x0
{ f [t , (t )] f [t , n 1 (t )]}dt
f [t , (t )] f [t , n 1 (t ) ] dt
x0
k n b (t x0 ) n x k n b( x x0 ) n = n! (n 1)! n x0
189
which is (13) with (n-1) replaced by n. When n = 1, we have |(x) - 1(x)| f [t , (t )] f [t , y0 ] dt
x0 x
k (t ) y0 dt kb( x x0 ) ,
x0
which is (12) for n = 1. Thus by induction the inequality (12) holds for all n on [x0, x0 + h]. Hence we have. |(x) - n(x)| b for n = 1, 2, 3. on [x0, x0 +h]. Now the series (kh) n n! (14)
b
n=0
(x) = lim n(x) on [x0, x0 +h]. But (x) = lim n(x) on this interval Thus,
n
(x) = (x) on [x0, x0 +h]. Thus the solution of the basic initial value problem is unique on [x0, x0 +h]. We have thus proved that the basic initial value problem has a unique solution on [x0, x0 +h], we can carry through similar arguments on the interval [x0-h, x0]. We thus conclude that the differential equation dy/dx = f(x,y) has a unique solution such that (x0) = y0 on |x x0| h.
The Method of Successive Approximation or Picard Iteration Method
This is very useful method to deduce the existence of solution. We know that is a solution of IVP x' = f(t, x) on |t | s.t. () = (E)
190
Definition
The successive approximations for (E) are defined to be the functions i.e. a sequence of functions 0(t), 1(t), (called successive approximations) as follows 0(t) = and i.e. k+1(t) = + f (s, k (s)) ds , k = 0, 1, 2, .;
|t |
t t
= + f(s, ) ds
Explanation
If () = , let us define the constant function 0(t) = . Though this constant function satisfies the initial condition, it does not in general satisfies the integral equation. But if we find 1(t) = +
f(s, (s)) ds , then 1(t) may be a little more closer to (t). In a
0 t
similar manner, we can find 2(t), 3(t), Continuing this process successively, we can obtain
191
n(t) = + f(s, n-1 (s)) ds
Now the crux of the Picard Theorem is that n(t) (t) as n giving the unique solution of the IVP.
Note: If one solves the given linear equation, the Picard Successive Approximations
In geometrical language, we are to devise a method for constructing a function x = x(t) whose graph passes through the point (, ) and that satisfies the differential equation x' = f (t, x) in some neighborhood of .
Process of Iteration
We begin with a crude approximation to a solution and prove it step by step by applying a repeatable operation, which will bring us as close as we please to an exact solution. In the integral equation x(t) = + f (s, x(s)) ds,
(1)
the dummy variable s is used to avoid confusion with the variable upper limit t on the integral I. A rough approximation to a solution is given by the constant 0(t) = , which is simply a horizontal straight line through the point (, ). We insert this approximation in the right side of equation (1) in order to obtain a new and perhaps better approximation x1(t) as follows. x1(t) = + f (s, x 0 (s)) ds
The next step is to use x1(t) to generate another and perhaps even better approximation x2(t) in the same way x2(t) = + f(s, x1 (s)) ds
192
Solution of Initial Value Problems by Picard method
dx = x , x(0) = 1. dt
x(s) ds
0
which is (Volterra) integral equation of 2nd kind. Let 0(t) = 1, then by Picards method 1(t) = 1 + 0 (s) ds = 1 + 1 ds = 1 + t
0 0 t t t t
2(t) = 1 + 1 (s) ds = 1 + (1 + s ) ds
0 0
=1+t+
t
t2 2!
t
3(t) = 1 + 2 (s) ds = 1 + (1 + s +
0 0
s2 ) ds 2!
=1+t+
t2 t3 + 2! 3!
t2 t3 tn + + .+ 2! 3! n!
(t) = et is the unique solution of the given IVP, by the Picards method of successive approximations.
193
2. Sol
dx = t2 x , x(0) = 1 dt The corresponding integral equation is x(t) = 1 + Picard iterates are 0(t) = 1, 1(t) = 1 + s 2 . 1 ds = 1 +
0 t t
s
0
x(s) ds
t3 3
2
t3 1 t3 s3 2(t) = 1 + s (1+ ) ds = 1 + + 3 2! 3 3 0
2
2 s3 1 s3 3(t) = 1 + s 1+ + ds 3 2! 3 0 t 2 2 3
t3 1 t3 1 t3 + =1+ + 3 2! 3 3! 3
t3 1 t3 1 t3 + .. + =1+ + 3 2! 3 n! 3
n
The exact solution is obtained by taking limit n , i.e. lim n (t) = (t) = e t 3 , to which the above approximate solution converges.
3. Sol
3
dx = t(x - t2 + 2) , x(0) = 1 dt The integral equation, equivalent to the above IVP is x(t) = 1 +
s [x(s) s
0
+ 2] ds
194
The approximate solutions are
0(t) = 1, 1(t) = 1 + s (1 - s 2 + 2) ds
0
3t 2 t 4 = 1 + s (3 - s ) ds = 1 + 2 4 0
2 t 3s 2 s 4 2 2(t) = 1 + s 1+ - s + 2 ds 2 4 0
s2 s4 = 1 + s 3 + ds 2 4 0
t
=1+
3t 2 t 4 t6 + 2 8 24
t
s2 s 4 s6 3(t) = 1 + s 3 + + ds 2 8 24 0
3t 2 t 4 t6 t8 + + , 2 8 48 192
(t) = t2 +(1 +
i.e.
4. Sol
(t) = t 2 + e t
dx = t x , x(0) = 1 dt We write f (t, x) = t x and the integral equation corresponding to the initial
t x(t) dt or = 1 +
0
s x(s) ds
0
0(t) = 1
195
n(t) = 1 +
Thus
t
0
n-1
(t) dt
for n = 1, 2, 3,
0(t) = 1 1(t) = 1 + 1. s ds = 1 +
0 t t
t2 2
s2 2(t) = 1 + s 1+ ds = 1 + 2 0
s3 s + ds 2
=1+
t2 t4 + 2 2.4
(A)
t2 t2 1 t2 1 n-1(t) = 1 + + + .. + 2 2! 2 (n-1)! 2
t
n-1
Then
=1+
t2 1 t2 1 t2 1 t2 n(t) = 1 + + + + .. + n! 2 2 2! 2 3! 2
196
5. Sol
x(s)
1
ds
for n = 1, 2, 3,
We find
0(t) = 1 1(t) = 1 + ds = t
1 t t
2(t) = 1 + s ds = 1 +
1
[(s - 1) + 1] ds
1
Here, it is convenient to have integrand occurring in the successive approximations in powers of (s 1) rather than in powers of s (since t0 = 1 and not zero).
2 s - 1) ( 2(t) = 1 + s + 2 1
= 1 + (t 1) +
t
(t - 1) 2 2!
3(t) = 1 +
(s - 1) 2 1 + (s 1) + ds 2
(t - 1) 2 (t - 1)3 + 2! 3!
197
n(t) = 1 + (t 1) + (t - 1) 2 (t - 1)3 (t - 1) n + + .. + 2! 3! n! (A)
We note that n(t) is the partial sum of the first (n+1) terms of the infinite series expansion of the function (t) = e t-1 . Moreover, this series converges for all real t. n(t) (t) = e t-1 t.
Hence, the function (t), is the required solution of the given problem.
6.
Find the first four approximations of the initial value problem x'(t) = 1 + tx, x(0) = 1. Let us find the first approximation as x0(t) = x(0) =1. The second approximation is x1(t) = x0 + f (s, x 0 (s)) ds = x 0 + [1 + s x 0 (s)] ds .
t0
0
t2 . 2
t
(1)
Now
(2)
= 1+t+
t2 t3 t4 + + 2 3 8
198
Now x3(t) = x0 + f (s, x 2 (s)) ds = x 0 + [1 + s x 2 (s)] ds
t0
0
(3)
s2 s3 s4 1 + s + + + ds 2 3 8
= 1+t+
7.
t2 t3 t4 t5 t6 + + + + 2 3 8 15 48
Find the first four Picard successive approximations of the initial value problem x'(t) = x + t, x(0) = 1 and find the n-th approximation xn(t). Find the limit of the sequence xn(t). The given equation is equivalent to the integral equation x(t) = 1 + (s + x(s)) ds
0 t
Let us first define the first approximation x0(t) = x(0) = 1. The successive approximations are given by x1(t) = 1 + (s + 1) ds = 1 + t +
0 t s2 x2(t) = 1 + s + 1 + s + ds 2! 0 t
t2 2!
= 1 + t + t2 +
s + 0
t
t3 3!
x3(t) = 1 +
s3 2 1 + s + s + ds 3!
2 t3 t 4 = 1 + t + t + + 3 4!
199
x4(t) = 1 +
s + 0
t
s3 s 4 2 1 + s + s + + ds 3 4!
2 t3 t3 t5 = 1 + t + t + + + 3 3.4 5!
since 0 < t < 1 Hence x(t) = 2et t 1 which can be easily seen to be the exact solution of the given differential equation.
Note 1. If one solves the given linear equation, the Picard successive approximations
initial approximation other than a constant function also. In general we cannot say that such a sequence of approximation will converge to a solution of the initial value problem.
Exercise
Use the method of successive approximations to find the first three members
1, 2, 3 of a sequence of functions that approaches the exact solution of the problem.
1.
dy = xy, dx dy = x + y2 , dx dy = ex + y2 , dx
y(0) = 1
2.
dy = x + y, dx dy = 1 + xy 2 , dx dy = sin x + y 2 , dx
y(0) = 1
3.
y(0) = 0
4.
y(0) = 0
5.
y(0) = 0
6.
y(0) = 0
200
Summary
Existence and uniqueness theorem is the tool which makes it possible for us to conclude that there exists only one solution to a first order differential equation which satisfies a given initial condition, provided, the function in given differential equation satisfies the Lipschitz condition. Also, we present a method due to E. Picard, which gives approximate solution curves of a differential equation passing through a given point.
Keywords
201
Written by: Sunita Pannu Vetted by: Kapil Kumar
Chapter-3
Objectives
In many cases, the existence of solution is guaranteed by the theorems, but no method for obtaining those solutions in explicit and closed form is known. The main objective of this chapter is to gain ability to solve first order ordinary differential equations by using some approximate methods.
Introduction
The graphical methods of the preceding section for solving initial value problems are very general but they suffer from several serious disadvantages. Not only are they tedious and subject to possible errors of construction, but they merely provide us with the approximate graphs of the solutions and do not furnish any analytical expression for these solutions. Though we have studied an approximate method in the form of Picards method in Chapter 2, it involves evaluation of integrals at each step which may not be easy in certain cases. The advantage of the methods described in this chapter over the earlier methods, will lie in the fact that in getting the approximate numerical values of the unknown functions we shall only require the numerical values of the functions appearing as coefficients in the differential equation, apart from mere elementary arithmetical computations. We present some approximate methods of solving first order equations.
Power Series Methods
In order to explain the power series methods we shall assume that power series solutions actually do exist. We consider the initial value problem consisting of the differential equation
dy = f ( x, y ) dx
(1)
202
and the initial condition y(x0) = y0 (2)
and assume that the differential equation (1) possesses a solution that is representable as a power series in powers of (x x0). That is, we assume that the differential equation (1) has a solution of the form y = c0 + c1(x x0) + c2 (x x0)2 + =
c (x x )
n=0 n 0
(3)
that is valid in some interval about the point x0. We now consider methods of determining the coefficients c0, c1, c2,. in (3) so that the series (3) actually does satisfy the differential equation (1)
A. The Taylor Series Method
We thus assume that the initial value problem consisting of the differential equation (1) and the initial condition (2) has a solution of the form (3) that is valid in some interval about x0. Then by Taylors theorem, for each x in this interval the value y(x) of this solution is given by
y ( x) = y ( x0 ) + y '( x0 )( x x0 ) +
=
y ''( x0 ) ( x x0 ) 2 + ... 2!
n=0
y ( n ) ( x0 ) ( x x0 ) n n!
(4)
From the initial condition (2) we have y(x0) = y0 and from the differential equation (1) y '(x0) = f(x0, y0) Substituting these values of y(x0) and y'(x0) into the series in (4) we obtain the first two coefficients of the desired series solution (3). Now differentiating the differential equation (1), we obtain
d2y d dy = [ f ( x, y )] = f x ( x, y ) + f y ( x, y ) 2 dx dx dx
= f x ( x, y ) + f y ( x, y ) f ( x, y ).
(5)
203
where we use subscripts to denote partial differentiations. From this we get
y ''( x0 ) = f x ( x0 , y0 ) + f y ( x0 , y0 ) f ( x0 , y0 )
Substituting this value of y''(x0) into (4) we obtain the third coefficient in the series solution (3). Proceeding in like manner, we differentiate (5) successively to obtain
d3y d4y dny , ....., ....... dx3 dx 4 dx n
From these we obtain the values y''' (x0), y(iv) (x0).., y(n)(x0),. Substituting these values into (4) we obtain the fourth and following coefficients in the series solution (3).
Example
Use the Taylor series method to obtain a power series solution of the initial value problem
dy = x2 + y 2 , dx
(1) (2)
y (0) = 1 in powers of x.
Solution
Since we seek a solution in powers of x, we set x0 = 0 and thus assume a solution of the form y = c0 + c1 x + c2 x 2 + ... = cn x n
n=0
By Taylors theorem, we know that for each x in the interval where this solution is valid
y ''(0) 2 y ( x) = y (0) + y '(0) x + x + ... = 2!
n=0
y ( n ) (0) n x n!
(3)
Using the initial condition (2) in the differential equation (1) we see that
204
y'(0) = 02 + 12 = 1 Differentiating (1) successively, we obtain d2 y dy = 2x + 2y 2 dx dx d3 y d2 y dy = 2 + 2y + 2 3 2 dx dx dx d4 y d3 y dy d 2 y = 2y + 6 dx 4 dx 3 dx dx 2 Substituting x = 0, y = 1, dy =1 dx into (5), we obtain (8)
2
(4)
(5)
(6)
(7)
y'''(0) = 2+2(1)(2) + 2(1)2 = 8. Finally, substituting y = 1, dy d2y d3y = 1, 2 = 2, = 8 into (7), we find that dx dx dx 3
(9)
y(iv)(0) = (2)(1)(8) + (6)(1)(2) = 28. By successive differentiation of (7), we could proceed to determine d5y d6y , ,......., dx 5 dx 6 and hence obtain y(v)(0), y(vi)(0),.....
(10)
Substituting the values given by (2), (4), (8), (9) and (10) into (3), we obtain the first five coefficients of the desired series solution. We thus find the solution y=1+x+ 2 2 8 3 28 4 x + x + x + ... 2! 3! 4! 4 3 7 4 x + x + ... 3 6
= 1 + x + x2 +
205
B. The Method of Undetermined Coefficients
We consider an alternative method for obtaining the coefficients c0, c1, c2,... in the assumed series solution y = c0 + c1(x x0) + c2 (x x0)2 + .... = of the problem consisting of differential equation dy = f ( x, y ) dx and the initial condition y(x0) = y0 . (3) (2)
c
n=0
( x x0 ) n
(1)
We shall refer to this alternative method as the method of undetermined coefficients. In order to apply it we assume that f(x,y) in the differential equation (2) can be represented in the form f(x,y) = a00 + a10(x - x0) + a01(y-y0) + a20( x - x0)2 + a11(x x0) (y y0) + a02 (y y0)2+... (4)
The coefficients aij in (4) may be found by Taylors formula for functions of two variables. Using the representation (4) for f(x,y) the differential equation (2) takes the form dy = a00 + a10 ( x x0 ) + a01 ( y y0 ) + a20 ( x x0 ) 2 dx +a11(x x0) (y y0) + a02 (y y0)2+... (5)
Now assuming that the series (1) converges in some interval |x x0| < r (r > 0) about x0, we may differentiate (1) term by term and the resulting series will also converge on |x x0| < r and represent y'(x) there. Doing this we thus obtain dy = c1 + 2c2 ( x x0 ) + 3c3 ( x x0 ) 2 + ... dx (6)
We note that in order for the series (1) to satisfy the initial condition (2) that y = y0 at x = x0, we must have c0 = y0 and hence
206
y y0 = c1(x x0) + c2 (x x0)2 + ... (7)
Now substituting (1) and (6) into the differential equation (5), and making use of (7), we obtain c1+2c2 (x - x0) + 3c3 (x x0)2 + = a00 + a10 (x - x0) + a01 [c1 (x - x0) + c2 (x - x0)2 + ...] + a20 (x - x0)2 + a11 (x - x0) [c1 (x - x0) + c2 (x - x0)2 + ...] + a02 [c1 (x - x0) + c2 (x - x0)2 + ...]2 + ... (8)
Performing the multiplications and then combining like powers of (x x0), we see that (8) takes the form c1+2c2 (x - x0) + 3c3 (x x0)2 + . = a00 + (a10 + a01c1) (x x0)
2 (x x0)2 + ... . + (a01c2 + a20 + a11c1 + a02 c1
(9)
In order that (9) be satisfied for all values of x in the interval |x x0| < r, the coefficients of like powers of (x x0) on both side of (9) must be equal. Equating these coefficients, we obtain c1 = a00, 2c2 = a10 + a01c1,
2 , 3c3 = a01c2 + a20 + a11c1 + a02 c1
(10)
From the conditions (10) we determine successively the coefficients c1, c2, c3... of the series solution (1). From the first condition, we obtain c1 as the known coefficient a00. Then from the second condition, we obtain c2 in terms of the known coefficients a10 and a01 and the coefficient c1 just determined. Thus we obtain c2 = 1 (a10 + a01a00 ). Similarly, we proceed to determine c3, c4.... We observe that in 2
general each coefficient cn is thus given in terms of the known coefficients aij in the expansion (4) and the previously determined coefficients c1, c2, .....cn-1 .
207
Finally, we substitute the coefficients c0, c1, c2,..... so determined into the series (1) and thereby obtain the desired solution.
Example
Use the method of undetermined coefficients to obtain a power series solution of the initial value problem. dy = x2 + y 2 dx y(0) = 1, (2) in powers of x .
Solution
(1)
Since x0 = 0, the assumed solution is of the form y = c0 + c1x + c2x2 + c3x3 + ... (3)
In order to satisfy the initial condition (2) we must have c0 = 1 and hence the series (3) takes the form y = 1 + c1x + c2x2 + c3x3 + .... Differentiating (4) we obtain dy = c1 + 2c2 x + 3c3 x 2 + 4c4 x3 + ... dx (5) (4)
For the differential equation (1) we have f(x, y) = x2 + y2. Since x0 = 0 and y0 = 1, we must expand x2 + y2 in the form
i , j =0
a x ( y 1) .
i j ij
Since
y2 = (y -1)2 +2 (y 1) + 1,
the desired expansion is given by x2 + y2 = 1 + 2 (y 1) + x2 + (y 1)2. Thus the differential equation (1) takes the form
208
dy = 1 + 2( y 1) + x 2 + ( y 1) 2 . dx Now substituting (4) and (5) into the differential equation (6), we obtain c1 + 2c2 x + 3c3x2 + 4c4x3 + .... = 1 + 2 (c1x + c2x2 + c3x3 + ...) + x2 + (c1x + c2x2 + ...)2. (7) Collecting like powers of x in the right hand side of (7), it takes the form c1 + 2c2x + 3c3x2 + 4c4x3
2 ) x2 + (2c3 + 2c1c2)x3+... = 1 + 2c1x + (2c2 + 1 + c1
(6)
(8)
Equating the coefficients of the like powers of x in (8), we obtain the conditions c1 = 1 2c2 = 2c1
2 , 3c3 = 2c2 + 1 + c1
4c4 = 2c3 + 2c1c2, From the conditions (9) we obtain successively c1 = 1 c2 = c1 = 1, 1 4 c3 = (2c2 + 1 + c12 ) = , 3 3 c4 = 1 1 14 7 (2c3 + 2c1c2 ) = = 4 4 3 6
(9)
(10)
Substituting these coefficients into the series (4), we obtain the first five terms of the desired series solution. We thus find y = 1 + x + x2 + 4 3 7 4 x + x + ... 3 6
We note that this is of course the same series previously obtained by the Taylor series method.
209
Exercise
Obtain a power series solution in power of x of each of the initial value problems in questions 1-5 by (a) the Taylor series method and (b) the method of undetermined coefficients. 1. 3. 5. dy = x + y, dx dy = 1 + xy 2 , dx y(0) = 1 y(0) = 2 2. 4. dy = x2 + 2 y 2 , dx dy = x3 + y 3 , dx y(0) = 4 y(0) = 3
dy = x + sin y, y(0) = 0 dx
Obtain a power series solution in powers of x 1 of each of the initial value problems in questions 6-9 by (a) the Taylor series method and (b) method of undetermined coefficients. 6. 8. dy = x2 + y 2 , dx y(1) = 4 7. 9. dy = x3 + y 2 , dx dy = x + cos y, dx y(1) = 1 y(1) =
dy = x + y + y 2 , y(1) = 1 dx
In this section we introduce certain basic numerical methods for approximating the solution of the initial value problem consisting of the differential equation. dy = f ( x, y ) dx and the initial condition y(x0) = y0 . (2) (1)
Numerical methods employ the differential equation (1) and the condition (2) to obtain approximations to the values of the solution corresponding to various, selected values of x. To be more specific, let y denote the solution of the problem and let h denote a positive increment in x. The initial condition tells us that y = y0 at x = x0. A numerical method will employ the differential equation (1) and the condition (2) to approximate successively the values of y at x1 = x0 + h, x2 = x1 + h, x3 = x2 + h, ....
210
A. The Euler Method
The Euler method is very simple but not very practical. Let y denote the exact solution of the initial value problem that consists of the differential equation. dy = f ( x, y ) dx and the initial condition y(x0) = y0. (1) (2)
x0
x1
f ( x, y )dx =
x0
dx dx = y( x ) y( x )
1 0
dy
Since y0 denotes the value y(x0) of the exact solution y at x0, we have
x1
y(x1) = y0 +
x0
f ( x, y)dx .
(3)
If we assume that f(x, y) varies slowly on the interval x0 x x1, then we can approximate f(x, y) in (3) by its value f(x0, y0) at the left endpoint x0. Then y ( x1 ) y0 + f ( x0 , y0 )dx
x0 x1 x1
But Thus
x0
f ( x , y )dx = f ( x , y )( x x ) = hf ( x , y ).
0 0 0 0 1 0 0 0
y ( x1 ) y0 + hf ( x0 , y0 ).
Thus we obtain the approximate value y1 of y at x1 = x0 + h by the formula y1 = y0 + hf (x0, y0). (4)
Having obtained y1, we proceed in like manner to obtain y2 by the formula y2 = y1 + hf(x1, y1), y3 by the formula y3 = y2 + hf(x2, y2), and so forth. In general we find yn+1 in terms of yn by the formula yn+1 = yn + hf(xn, yn). (5)
The graph of the exact solution y is a curve C in the xy plane (see figure) Let P denotes the initial point (x0, y0) and let T be the tangent to C at P. Let Q be the point at which the line x = x1 intersects C and let N be the point at which this line intersects T. Then the exact value of y at x1 is represented by LQ. The approximate value y1 is
211
represented by LN, since LN = LM + MN = y0 + PM tan = y0 + hf(x0, y0). The error in approximating the exact value of y at x1 by y1 is thus represented by NQ. The figure suggests that if h is sufficiently small, then this error NQ will also be small and hence the approximation will be good.
Figure
But if the increment h is very small, then the computations will be more lengthy and so the method will involve tedious and time consuming labour. Thus, this method is not very practical.
Example
Apply the Euler method to the initial value problem dy = 2x + y dx y(0) = 1 (1) (2)
Employ the method to approximate the value of the solution y at x = 0.2, 0.4, 0.6, 0.8, and 1.0 using (1) h = 0.2, and (2) h = 0.1. Obtain results to three figures after the decimal point. Compare with the exact value obtained.
212
Solution 1
have x0 = 0, y0 = 1. We now proceed with the calculations. (a) x1 = x0 + h = 0.2, f(x0, y0) = f(0, 1) = 1.000, y1 = y0 + hf(x0, y0) = 1.000 + 0.2 (1.000) = 1.200 (b) x2 = x1 + h = 0.4, f(x1, y1) = f(0.2, 1.200) = 1.600, y2 = y1 + hf(x1, y1) = 1.200 + 0.2 (1.600) = 1.520. (c) x3 = x2 + h = 0.6, f(x2, y2) = f(0.4, 1.520) = 2.320, y3 = y2 + hf(x2, y2) = 1.520 + 0.2 (2.320) = 1.984. (d) x4 = x3 + h = 0.8, f(x3, y3) = f(0.6, 1.984) = 3.184, y4 = y3 + hf(x3, y3) = 1.984 + 0.2 (3.184) = 2.621. (e) x5 = x4 + h = 1.0, f(x4, y4) = f(0.8, 2.621) = 4.221, y5 = y4 + hf(x4, y4) = 2.621 + 0.2 (4.221) = 3.465. These results, corresponding to the various values of xn, are collected in the second column of Table 1.
Table 1 xn yn using h = 0.2 yn using h = 0.1 y
0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0
1.000 1.100 1.230 1.393 1.592 1.831 2.114 2.445 2.830 3.273 3.780
1.000 1.116 1.264 1.450 1.675 1.946 2.266 2.641 3.076 3.579 4.155
213
2. In this case we have f(x, y) = 2x + y and h = 0.1. Again we have x0 = 0, y0 = 1. The calculations are as follows: (a) x1 = x0 + h = 0.1, f(x0, y0) = f(0, 1) = 1.000, y1 = y0 + hf(x0, y0) = 1.000 + 0.1 (1.000) = 1.100 (b) x2 = x1 + h = 0.2, f(x1, y1) = f(0.1, 1.100) = 1.300, y2 = y1 + hf(x1, y1) = 1.100 + 0.1 (1.300) = 1.230. (c) x3 = x2 + h = 0.3, f(x2, y2) = f(0.2, 1.230) = 1.630, y3 = y2 + hf(x2, y2) = 1.230 + 0.1 (1.630) = 1.393. (d) x4 = x3 + h = 0.4, f(x3, y3) = f(0.3, 1.393) = 1.993, y4 = y3 + hf(x3, y3) = 1.393 + 0.1 (1.993) = 1.592. (e) x5 = x4 + h = 0.5, f(x4, y4) = f(0.4, 1.592) = 2.392, y5 = y4 + hf(x4, y4) = 1.592+ 0.1 (2.392) = 1.831. (f) x6 = x5 + h = 0.6, f(x5, y5) = f(0.5, 1.831) = 2.831, y6 = y5 + hf(x5, y5) = 1.831 + 0.1 (2.831) = 2.114. (g) x7 = x6 + h = 0.7, f(x6, y6) = f(0.6, 2.114) = 3.314, y7 = y6 + hf(x6, y6) = 2.114+ 0.1 (3.314) = 2.445. (h) x8 = x7 + h = 0.8, f(x7, y7) = f(0.7, 2.445) = 3.845, y8 = y7 + hf(x7, y7) = 2.445 + 0.1 (3.845) = 2.830. (i) x9 = x8 + h = 0.9, f(x8, y8) = f(0.8, 2.830) = 4.430, y9 = y8 + hf(x8, y8) = 2.830 + 0.1 (4.430) = 3.273. (j) x10 = x9 + h = 1.0, f(x9, y9) = f(0.9, 3.273) = 5.073, y10 = y9 + hf(x9, y9) = 3.273 + 0.1 (5.073) = 3.780. These results are collected in the third column of Table 1. The values of the exact solution y, computed to three figures after the decimal point, are listed in the fourth column of Table 1. From this table we compute the errors involved in both approximations at x = 0.2, 0.4, 0.6, 0.8, and 1.0. These errors are tabulated in Table 2.
214
A study of these tables illustrates two important facts concerning the Euler method. First, for a fixed value of h, the error becomes greater and greater as we proceed over a larger and larger range away from the initial point. Second, for a fixed value of xn, the error is smaller if the value of h is smaller.
Table 2 xn Error using h = 0.2 Error using h = 0.1
Exercise
1.
Consider the initial value problem dy = x 2 y , y(0) = 1. dx (a) Apply the Euler method to approximate the values of the solution y at x = 0.1, 0.2, 0.3, and 0.4, using h = 0.1. Obtain results to three figures after the decimal point. (b) (c) Proceed as in part(a) using h = 0.05 Find the exact solution of the problem and determine its values at x = 0.1. 0.2, 0.3 and 0.4 (to three figures after the decimal point). (d) Compare the results obtained in parts (a), (b), and (c). Tabulate errors as in table 2.
2.
215
B. The Modified Euler Method
In Section A we observed that the value y(x1) of the exact solution y of the initial value problem dy = f ( x, y ), dx y(x0) = y0, at x1 = x0 + h is given by y(x1) = y0 + (1) (2)
x1
x0
f ( x, y )dx.
(3)
In the Euler method we approximated f(x, y) in (3) by its value f(x0, y0) at the left endpoint of the interval x0 x x1 and obtained the approximation y1 = y0 + hf(x0, y0) (4)
for y at x1. It seems that a more accurate value would be obtained if we were to approximate f(x, y) by the average of its values at the left and right endpoints of x0 x x1, instead of simply by its value at the left endpoint x0. This is essentially what is done in the modified Euler method. In order to approximate f(x, y) by the average of its values at x0 and x1, we need to know its value f[x1, y(x1)] of y at x1. We must find a first approximation y1(1) for y(x1), and to find this, we take y1(1) = y0 + hf (x0, y0) (5)
as the first approximation to the value of y at x1. Then we approximate f[x1, y(x1)] by f(x1, y1(1) ), using the value y1(1) found by (5). From this we obtain. f ( x0 , y0 ) + f ( x1 , y1(1) ) 2 (6)
which is approximately the average of the values of f(x, y) at the endpoints x0 and x1. We now replace f(x, y) in (3) by (6) and thereby obtain y1(2) = y0 + f ( x0 , y0 ) + f ( x1 , y1(1) ) h 2 (7)
216
We now use the second approximation y1(2) to obtain a second approximation f ( x1 , y1(2) ) for the value of f (x, y) at x1. From this we obtain y1(3) = y0 + f ( x0 , y0 ) + f ( x1 , y1(2) ) h 2 (8)
as the third approximation of the value of y at x1. Proceeding in this way we obtain a sequence of approximations y1(1) , y1(2) , y1(3) ,..... to the value of the exact solution y at x1. We proceed to compute the successive terms of the sequence until we obtain two consecutive members that have the same value to the number of decimal places required. Let that value of the solution y at x1 be denoted by y1. We now proceed to approximate y at x2 = x1 + h, in exactly the same way as we did in finding y1. We find successively
(1) y2 = y1 + hf ( x1 , y1 ), (1) f ( x1 , y1 ) + f ( x2 , y2 ) h, 2 (2) f ( x1 , y1 ) + f ( x2 , y2 ) h, 2
(2) y2 = y1 +
(3) y2 = y1 +
(9)
..
..
..
..
..
until two consecutive members of this sequence agree, thereby obtaining an approximation y2 to the value of y at x2. Proceeding further in like manner one obtains an approximation y3 to the value of y at x3, and so forth.
Example
Apply the modified Euler method to the initial value problem dy = 2 x + y, dx y(0) = 1. (1) (2)
Employ the method to approximate the value of the solution y at x = 0.2 and x = 0.4 using h = 0.2. Obtain results to three figures after the decimal point. Compare
217
with the results obtained using the basic Euler method with h = 0.1 and with the exact values.
Solution
Here f(x, y) = 2x + y, x0 = 0, and y0 = 1, and we are to use h = 0.2. We begin by approximating the value of y at x1 = x0 + h = 0.2. A first approximation y1(1) is found using Euler Method. Since f(x0, y0) = f(0, 1) = 1.000, we have y1(1) = y0 + hf ( x0 , y0 ) = 1.000 + 0.2(1.000) = 1.200. We now use modified Euler method to find a second approximation y1(2) to the desired value. Since f ( x1 , y1(1) ) = f (0.2,1.200) = 1.600, we have y1(2) = y0 + f ( x0 , y0 ) + f ( x1 , y1(1) ) 1.000 + 1.600 h = 1.000 + (0.2) = 1.260 2 2
We next find a third approximation y1(3) . Since f ( x1 , y1(2) ) = f (0.2,1.260) = 1.660, we find y1(3) = y0 + Similarly, we find y and y
(5) 1 (4) 1
Since the approximations y1(4) and y1(5) are the same to the number of decimal places required, we take their common value as the approximation y1 to the value of the solution y at x1 = 0.2. That is, we take y1 = 1.267 (3)
218
We now proceed to approximate the value of y at x2 = x1 + h = 0.4. We find successively
(1) y2 = y1 + hf ( x1 , y1 ) = 1.267 + 0.2(1.667) = 1.600 (1) f ( x1 , y1 ) + f ( x2 , y2 ) 1.667 + 2.400 = y1 + h = 1.267 + (0.2) = 1.674. 2 2 (2) f ( x1 , y1 ) + f ( x2 , y2 ) 1.667 + 2.474 h = 1.267 + (0.2) = 1.681. 2 2 (3) f ( x1 , y1 ) + f ( x2 , y2 ) 1.667 + 2.481 h = 1.267 + (0.2) = 1.682. 2 2 (4) f ( x1 , y1 ) + f ( x2 , y2 ) 1.667 + 2.482 h = 1.267 + (0.2) = 1.682. 2 2 (4) (5) y2 and y2 are both the same to the required
(2) 2
(3) y2 = y1 +
(4) y2 = y1 +
(5) y2 = y1 +
number of decimal places, we take their common values as the approximation y2 to the value of the solution y at x2 = 0.4. That is, we take y2 = 1.682 (4)
We compare the results (3) and (4) with those obtained using the basic Euler method with h = 0.1 and with the exact values. For this purpose the various results and the corresponding errors are listed in Table 3. The principal advantage of the modified Euler method over the basic Euler method is immediately apparent from a study of Table 3.
Table 3 xn Exact value of y (to three decimal places) Using basic Euler method with h = 0.1 Approximation Using modified Euler with h = 0.2 Approximation
yn 1.267 1.682
219
Exercise
1.
(a)
Apply the modified Euler method to approximate the values of the solution y at x = 0.1, 0.2 and 0.3 using h = 0.1. Obtain results to three figures after the decimal point.
(b) (c)
Proceed as in part (a) using h = 0.05 Find the exact solution of the problem and determine its value at x = 0.1, 0.2 and 0.3 (to three figures after the decimal point).
(d) 2.
Compare the results obtained in parts (a), (b) and (c) and the tabulate errors. Consider the initial value problem dy = x2 + y 2 , dx y(0) = 1
(a)
Apply the modified Euler method to approximate the values of the solution y at x = 0.1, 0.2 and 0.3 using h = 0.1. Obtain results to three figures after the decimal point.
(b)
Apply the Euler method to approximate the values of the solution y at x = 0.1, 0.2 and 0.3 using h = 0.1. Obtain results to three figures after the decimal point.
(c)
C.
Compare the results obtained in parts (a) and (b) and the tabulate errors.
The Runge-Kutta Method
We now consider the Runge-Kutta method for approximating the values of the solution of the initial value problem. dy = f ( x, y ) dx y(x0) = y0 (1) (2)
at x1 = x0 + h, x2 = x1 + h, and so forth. This method gives accurate results without the need of using extremely small values of the interval h.
220
To approximate the value of the solution of the initial value problem under consideration at x1 = x0 + h by the Runge-Kutta method, we calculate successively the coefficients k1, k2, k3, k4, and K0 defined by the formulas k1 = hf(x0, y0), h k k2 = hf x0 + , y0 + 1 , 2 2 h k k3 = hf x0 + , y0 + 2 , 2 2 k4 = hf ( x0 + h, y0 + k3 ), and Then we set 1 K 0 = (k1 + 2k2 + 2k3 + k4 ) . 6 y1 = y0 + K0 (4) (3)
and take this as the approximate value of the exact solution at x1 = x0 + h. We proceed to approximate the value of the solution at x2 = x1 + h in an exactly similar manner. Using x1 = x0 + h and y1 as determined by (4), we calculate successively the coefficients k1, k2, k3, k4 and K1 defined by k1 = hf(x1, y1), h k k2 = hf x1 + , y1 + 1 , 2 2 h k k3 = hf x1 + , y1 + 2 , 2 2 k4 = hf ( x1 + h, y1 + k3 ), and Then we set 1 K1 = (k1 + 2k2 + 2k3 + k4 ) . 6 y2 = y1 + K1 (6) (5)
and take this as the approximate value of the exact solution at x2 = x1 + h. Similarly, we proceed to approximate the value of the solution at x3 = x2 + h, x4 = x3 + h, and so forth. Let yn denotes the approximate value obtained for the solution at xn = x0 + nh, we calculate successively k1, k2, k3, k4, and Kn defined by
221
k1 = hf(xn, yn), h k k2 = hf xn + , yn + 1 , 2 2 h k k3 = hf xn + , yn + 2 , 2 2 k4 = hf ( xn + h, yn + k3 ), and Then we set 1 K n = (k1 + 2k2 + 2k3 + k4 ) . 6 yn+1 = yn + Kn
and take this as the approximate value of the exact solution at xn+1 = xn + h.
Example
Apply the Runge-Kutta method to the initial value problem dy = 2x + y dx y(0) = 1 (1) (2)
Employ the method to approximate the value of the solution y at x = 0.2 and x = 0.4 using h = 0.2. Carry the intermediate calculations in each step to five figures after the decimal point, and round off the final results of each step to four such places. Compare with the exact value.
Solution
Here f(x, y) = 2x + y, x0 = 0, y0 = 1, and we are to use h = 0.2. Using these quantities we calculate successively k1, k2, k3, k4, and K0. We first find k1 = hf(x0, y0) = 0.2 f(0, 1) = 0.2(1) = 0.20000 Then since x0 + and y0 + h 1 = 0 + (0.2) = 0.1 2 2 k1 1 = 1.00000 + (0.20000) = 1.10000, 2 2
222
we find h k k2 = hf x0 + , y0 + 1 = 0.2 f (0.1, 1.10000) 2 2 = 0.2 (1.30000) = 0.26000. Next, since y0 + we find h k k3 = hf x0 + , y0 + 2 = 0.2 f (0.1, 1.13000) 2 2 = 0.2(1.33000) = 0.26600. Since x0 + h = 0.2 and y0 + k3 = 1.00000 + 0.26600 = 1.26600, we obtain k4 = hf(x0 + h, y0 + k3) = 0.2f (0.2, 1.26600) = 0.2 (1.66600) = 0.33320. Finally, we find 1 1 K 0 = (k1 + 2k2 + 2k3 + k4 ) = (0.20000 + 0.52000 + 0.53200 + 0.33320) 6 6 = 0.26420. Then the approximate value of the solution at x1 = 0.2 is y1 = 1 + 0.2642 = 1.2642 (3) k2 1 = 1.00000 + (0.26000) = 1.13000, 2 2
Now using y1 as given by (3), we calculate successively k1, k2, k3, k4, and K1. We first find k1 = hf(x1, y1) = 0.2f(0.2, 1.2642) = 0.2 (1.6642) = 0.33284 Then since x1 + and y1 + k1 1 = 1.26420 + (0.33284) = 1.43062 2 2 h 1 = 0.2 + (0.2) = 0.3 2 2
223
we find h k k2 = hf x1 + , y1 + 1 = 0.2 f (0.3, 1.43062) 2 2 = 0.2(2.03062) = 0.40612 Next, since y1 + we find h k k3 = hf x1 + , y1 + 2 = 0.2 f (0.3, 1.46726) 2 2 = 0.2(2.06726) = 0.41345 Since x1 + h = 0.4 and y1 + k3 = 1.26420 + 0.41345 = 1.67765, we obtain k4 = hf(x1 + h, y1 + k3) = 0.2f (0.4, 1.67765) = 0.2 (2.47765) = 0.49553 Finally, we find 1 K1 = (k1 + 2k2 + 2k3 + k4 ) 6 1 = (0.33284 + 0.81224 + 0.82690 + 0.49553) = 0.41125 . 6 Then, the approximate value of the solution at x2 = 0.4 is y2 = 1.2642 + 0.4112 = 1.6754 (4) k2 1 = 1.26420 + (0.40612) = 1.46726 2 2
Rounded off to four places after the decimal point, the exact values at x = 0.2 and x = 0.4 are 1.2642 and 1.6754, respectively. The approximate value at x = 0.2 as given by (3) is therefore correct to four places after the decimal point and the approximate value at x = 0.4 as given by (4) is likewise correct to four place. The remarkable accuracy of the Runge-Kutta method in this problem is certainly apparent. In fact, we employ the method to approximate the solution at x = 0.4 using h = 0.4 (that is, in only one step), we obtain the value 1.6752, which differs from the exact value 1.6754 by merely 0.0002.
224
Exercise
1.
(a)
Apply the Runge-Kutta method to approximate the values of the solution y at x = 0.1, 0.2 and 0.3 using h = 0.1. Carry the intermediate calculations in each step to five figures after the decimal point and round off the final results of each step to four such places.
(c)
Find the exact solution of the problem and compare the results obtained in part(a) with the exact values.
Summary
In this chapter several more approximate methods, namely, power series method, Taylor Series method, method of undetermined coefficients, Euler method and Runge-Kutta method, for the solution of arbitrary first order ordinary differential equations are considered. In the study of each method in this chapter, the primary concern is to obtain familiarity with the procedure itself and to develop skill in applying it.
Keywords
Approximate methods, Taylor Series, Undetermined coefficients, Euler method, Runge-Kutta method.
225
Written by: Sunita Pannu Vetted by: Kapil Kumar
Chapter- 4
Objectives
The main objective of this chapter is to find out the maximal interval of existence of solution of initial value problems. Equal emphasis is on the study of dependence of solution on initial conditions and functions of IVP.
Introduction
In the Picards theorem, we obtained the solution of the initial value problem near the initial point (x0, y0) in a closed rectangle R contained in the domain D and the b interval of convergence was determined by h = min a, , where |f(x, y)| M on M R. We obtained the best possible h determined by the closed rectangle R as a part of the domain and we obtained the solution on (x0 h, x0) on the left side and (x0, x0 + h) on the right side unmindful of points outside the rectangle. Hence the question arises whether the solution exists outside the best possible interval, the answer is yes and is explained in the first three theorems of this chapter. The fact that we are using initial conditions to construct Picard successive approximations shows that the solutions are functions of the initial conditions. Thus, it is evident that we get different solutions of same equations for different initial conditions. Also, we shall obtain the relation between two initial value problems of two different functions and this will explain how the solutions change when the functions are slightly changed. This chapter also extends the theory to a system of equations, which give rise to the study of matrix differential equation. The existence and uniqueness of solutions of the initial value problem of such a vector differential equation will be studied.
226
Theorem 4.1
The largest open interval over which the solution y(x) with y(x0) = y0 is defined is any one of the following two types. (i) (ii)
Proof
(a, b) where both a and b are finite or either a is finite or b is finite. The entire x-axis in the sense - < x < .
Suppose that f satisfies the hypothesis of Picard theorem in D and that (x0, y0) D. Let R:|x- x0)| a, |y y0| b be a rectangle lying in D which gives rise to the best possible h of the conclusion of Picard theorem. The Picard theorem asserts that the initial value problem. dy = f ( x, y ) dx y(x0) = y0 (1)
possesses a unique solution 0 on |x x0| h, but nothing is implied about 0 outside this interval. Now let us consider the extreme right hand point for which 0 is defined. This is the point (x1, y1), where x1 = x0 + h, y1 = 0(x1). Since this point is a point of R, it is certainly a point of the domain D in which the hypotheses of Picard Theorem are satisfied. Thus we can reapply Picard Theorem at the point (x1, y1) and can prove that the differential equation dy/dx = f(x, y) possesses a unique solution 1 such that 1(x1) = y1, which is defined on some interval x1 x x1 + h1, where h1 > 0. Now let us define as follows:
( x), x0 h x x0 + h = x1 , ( x) = 0 1 ( x), x1 x x1 + h1
We now assert that is a solution of problem (1) on the extended interval x0 h x x1 + h1 (see Figure). The function is continuous on this interval and is such that (x0) = y0. For x0 h x x0 + h we have
0 ( x) = y0 + f [t , 0 (t )]dt
x0
227
Figure
and hence
( x) = y0 + f [t , (t )]dt
x0
(2)
1 ( x) = y1 + f [t , 1 (t )]dt ,
x1
or Since y1 = 0(x1) = y0 +
( x) = y1 + f [t , (t )]dt .
x1
x1
x0
( x) = y0 + f [t , (t )]dt
x0
(3)
on the interval x0 + h < x x1 + h1 . Thus, combining the results of (2) and (3) we see that satisfies the integral equation (3) on the extended interval x0 h x x1 + h1 . Since is continuous on this interval, so is f [x,(x)]. Thus, d ( x ) = f [ x, ( x)] dx on [ x0 h, x1 + h1 ]. Therefore is a solution of problem (1) on this larger interval. The function so defined is called a continuation of the solution 0 to the interval [x0 h , x1 + h1]. If we now apply Picard Theorem again at the point [x1+h1, (x1 + h1)], we may thus obtain the continuation over the still larger interval
228
x0 h x x2 + h2 where x2 = x1 + h1 and h2 is positive. Repeating this process further, we may continue the solution over successively larger intervals x0 h x xn + hn extending farther and farther to the right of x0 + h. Also, in like manner, it may be continued over successively larger intervals extending farther and farther to the left of x0 h. Thus repeating the process indefinitely on both the left and the right, we extend the solution to successively larger intervals [an, bn], where [x0 h, x0 + h] = [a0, b0] [a1, b1] [a2, b2] K
n n
[an , bn ] K
Let a = lim an and b = lim bn , where the limit exists finitely or infinitely. We thus obtain a largest open interval a< x <b over which the solution such that (x0) = y0 may be defined. It is clear that two cases are possible. 1. 2. a = - and b = + , in which case is defined for all x, - < x < + . Either a is finite or b is finite or both.
This completes the proof of theorem. We can be more definite concerning the largest open interval over which the solution of this initial-value problem is defined. In this connection we state the following theorem.
Theorem 4.2 Hypothesis
1. 2.
Let f be continuous in the unbounded domain D: a< x < b, - < y < + . Let f satisfies a Lipschitz condition (with respect to y) in this unbounded domain. That is, assume there exists k > 0 such that |f(x, y1) f(x, y2)| k |y1 y2|
A solution of dy/dx = f(x, y) such that (x0) = y0, where (x0, y0) is
any point of D, is defined on the entire open interval a < x < b. In particular, if a = - and b = + , then is defined for all x, - < x < + .
229
Example
y (-1) = 1. It has a solution (x) = 1 through the point (-1, 1) and this solution exists x
on the interval [-1, 0] but cannot be continued further to the right. Because, in that case does not stay within the region D, where f(x, y) = y2 is bounded.
Maximal Interval of Existence
Let f(x, y) be a continuous function on a (x, y)- set E. Let = (x) be a solution of the differential equation
dy = f ( x, y ) dx
(1)
on an interval I. The interval I is called a right maximal interval of existence for if there does not exist an extension of (x) over an interval, say I1, so that = (x) remains a solution of (1), where I is a proper subset of I1 and I, I1 have different right end points. Similarly, a left maximal interval of existence for can be defined.
Definition
Let f (x, y) be continuous on an open (x, y) - set E and let y(x) be a solution of differential equation
dy = f ( x, y ) dx
(1)
on some interval. Then y(x) can be extended as a solution over a maximal interval of existence (- , +). Also if (- , +) is a maximal interval of existence, then y(x) tends to the boundary E of E as x - or x +.
230
Proof Let E1, E2, .En be open subsets of the given set E such that E = E n and
n
let E1 , E 2 ,.......E n be the closures of these open sets. Then E1 , E 2 ,.......E n are compact subsets of E and let us choose E n s.t. E n E n+1 . By Cauchy Peano Existence theorem, if (x0, y0) is any point of E n , there exists an n > 0 such that all solutions of given differential equation through (x0, y0) exist on the interval |x x0| n . Now consider a given solution y = y(x) of differential equation (1) on an open interval J. If J is the right maximal interval of existence, then nothing to prove. If J is not the right maximal interval of existence, then by the theorem on continuation of solution, this solution y(x) can be extended to an interval such that interval contains the right end point of interval J. Thus in proving the existence of a right maximal interval of existence, it can be supposed that y(x) is defined on a closed interval [a, b0]. Let us denote n(1) be the integer so large that (b0, y(b0)) E n(1) . Then y(x) can be extended over an interval [b0, b0 + n(1)]. Now if (b0 + n(1), y(b0 + n(1))) E n(1) , then y(x) can be extended over an interval (another) [b0 + n(1), b0 + 2n(1)] of length n(1). Continuing this way, we can say that there exist an integer j(1) 1 such that y(x) can be extended over a x b1, where b1 = b0 + j(1)n(1) and this (b1 , y(b1)) E n(1) . Let n(2) be so large that (b1 , y(b1)) E n(2) . Applying the same procedure we can say that there exist an integer j(2) 1 such that y(x) can be extended over a x b2, where b2 = b1 + j(2)n(2) such that (b2 , y(b2)) E n(2) . Continuing in this way, we get a sequence of integers n(1), n(2), such that n(1) < n(2) < n(3), and numbers b0 < b1 < b2 . such that y(x) can be extended over [a, +] where bk + as k and (bk , y(bk)) E n(k) . This sequence of points (b1,y(b1)), (b2,y(b2)), . (bk,y(bk)), . is either unbounded or bounded. If bounded, then by Bolzano Weierstrass Theorem this sequence has a limit point (say y0). [If unbounded, we can apply theorem 2]. We claim that this limit point lies on the boundary of set E and cannot be an interior point of E.
231
If this is possible, then there exist a neighbourhood N of the limit point (+ , y0) such that N is contained in E i.e. N(+ , y0) E. Since (+ , y0) is the limit point of the sequence, therefore N (+ , y0) contains infinitely many terms of the sequence (b1,y(b1)), (b2,y(b2)), . If (+ , y0) is not on E, then N (+ , y0) contains some term (bk , y(bk)) corresponding to 1, on the right of + . This is a contradiction as [a, +] is right maximal interval of existence. (+ , y0) lies on E and y(x) E as x +. Similarly, it can be proved that y(x) can be extended over a left maximal interval of existence say (- , a). Therefore, y(x) can be extended over a maximal interval of existence (- , +) and y(x) E as x - or x +. This completes the proof of theorem.
Dependence of solutions on Initial conditions and on the function f A. Dependence on Initial Conditions
We now consider how the solution of the differential equation dy/dx = f (x, y) depends upon a slight change in the initial conditions or upon a slight change in the function f. We will show that under suitable restrictions such slight changes would cause only slight changes in the solution. We first consider the result of a slight change in the initial condition y(x0) = y0. Let f be continuous and satisfies a Lipschitz condition with respect to y in a domain D and let (x0, y0) be a fixed point of D. Then by existence and uniqueness theorem, the initial value problem
dy = f ( x, y ) dx
y(x0) = y0 has a unique solution defined on some sufficiently small interval |x x0| h0. Now suppose the initial y value is changed from y0 to Y0. If Y0 is such that |Y0 y0| is sufficiently small, then we can be certain that the new initial value problem
232
dy = f ( x, y ) dx
y(x0) = Y0 also has a unique solution on some sufficiently small interval |x x0| h1.
(1)
In fact, let the rectangle, R: |x x0| a,| y y0| b, lies in D and let Y0 be such that |Y0 y0| b/2. Then by existence and uniqueness theorem, this problem (1) has a unique solution which is defined and contained in R for |x - x0| h1, where h1 = min (a, b/2M) and M = max |f (x, y)| for (x, y) R. Thus we may assume that there exists > 0 and h > 0 such that for each Y0 satisfying |Y0 y0| , problem (1) possesses a unique solution (x, Y0) on |x x0| h (see Figure).
Figure
We are now in a position to state the basic theorem concerning the dependence of solutions on initial conditions.
Theorem 4.4 Hypothesis
1.
Let f be continuous and satisfies Lipschitz condition with respect to y, with Lipschitz constant k, in a domain D of the xy plane and let (x0, y0) be a fixed point of D.
2.
Assume there exists > 0 and h > 0 such that for each Y0 satisfying |Y0 y0| , the initial value problem.
233
dy = f ( x, y ) dx
y(x0) = Y0
(1)
If denotes the unique solution of (1) when Y0 = y0, and denotes the unique solution of (1) when Y0 = y0 , where y0 y0 = 1 , then
( x) ( x)
1 e kh
on |x - x0| h.
= lim n
n
where
and Similarly,
0 (x) = y0 ; |x - x0| h.
= lim n
n
where
and
0 (x) = y0 ; |x - x 0 | h.
We shall show by induction that
n (x) - n (x) 1
j=0
k j (x - x 0 ) j j!
(2)
on [x0 , x0 + h], where k is the Lipschitz constant. We assume that on [x0 , x0 + h],
234
n-1 (x) - n-1 (x) 1
Then
j=0
n-1
k j (x - x 0 ) j j!
(3)
x0
f [t,
n-1
(t)] dt
y0 y0 +
x0
f [t,
n-1
Applying Lipschitz condition, we have f [x, n-1 (x)] f [x, n-1 (x)] k n-1 (x) - n-1 (x) and since y0 y0 = 1 , so
n (x) - n (x) 1 + k
Using the assumption (3), we have
x0
n-1
n (x) - n (x) 1 + k 1
x0
j=0
n-1
k j (t - x 0 ) j dt j!
= 1 + k 1
= 1 1 +
j=0 n-1
n-1
kj j!
x0
(t - x 0 ) j dt
Since
1 1 +
j=0
n-1
j=0
k j (x - x 0 ) j , j!
we have
n (x) - n (x) 1
j=0
k j (x - x 0 ) j , j!
235
which is (3) with (n 1) replaced by n. Also, on [x0 , x0 + h], we have
x0
f [t, y ] dt
0
y0 y0 +
x0
f [t, y ]
0
f [t, y0 ] dt
1 +
x0
k y
- y0 dt = 1 + k 1 (x x0).
Thus (2) holds for n = 1. Hence the induction is complete and (2) holds on [x0, x0 + h]. Using similar arguments on [x0 h, x0], we have
n (x) - n (x) 1
j=0
k j |x - x 0 | j 1 j!
j=0
(kh) j j!
(x) - (x) 1
j=0
(kh) j . j!
But
j=0
and differ by a sufficiently small amount, then their values will differ by an arbitrary small amount at every point of | x- x0| h. Geometrically, this means that if the corresponding integral curves are sufficiently close to each other initially, then they will be arbitrarily close to each other for all x such that | x- x0| h.
B. Dependence on the Function f
We now consider how the solution of dy/dx = f(x, y) will change if the function f is slightly changed. In this connection we have the following theorem.
236
Theorem 4.5 Hypothesis
1.
In a domain D of the xy plane, assume that (i) f is continuous and satisfies Lipschitz condition with respect to y, with Lipschitz constant k. (ii) (iii) F is continuous. |F(x, y)- f(x, y)| for (x, y) D.
2.
Let (x0, y0) be a point of D and let (i) be the solution of the initial value problem
dy = f ( x, y ) dx y ( x0 ) = y0
(ii)
(iii)
Conclusion
( x) ( x) (e kh 1) on x x0 h.
k
Proof Let 0 ( x) = ( x) and define a sequence of function {n } by
n ( x) = y0 + f [t , n 1 (t )]dt ,
x0 n
x x0 h
(n = 1, 2, 3, )
Then = lim n is a solution of dy/dx = f(x, y) such that ( x0 ) = y0 on |x x0| h. By Hypothesis 1(i), the initial-value problem dy/dx = f(x, y), y(x0) = y0, has a unique solution on |x x0| h. Thus from Hypothesis 2(i) ( x) = ( x) on so lim n = .
n
x x0 h , and
237
( x) = y0 + F [t , (t )]dt ,
x0 x
x x0 h .
n ( x) ( x)
k j-1 (x - x 0 ) j j! j =1
n
(1)
(2)
Then
n ( x) ( x) = y0 + f [t , n 1 (t )]dt y0 F [t , (t )] dt
x0 x0
x0
f [t , n 1 (t )] F [t , (t )] dt .
n ( x) ( x)
x0
f [t , n 1 (t )] f [t , (t )] - [t , (t )] dt.
Applying the inequality |A B| |A| + |B| and then the Lipschitz condition satisfied by f, we have
n ( x) ( x)
x0
f [t , n 1 (t )] f [t , (t )] dt +
x
x0
[t , (t )] dt
x0
n 1 (t ) - (t ) dt +
x0
[t , (t )] dt.
Now using the assumption (2) and the fact that | (x, y)| = |F(x, y)- f (x, y)| , we obtain
n ( x) ( x) k
j=1
n 1
k j-1 j!
x0
(t - x 0 ) j dt +
x0
dt
n 1
j=1
k j (x - x 0 ) j + 1 (j + 1)!
+ (x - x 0 )
238
=
j=1
k j -1 (x - x 0 ) j j!
Thus (2) holds with (n 1) replaced by n. Also Hypothesis 1 (iii) shows that
1 ( x) ( x)
x0
f [t , (t )] F [t , (t )] dt
x0
dt = (x - x 0 )
on [x0, x0 + h]. Thus (1) holds for n = 1. Thus the induction is complete and so (1) holds on [x0, x0 + h] for n = 1, 2, 3, .. Using similar arguments on [x0 h, x0], we thus have
n ( x) ( x)
k j-1 |x - x 0 | j j! k j =1
n
(kh) j j! j =1
n
(kh) j j! j =1
( x ) ( x )
kh (e 1) on |x x0| h. k
Thus, under the hypotheses stated, if is sufficiently small, the difference between the solutions and will be arbitrarily small on |x x0| h. This completes the proof of the theorem.
Upto now, we obtained solutions of single differential equation of different types and obtained the existence and uniqueness of solution of the initial value problem of first order equation which are not necessarily linear. However, we come across practical situations where we have to deal with more than one differential equation with many variables or depending upon a single variable. For example, if we consider the motion of a particle in three dimensions, we get one such situation.
239
Systems of First Order Equations
In analogy with the system of the single equation x'(t) = f (t, x), t I, we shall study the system of equations by considering the following n-equations:
= f1(t, x1, x2, . xn) x1
where f1, f2, ... fn are the given n-real valued functions defined on the domain D in R
n+1
Our problem is to find an interval I and n-differentiable functions 1, 2, . n on I such that (i) (ii) (t, 1(t), 2(t), ..., n(t)) is in D for t I and
i'(t) = fi (t, 1(t), 2(t), ..., n(t)) for all t I and i = 1, 2, 3, . n.
When n such differentiable functions (1, 2, . n) exist, (1, 2, . n) is called a solution of (1) on I. Using the vector notation x = (x1, x2, ., xn) and F = (f1(t), f2(t), . fn(t)), the system can be written in the form x' = F(t, x) and the solution is denoted by
= (1, 2, . n).
which is a linear system of n first order equations in n unknown functions x1, x2, ... xn and aij(t), bj(t) i, j = 1, 2, ... n are all given functions on I. The above system (2) is a particular case of (1), since x i (t) can be taken as fi (t, x1, x2, . xn) where each fi is linear on I. By using matrices, we can represent (2) as x'(t) = A (t) x(t) + B(t), t I (3)
240
where
x1 x x = 2 , A(t) = .... xn a11 a12 a 21 a 22 ... ... a n1 a n2 ... a1n ... a 2n and ... ... ... a nn b1 (t) b (t) B(t) = 2 . ... b n (t)
(3) is a linear equation in x and it is a matrix representation of a linear nonhomogeneous system. If B(t) = 0, then (3) reduces to a homogeneous system x'(t) = A(t) x(t).
Matrix Preliminaries
Let A(t) = [aij(t)], i,j = 1, 2, 3, .. n be an n x n matrix of functions defined for t I = [a, b].
Definition 1
The matrix A(t) = [aij(t)], t I is said to be continuous or differentiable, if every element aij(t), i, j = 1, 2, ... n of the matrix A(t) is continuous or differentiable on I. The derivative A'(t) is obtained by differentiating every element of A(t), that is, A'(t) = [a'ij (t)], i, j = 1, 2, 3, ... n.
Definition 2
If x = (x1, x2, . xn) Rn , then the norm of x denoted by |x| is defined as |x| = |x1| + |x2| + + |xn| =
Definition 3
x
i=1
Let A = [aij] denotes an n x n matrix then the norm of the matrix A denoted by |A| is defined as |A| =
i, j = 1
a ij .
If A(t) = [aij(t)] is a continuous matrix on I, then |A(t)| is also continuous on I. Also, (i) (ii) |A + B| |A| + |B| |AB| |A| |B|
241
(iii) (iv) |A | = || |A| for every scalar |Ax| |A| |x| for any vector x.
Definition 4
A sequence {An} of matrices is said to be convergent, if given any > 0, there exists a positive integer n0 such that |Am An| < for all m, n n0.
Definition 5
A sequence {An} of matrices is said to tend to a limit A, if given any > 0, there exists a positive integer n0 such that |An A| < for all n n0.
Definition 6
n=1
of partial sums of the series is convergent and the sum of the series is the limit matrix of partial sums. The exponential of a matrix A is defined as e A = I +
An , where the n = 1 n!
convergence of the series is in the sense of the Definition 6 and An represents the n-th power of A and I is the identity matrix. From the above definition, the following important properties are observed: (i) |eA| (n - 1) + e|A| where | | denotes the norm.
An , where I is the identity matrix of order n. n = 1 n!
Proof e A = I +
Hence
|e | | I | +
n=1
n!
Now for any matrix A = [aij], |A| = | a ij | so that we have |I| = n. Hence we
i, j
have |e | n 1 + 1 +
A
n=1
n!
= (n 1) +
n=0
n!
= (n 1) + e|A|
242
which gives |eA| (n 1) + e|A| . (ii) For matrices A and B, it is not true in general that eA+B = eA . eB but this
The sum of the diagonal elements of a matrix A is called the trace of A. If A is the given matrix, then the trace of A is denoted by trA.
Representation of n-th Order Equation as a System
Any differential equation of order n can be written as a system of n first order differential equations.
Theorem 4.6
The general n-th order initial value problem x(n) = f (t, x, x', .. x(n-1)), t I x (t0) = a0, x' (t0) = a1, ... x(n-1) (t0) = an-1, t0 I where a0, a1, a2, an-1, are given constants, is equivalent to a system of nlinear equations.
Proof We define a new family of unknown functions xn = x(n 1) .
(1)
i.e.
We can then rewrite the original differential equation as a system of differential equations with order 1 and dimension n. Thus we have the system of equations
= x' = x2 x1
x 2 = x'' = x3 x 3 = x''' = x4 . .
(n-1) x = xn n -1 = x
and
(n) x . n = x
(2)
243
x n (t) = f (t, x1, x2, .., xn). Let = (1, 2, ... n) be a solution of (2). Then
= 1, ....., n = 1( n 1) 2 = 1 , 3 = 2
(3)
(4)
Hence
f (t, 1 , 2 , ... n ) = f [t, 1 , 1 , ... 1( n 1) (t)] = 1(n) (t) which shows that the component 1 is a solution of (1). Conversely, if 1(t) is a
solution of (1), then the vector = (1, 2, ... n) is a solution of (2). Thus, system (2) is equivalent to (1). Further we obtain the initial conditions of the system as a vector
The general n-th order equation is a0(t) x(n) + a1(t) x(n-1) + . + an(t) x = b(t), t I where a0(t) 0 for any t I. Now make the following substitutions: x(t) = x1(t)
(t) = x'(t) = x2 x1
(1)
(2)
244
x (n) =
a n (t) a (t) a (t) (n-1) b(t) x n-1 x' ... 1 x + a 0 (t) a 0 (t) a 0 (t) a 0 (t)
x 2 (t) = 0 x1 + 0 x2 + 1 x3 + + 0 . . .. . . . x n (t) =
a n (t) a (t) a (t) b(t) x1 n-1 x 2 ... 1 xn + a 0 (t) a 0 (t) a 0 (t) a 0 (t)
(3)
1 0 ... 0 a n-1 a0
0 1 ... 0 a n-2 a0
0 x1 0 x 2 .. .. + .. .. b(t) x n a0
(4)
The n-th order linear equation (1) is equivalent to the system (2) leading to the matrix differential equation (4).
Example 1
is
Example 2
Find the matrix form of the linear equation x"' - 4x" + 10x' - 6x = 9t (1)
245
Let us make the following substitution x1 = x, x2 = x', x3 = x'' and express x, x', x'' and x"' in terms of x1, x2, x3 and their derivatives. From (2)
= x2 x' = x1 x" = x 2 = x3 and x''' = x 3 .
(2)
(3)
x 2 = 0 + 0 + 1x3 x 3 = 6x1- 10x2 + 4x3 + 9t With the help of (4), (5) and (6), we get
x1 0 1 0 x1 0 x = 0 0 1 x + 2 2 0 9t 6 10 4 x3 x3
In this section we will solve the initial value problem of the vector differential equation x'(t) = A(t)x(t), x(t0) = x0, tI
where A(t) is an n x n matrix defined over I and x(t) is a vector function on I. For a fixed t0, x(t0) is a fixed vector. We shall state and prove the following theorem for a system of equations given in the vector form.
Theorem 4.7 Let A(t) be a continuous n x n matrix defined on a closed and bounded
interval I. Then the initial value problem x'(t) = A (t) x(t), x(t0) = x0, has a unique solution on I.
Proof We first note that the initial value problem (1) is equivalent to the solution of
t, t0 I
(1)
246
x(t) = x 0 + A(s) x(s) ds
t0
(2)
which gives x(t0) = x0. On differentiating (2), we get x'(t) = A(t) x(t), t I If x(t) is a vector solution of (1), then integrating (1), we get x(t) = x(t 0 ) + A(s) x(s) ds.
t0
With the help of the successive approximations, we shall show that {xn(t)} is a Cauchy sequence in Rn for every t 1. Since Rn is complete, {xn(t)} converges uniformly to a limit x(t) and then we prove its uniqueness. Let us define the approximations by x(t 0 ) = x 0 , x n+1 (t) = x 0 + A(s) x n (s) ds , t t0, t I
t0
for n = 0, 1, 2, 3, .. Since x0 is a given vector, the sequence {xn(t)} is well-defined. First we shall show that {xn(t)} converges uniformly on I to a function x(t) which is the solution of the initial value problem (1) and for this it is enough to show that {xn(t)} is a Cauchy sequence in Rn. Since Rn is complete {xn(t)} converges to x(t). To show that xn(t) is a Cauchy sequence, let us consider x 0 (t) + ( x n+1 (t) - x n (t) )
n =1
where sequence of partial sums is xn(t). Hence, consider x n+1 (t) - x n (t) = A(s) ( x n (s) - x n-1 (s) ) ds
t0
We shall construct an upper bound for | x n+1 (t) - x n (t) | by induction method where | | is the norm. Since A(t) is a continuous matrix defined on I, there exists a constant M such that |A(t)| M, t I (4)
247
Now x1 (t) - x 0 (t) = A(s) x 0 (s) ds
t0
x1 (t) - x 0 (t)
t0
A(s)
x 0 ds (5)
x 2 (t) - x1 (t)
t0
A(s) x1 x 0 ds
t
x 2 (t) - x1 (t) M M x 0 s t 0 ds
t0
=M
x 0 (t - t 0 ) 2 2!
Proceeding in same manner, we have by induction x n+1 (t) x n (t) M n+1 x 0 (t - t 0 ) n+1 (n + 1)! (6)
n =1
of
positive constants whose sum is x 0 e M (t - t 0 ) . Since the n-th term of a convergent series tends to zero as n , the right hand side of (6) tends to zero as n . Hence {xn(t)} is a Cauchy sequence in Rn for each t I. Since Rn is complete, xn(t) x(t) in the norm | |. Since the norm convergence is uniform, xn x uniformly on I. Now consider x n+1 (t) = x 0 + A(s) x n (s) ds
t0
(7)
Taking the limit as n on both sides of (7), we have x(t) = x 0 + lim A(s) x n (s) ds
n
t0
(8)
248
Since xn(t) x(t) uniformly on I, we can take the limit under the integral sign of (8) to get x(t) = x 0 + A(s) x(s) ds
t0
which proves that x(t) is the solution of the integral equation. Since the solution of the above integral equation and the initial value problem (1) are equivalent, x(t) gives the solution of the initial value problem. Now, we shall prove the uniqueness of the solution. If the solution x(t) is not unique, let y(t) be another solution of (1). Then we get x(t) = x 0 + A(s) x(s) ds
t0
Or
x(t) - y(t) =
t0
so that
x(t) - y(t) M
x(s) - y(s)
ds
Thus, for any given > 0, we get from the above inequality x(t) - y(t) < + M Let us take z(t) = |x(t) - y(t)|. Then we have z(t) < + M
t0
x(s) - y(s) ds
t0
z(s) ds
, tI
If r(t) = + M z(s) ds , then r(t0) = and z(t) < r(t). From the definition
t0
r'(t) = M z(t) < M r(t) so that we have r'(t) - M r(t) < 0. Integrating the left hand side of the above inequality, we get r(t) e Mt - r(t 0 ) e Mt 0 < 0 from which we have r(t) < r(t 0 ) e M(t - t 0 )
249
Hence z(t) < r(t) < e M(t - t 0 ) , since r(t0) = . Since the above inequality is true for each > 0, we get z(t) < 0 which implies |x(t) - y(t)| = 0 which gives x(t) = y(t) on I. This proves that the solution is unique. Hence the proof of the theorem is complete.
Note
The zero vector function on I is always a solution of (1). If the solution of (1)
is zero for any to t0 I, and since the solution is unique, then it must be zero throughout I.
Corollary
x(t) = 0 is the only solution of the initial value problem x' = A(t)
(1)
(2)
In a similar manner, we can show that xn(t) = 0 for all n. Hence xn(t) 0 as n for all n implies x(t) = 0. Thus x(t) = 0 is the only solution.
Summary
This chapter is devoted to the study of finding the largest open interval over which the solution of IVP is defined and then establishing the conditions for the continuation of solution in a general domain. Further, dependence of solutions of initial value problems on initial conditions and functions are also proved. In last, the study of solution of the initial value problem of first order equation is extended to analyse systems of first order differential equations. The criterion for existence and uniqueness of solutions for linear system of equations is presented.
Keywords
250
Written by: Sunita Pannu Vetted by: Kapil Kumar
Chapter-5
Objectives
The purpose of this chapter is to introduce the reader to Pfaffian differential equations, their condition of integrability and some methods to solve such equations. Some theorems for differential inequations are also presented.
Introduction
The purpose of this chapter is to study Pfaffian (Total) differential equations and differential inequations. Certain methods to solve Pfaffian differential equations in three variables are explained. The most important techniques in the theory of differential equations involve the integration of differential inequalities Gronwall inequality is proved in this context. Comparison theorems are also presented to compare the unknown solutions of one differential equation with the known solutions of another.
Pfaffian Differential Forms and Equations
The expression
F (x , x
i 1 i=1
, ........, x n ) dx i
(1)
where Fi (i = 1, 2, .., n) are continuous functions of some or all of the n independent variables x1, x2, ..xn, is called a Pfaffian differential form of n variables. The relation
F dx
i i=1
=0
(2)
251
We shall consider Pfaffian differential equations in two variables and those in a higher number of variables, separately. In the case of two variables we may write equation (2) in the form P(x, y) dx + Q(x, y) dy = 0 which is equivalent to
dy = f ( x, y ) , dx
(3)
(4)
if we write f(x, y) = - P/Q. Now the functions P(x, y) and Q(x, y) are known functions of x and y, so that f(x, y) is defined uniquely at each point of the xy plane at which the functions P(x, y) and Q(x, y) are defined. The fundamental existence theorem in the theory of ordinary differential equations is of the form:
Theorem 5.1
A Pfaffian differential equation in two variables is P(x, y) dx + Q(x, y) dy = 0 where the functions P and Q are continuously differentiable. If Q(x, y) 0, then dy - P(x, y) = dx Q(x, y) (1)
Then, by existence theorem, this equation has a solution. Let (x, y) = c be a solution of this equation.
d(x, y) = 0
dx + dy = 0 x y
(2)
252
P(x, y) =
, x
Q(x, y) =
, y
In case, given equation (1) is not exact, then multiplying (1) by i.e.
(x, y) P(x, y) dx + (x, y) Q(x, y) dy = 0
(3)
Find value of (x, y) from (3), which when multiplied to given equation, makes it exact. Then the equation is said to be integrable and will possess an integrating factor.
Pfaffian Differential Equation in Three Variables
When there are three variables, the Pfaffian differential equation is of the form P dx + Q dy + R dz = 0, (1)
as
X. dr = 0
(2)
A necessary and sufficient condition that there exists between two functions u(x, y) and v(x, y) a relation F(u, v) = 0, not involving x or y explicitly is that
(u,v) = 0. (x,y)
Proof
253
F(u, v) = 0 Differentiating this identity with respect to x, we get
F u F v + =0 u x v x
(1)
(2)
(3)
Eliminating
But the relation (1) involves both u and v, it follows that zero, so that
(u,v) =0 (x,y)
F is not identically u
(4)
Second, the condition is sufficient. We may easily eliminate y from the equations u = u (x, y), v = v (x, y). Let the resultant eliminant be F (u, v, x) = 0. Differentiating this relation with respect to x, we get
F F u F v + + =0 x u x v x
(5)
254
Eliminating
F from these two equations, we find that v F v (u, v) F + =0 x y (x, y) u
The function v is a function of both x and y, so that v/y can not be identically zero. Hence
F = 0, x
which shows that the function F does not contain the variable x explicitly. Thus from (5), we get F(u, v) = 0.
Lemma 2
If X is a vector such that X. curl X = 0 and is an arbitrary function of x, y, z then (X). curl (X) = 0.
Proof
x, y, z
(P)
(R) (Q) z y
where X has components (P, Q, R). The right-hand side of equation may be written in the form
R Q 2 P - PR - PQ z z y x, y, z x, y, z y
255
X. curl X = {X. curl X} 2 = 0 as X. curl X = 0 The converse of this theorem is also true, as can be seen by applying the factor 1/ to the vector X.
Note
Equations of the form P dx + Q dy + R dz = 0, do not always possess integrals. To find out the criterion for determining,
whether or not an equation of this type is integrable, the following theorem is proved.
Theorem 5.2
The condition is necessary, for if the equation P dx + Q dy + R dz = 0 is integrable, there exists among the variables x, y, z a relation of the type F (x, y, z) = C where C is a constant. Writing this in the differential form
F F F dx + dy + dz = 0 x y z
(1)
(2)
(3)
Now, equations (1) and (3) must be identical, i.e. there must exist a function
(x, y, z) such that
F F x = y P Q i.e. i.e., such that Now, since P = F , x F = z = (x, y, z) (say) R Q = F , y R = F z
256
we have so that curl(X) = 0 X. curl (X) = 0
Sufficient condition. If z is treated as a constant, the differential equation (1) becomes P(x,y,z) dx + Q(x,y,z) dy = 0, (4)
which is a Pfaffian differential equation in two variables and by Lemma 1, it possesses a solution of the form U(x, y, z) = c1 where the constant c1 may involve z. In differential form (5) can be written as d U(x,y,z) = 0 i.e. U U dx + dy = 0 x y (6) (5)
Now (4) and (6) must be identical. Thus there must exist a function such that
U = P, x
U = Q . y
(7)
Substituting from the equations (7) into equation (1), we see that the latter equation may be written in the form U U U dx + dy + dz + x y z which is equivalent to the equation dU + K dz = 0, if we write K = R U z (9) (8) U R dz = 0 z
Now we are given that X . curl X=0, and it follows from Lemma 2 that
257
X. curl X = 0 Now U U U , , + K X = (P, Q, R) = z x y = grad U + (0, 0, K) Hence U U U K K , , + K . , , 0 X. curl (X) = z x x y y = U K U K . x y y x
Thus the condition X. curl X = 0 is equivalent to the relation (U,K) =0 (x,y) From Lemma 1, it follows that there exists between U and K a relation independent of x and y but not necessarily of z. In other words, K can be expressed as a function K (U, z) of U and z alone, and equation (8) takes the form dU + K (U , z ) = 0 dz which, by Theorem 1, has a solution of the form (U, z) = c where c is an arbitrary constant. Now, replacing U by its expression in terms of x, y, and z, we obtain the solution in the form F(x, y, z) = c which shows that the equation (1) is integrable. Once it has been established that the equation is integrable, it only remains to determine an appropriate integrating factor (x, y, z). We shall discuss the solution of Pfaffian differential equations in three variables.
258
Solution of Pfaffian Differential Equations in Three Variables
We shall now consider some methods by which the solutions of Pfaffian differential equations in three variables x, y, z may be derived.
(a) By Inspection Once the condition of integrability is satisfied, one may find
by rearranging the terms or dividing or multiplying by some suitable function to reduce to a form containing some or several parts, which are exact differentials. It is often possible to derive the primitive of the equation by inspection. In particular, we may also have curl X = 0, then X must be of the form grad u, and hence the equation
X . dr = 0 reduces to the form
curl X = (0, -2x2, 6y2), and hence X. curl X = 0. We may write the equation in the form x2(z dx + x dz) y3 dx + 3xy2 dy = 0 i.e. z dx + x dz y3 3y 2 dx + dy = 0 x2 x
i.e.
y3 d(xz) + d = 0 x
259
by showing that it is integrable.
Solution
Here P = 1, Q = 1, R = y + x
X. curl X = P [Q z - R y ] + Q [R x - Pz ] + R [Py - Q x ] = 0
Hence, equation is integrable. Dividing the given equation by (y + x) throughout, we obtain dz + Hence dx + dy =0 x+y
(b)
differential equation in the form P(x) dx + Q(y) dy + R(z) dz = 0 which obviously gives on integration
P ( x ) dx
where c is a constant.
Q ( y ) dy
R ( z ) dz = c
Example 3
Solution
where k is a constant.
Example 4
260
Solution
It is easy to verify the condition of integrability. Dividing the given
(c)
One Variable Separable It may happen that the equation is of the form
P(x, y) dx + Q(x, y) dy + R(z) dz = 0 (1)
R ( z ) dz
= c
Example 5
Verify that the equation x(y2- a2) dx + y(x2- z2) dy - z(y2- a2) dz = 0
261
Solution
If we divide throughout by (y2- a2) (x2- z2), we see that the equation
assumes the form x dx - z dz y dy + 2 = 0 2 2 x z y a2 showing that it is separable in y. By the above argument it is therefore integrable if P R = z x which is readily shown to be true. To determine the solution of the equation we note that it is 1 1 d log (x 2 - z 2 ) + d log (y 2 - a 2 ) = 0 2 2 so that the solution is (x2 z2) (y2 a2) = c where c is a constant.
(d)
is said to be homogeneous if the functions P, Q, R are homogeneous in x, y, z of the same degree n. To solve such an equation we make the substitutions y = ux, z = vx. Substituting (3) into (2), we see that equation (2) takes the form P(1,u,v) dx + Q(1,u,v) (u dx + x du) + R(1,u,v) (x dv + v dx) = 0, a factor xn cancelling out. If we now write (3)
A(u, v) =
B (u, v) =
262
and can be solved by method (c).
Solution
It is easy to show that the condition of integrability is satisfied. Making the substitution y = ux, z = vx, we find that the equation satisfied by x, u, v is uv (u + v) dx + v (v + 1) (u dx + x du) + u (u + 1) (v dx + x dv) = 0 which reduces to
dx 1 1 1 1 + du + dv = 0 x u 1+ u + v v 1+ u + v
The solution of this equation is obviously x2uv = c( 1+ u + v), where c is a constant. In terms of original variables, we see that the solution of the given equation is xyz = c(x + y +z).
Example 7
Solution
263
and Substituting
X . curl X = 0
Splitting the factors of du and dv into partial fractions, we see that this is equivalent to dx du dv du + dv =0, + + x u v+1 u+v which on integration, gives the solution as x u(v + 1) =C u+v y(z + x) =C. z+y i.e.
(e) Natanis Method In this method, we assume one of the variables as constant. Let
us treat the variable z as constant, so that the resulting differential equation Pdx + Qdy = 0, is easily integrable. Let its integral be given by (x, y, z) = c1, c1 may involve z. (1)
The solution of equation P(x,y,z) dx + Q (x,y,z) dy + R(x,y,z) dz = 0 is then of the form ( , z ) = c2 where c2 is a constant, and we can express this solution in the form (x, y, z) = (z) where is a function of z alone. To determine the function (z) we observe that, if we give the variable x a fixed value, say, then (, y, z) = (z) is a solution of the differential equation (3) (2)
264
Q(, y, z) dy + R(, y, z) dz = 0 This equation will always have a solution of the form K(y, z ) = c by using the methods of the theory of first order differential equations. Since equations (3) and (5) represent general solutions of the same differential equation (4), they must be equivalent. Therefore, if we eliminate the variable y between (3) and (5), we obtain an expression for the function (z). Substituting this expression in equation (3), we obtain the solution of the Pfaffian differential equation. The method is often simplified by choosing a value for , such as 0 or 1, which makes the solving of differential equation (4) easy. It is necessary to verify in advance that the equation is integrable before using Natanis method. (5) (4)
Example 8
Solution
(1)
dx dy + =0 2 1+ x 1 + y2
(2)
265
This can be easily integrated to tan-1x + tan-1y = constant = tan-1 Now tan 1x + tan 1 y = tan -1 we see that the solution of equation (2) is 1 xy =c x+ y (3) 1 (say) c
x+ y 1 xy
This solution must be the form assumed by (1) in the case z = 1; or (3) must be equivalent to the relation
x( y 2 + 1) = f ( y) x+ y
(4)
Eliminating x between equations (3) and (4), we find that f(y) = 1 cy. Substituting this expression in equation (1), we find that the solution of the equation is x(y2 + z) = z(x + y) (1 cy).
Example 9
Solution
obtain
y dx - xdy = 0 This implies Putting y = 1, we obtain x = f(z) and (1 + z2) dx + (1 + x2) dz = 0 (2)
x = f ( z) y
(1)
266
which on integration gives (x + z)/ (1 xz) = c Elimination of x from (2) and (3) gives (3)
f ( z) =
Thus from (1), the primitive is
Cz 1 + Cz
x Cz = . y 1 + Cz
Problems
Verify that the following equations are integrable and find their primitives: 1. 2. 3. 4. 5. 6. 7. 8. 9. 10. 2y(a x) dx + [z y2 + (a x)2] dy y dz = 0 zy2dx + zx2dy x2y2dz = 0 (y2 + yz + z2) dx + (z2 + zx + x2)dy + (x2 + xy + y2)dz = 0 yz dx + xz dy + xy dz = 0 (1 + yz) dx + x (z x) dy (1 + xy) dz = 0 y (x + 4) (y + z) dx x (y + 3z) dy + 2xy dz = 0 yz dx + (x2y zx) dy + (x2z xy) dz = 0 2yz dx 2xz dy (x2 y2) (z 1) dz = 0 y(x + y) (z + y) dx - x (y + z2) dy + 2xy dz = 0 (2xyz + z2) dx + x2zdy + (xz + 1) dz = 0
Differential Inequations
The most important techniques in the theory of differential equations involve the integration of differential inequalities. The following integral inequality known as Gronwalls inequality is fundamental in the study of ordinary differential equations. It is one of the simplest and most useful result involving an integral inequation.
267
In the following r, u, v, U, V are scalars while y, z, f, g are ndimensional vectors.
closed interval [a,b]. Let c be any non-negative constant. Then the inequality
v(t ) c + v( s )u ( s )ds,
a
for a t b
for a t b
Proof
(1) (2)
Then
V(a) = c
and by hypothesis v(t) V(t), and V(t) c > 0 on [a, b], as u and v are non-negative functions. Also from (1), we have, on [a, b], V'(t) = v(t) u(t) V(t) u(t), using (3). This implies, using (4) (4) (3)
V '(t ) u (t ). V (t )
Integrating (5) over [a, t], we get
(5)
268
log V ( t ) a
t
u(s)ds
a
which gives log V(t) log V(a) Using V(a) = c, we have log V(t) - log c u ( s )ds .
a t
u(s)ds
a
(6)
Replacing the left hand side of (6) by lesser term, as in (3), we get
t v(t) V(t) c exp u ( s )ds a
or
(7)
Case II
269
0 r(t) exp {| t- t0|}.
Cor. 1 Let f(t, y) satisfies a Lipschitz condition with constant k for y D and
t t0 . Let y(t) and z(t) be solutions of problem
dy = f (t , y ) dt
for t- t0 such that y(t0) = y0, where y0, z0 D. Then z(t) y(t) |z0 y0| exp {kt - t0}. z(t0) = z0, (1)
Proof
Given that y(t) and z(t) are solutions of (1), the corresponding integral
equations are:
y (t ) = y0 + f ( s, y ( s ))ds,
t0 t
z (t ) = z0 + f ( s, z ( s ))ds,
t0
(2)
(3)
Taking the norm of both sides and applying the Lipschitz condition, it follows that
z (t ) y (t ) z0 y0 + [ f ( s, z ( s )) f ( s, y ( s ))]ds
t0 t
z0 y0 + f ( s, z ( s )) f ( s, y ( s )) ds
t0 t
0 z (t ) y (t ) z0 y0 + k z ( s ) y ( s ) ds
t0
270
From Gronwalls inequality, with r(t) = z(t) y(t), = z0 y0 and = k, the result follows immediately.
Cor. 2 Let f(t,y) satisfies a Lipschitz condition for y D and t t0 . Then the
initial value problem has a unique solution, that is, there is at most one continuous function y(t) which satisfies
dy = f (t , y ) , dt
y(t0) = y0
Proof Putting z0 = y0 in Cor. 1, we see that z(t) = y(t) for all t t0 , which
shows the uniqueness of the solution of initial value problem, whenever f(t, y) satisfies a Lipzchitz condition. Hence the result.
Theorem 5.4 Let f(t, y) satisfies a Lipschitz condition for t a. If the function u =
u(t) satisfies the differential inequality.
dy f (t , y ) dt
and v = v(t) is a solution of differential equation
for t a
(1)
dy = f (t , y ) dt
(2)
271
satisfying the initial conditions u(a) = v(a) = c0 then u(t) v(t) for t a. (3)
for some t1 in the given interval. Let t0 be the largest t in the interval [a, t1] such that u(t) v(t). Then u(t0) = v(t0). Let Then and (t) = u(t) v(t) . (t0) = 0, (t1) > 0 (5) (6) (7)
= K (t), where K is the Lipschitz constant for the function f. Multiplying both sides of (8) by e-Kt, we write 0 e-Kt{'(t) - K(t)} = This implies
d (t) e Kt } . { dt
d (t) e Kt } 0 in [t 0 , t1 ] { dt
(9)
272
So, (t). e-Kt is a decreasing function for [t0, t1]. Therefore (t) e-Kt (t0) e-Kt0 (t) (t 0 ) e K(t - t0 ) (t) 0 for all t in [t0, t1], using (6). for all t in [t0, t1]
(t) vanishes identically in [t0, t1]. This contradicts the assumption that (t1) > 0. Hence, we conclude that u(t) v(t) for all t in the given interval. This completes the proof
dy = U (t , y ), dt
respectively, where U (t, y) V (t, y)
dz = V (t , z ) dt
(1)
(2)
in the strip a t b and U or V satisfies a Lipschtiz condition, and u(a) = v(a). Then u(t) v(t) (4) for all t [a, b]. (3)
273
f(t) = - u(t), satisfy the differential equations g(t) = - v(t) (5)
du = U (t , u ), dt
and
dv = V (t , v), dt
- U (t, -v), using (2)
(6)
As g(t) is a solution of
dv U (t , v) dt
and f(t) is a solution of
du = U (t , u ) . dt
Therefore, by Theorem 5.4, we obtain the inequality g(t) f(t) -g(t) - f(t) v(t) u(t) u(t) v(t) This completes the proof. for t a for t a for t a for t a
Remark The inequality u(t) v(t) in this comparison theorem 5.5 can often be
replaced by a strict inequality.
Corollary 1
In theorem 5.5, for any t1 > a, either u(t1) < v(t1) or u(t) v(t) for a t t1 (1)
274
If u(t1) is not less than v(t1), then u(t1) = v(t1). Then, either u and v are identically equal for a t t1, or else u(t0) < v(t0) for some t0 in the interval (a, t1). Let 1(t) = v(t) u(t), (3) (2)
(4)
(5)
Then
and by theorem 5.5,
1(t0) > 0,
1'(t) = v'(t) u'(t) = V (t, v(t)) U (t, u(t)) V (t, v(t)) V (t, u(t)) (Q U V given) - K {v(t) u(t)}
1'(t) - K 1
(1' + K 1) 0.
(7)
Hence
{e
Kt
. 1 ( t )} = e Kt { 1 ( t ) + K 1 ( t )} 0,
using (7) for t [a, t1]. This shows that the function (t) = ekt 1(t) is an increasing function on the interval [a, t1]. So (t) (t0) for t [t0, t1]
ekt 1(t) e Kt 0 1(t0) 1(t) 1(t0) e- K(t-t0 ) > 0 1(t) > 0 in [t0, t1]
275
v(t) u(t) > 0 in [t0, t1] v(t) > u(t) in [t0, t1] [a, b],
which is a contradiction. Hence, u and v are identical for a t t1. This completes the proof. Cor 2 In theorem 5.5, assume that U as well as V, satisfies a Lipschitz conditions and, instead of u(a) = v(a), that u(a) < v(a). Then u(t) < v(t) for t > a.
Proof The proof will be by contradiction. If we had u(t) v(t) for some t > a, there would be a first t = t1 > a, where u(t) v(t). We define two functions y = (t) = u(-t), z = (t) = v(-t). Then and satisfy the differential equations
dy = U (t , y ), dt
and the respective initial conditions,
(1)
(2)
dz = V (t , z ) dt
(3)
(4)
we can apply theorem 5.5 in the interval [-t1, -a], knowing that the function U (-t, y) satisfies a Lipschitz condition. So, by theorem 5.5, we conclude that (t) (t) in [-t1, -a]
276
(-a) (-a) u(a) v(a), (5)
which is a contradiction (to the given assumption that u(a) < v(a)). Therefore, the assumption that u(t) v(t) is wrong. Thus, u(t) < v (t) Hence the result. for t > a.
Summary
In this chapter, after proving some preliminary results about Pfaffian differential equations, the criterion for determining the condition of integrability of such equations is explained. Some methods of solutions of Pfaffian differential equations in three variables are also discussed. Some theorems for differential inequations along with Gronwall inequality are presented at the end of chapter.
Keywords
Total differential equations, Integrability, Comparison theorems, Gronwall inequality.
277
Written by: Sunita Pannu Vetted by: Kapil Kumar
Chapter-6
Objectives
In this chapter, the reader is made familiar with Sturm theory, Sturm separation and comparison theorems describing the location of roots of homogeneous second order linear differential equations. Riccati Equations and Pruffer transformation are also important parts of this chapter.
Introduction
We know that a second order differential equation can be expressed in the form
d dx P(t ) + Q(t ) x = 0, dt dt
(1)
(called self adjoint form) where P(t) has continuous derivate, Q(t) is continuous and P(t) > 0 on a t b. We shall need a well-known theorem on point sets known as the Bolzano-Weierstrass theorem. Suppose E is a set of points on the t axis. A point t0 is called a limit point of E if there exists a sequence of distinct points t1, t2, t3 of E such that lim tn = t0 . The Bolzano-Weierstrass theorem states that every bounded
n
Theorem A
If the function is a solution of the homogeneous equation
dnx d n 1 x a0 (t ) n + a1 (t ) n 1 + ...... + an (t ) x = 0 dt dt
Such that (t0) = 0, '(t0) = 0, , (n-1)(t0) = 0,
278
where t0 is a point of an interval a t b on which the co-efficients a0, a1, , an are all continuous and a0(t) 0. Then (t) = 0 for all t such that a t b.
Theorem B
Let f1, f2, , fn be n solutions of the homogeneous linear differential equation
a0 (t )
dependent on [a, b] iff W( f1 , f 2 ,.., f n )(t ) = 0, t [a, b] , where W( f1 , f 2 ,.., f n )(t ) denotes the Wronskian of solutions f1, f2,., fn at t.
d dx P(t ) + Q(t ) x = 0, dt dt
having first derivative f ' on a t b, 2. and let f has an infinite number of zeros on a t b. Then f(t) = 0 for all t on a t b. (1)
Conclusion
Proof Since f has an infinite number of zeros on [a, b], by the Bolzano-Weierstrass
theorem the set of zeros has a limit point t0 [a, b]. Thus there exists a sequence {tn} of zeros which converges to t0 (where tn t0). Since f is continuous lim f(t) = f(t 0 ) ,
t t0
where t t0 through any sequence of points on [a, b]. Let t t0 through the sequence of zeros {tn}. Then
lim f (t ) = 0 = f (t0 ) .
t t0
f (t ) f (t0 ) , t t0
279
f (t ) f (t0 ) = 0 and thus f '(t0} = 0 t t0
Thus f is a solution of Equation (1) such that f(t0) = f '(t0) = 0. Hence, using theorem A f(t) = 0 for all t on a t b.
d dx P t ( ) + Q(t ) x = 0, dt dt
on the interval a t b.
(1)
Conclusion
where k is a constant.
(4)
f (t )
Integrating (5) from a to t, we obtain
(5)
or P(t)[f(t)g'(t) f '(t)g(t)] = P(a)[f(a)g'(a) f '(a)g(a)] = k (constant) and thus we have Abels formula (2).
280
Theorem 6.3 A. Hypothesis Let f and g be two solutions of
d dx P(t ) + Q(t ) x = 0 dt dt
such that f and g have a common zero on a t b. (1)
Conclusion
Conclusion Proof A.
Then g(t0) = 0
Let t0 [a, b] be the common zero of f and g, then f(t0)= g(t0) = 0. Letting t = t0 in the Abels formula, we obtain k = 0. Thus P(t)[f(t)g'(t) f '(t)g(t)] = 0 for t [a, b]. Since we have assumed throughout that P(t) > 0 on a t b, the quantity in brackets above must be zero for all t on a t b. But this quantity is W(f, g)(t). Thus by using the theorem B, the solution f and g are linearly dependent on a t b.
B.
for all t on a t b. Now given that neither f nor g is zero for all t on a t b. If c1 =0, then c2 g(t) = 0 for all t on a t b. Since g is not zero for all t on [a, b], we must have c2 = 0, which is a contradiction. Thus c1 0, and likewise c2 0. Thus
281
neither c1 nor c2 in (3) is zero. Since f(t0) = 0, letting t = t0 in (3) we have c2 g(t0) = 0. Thus g(t0) = 0.
Example 1
The equation
d2x +x=0 dt 2
is of the type (1) is above theorem, where P(t) = Q(t) = 1 on every interval a t b. The linearly dependent solutions are A sint and B sint, which have the common zeros t = n ( n= 0, 1, 2..) and no other zeros.
The Separation and Comparison Theorems Theorem 6.4 Sturm Separation Theorem Hypothesis
Let f and g be real linearly independent solutions of
d dx P(t ) + Q(t ) x = 0 dt dt
on the interval a t b.
(1)
Conclusion
Between any two consecutive zeros of f (or g) there is precisely one zero of g (or f).
Proof Let t1 and t2 be two consecutive zeros of f on [a, b]. Then by Theorem 6.3, Part
A, g(t1) 0 and g(t2) 0. Now assume that g has no zero in the open interval t1 < t < t2. Then since the solutions f and g have continuous derivatives on [a, b], the quotient f/g has a continuous derivative on the interval t1 t t2. Further, f(t)/g(t) is zero at the endpoints of this interval. Thus by Rolles theorem there exists , where t1 < < t2, such that
d f (t ) =0 dt g (t ) t =
But
282
d f (t ) W ( g , f )(t ) = [ g (t )]2 ; dt g (t )
and since f and g are linearly independent on a t b. W(f,g) (t) 0 on t1< t < t2 , i.e.
d f (t ) 0 dt g (t )
This contradiction shows that g has at least one zero in t1< t < t2 . Now suppose g has more than one zero in t1< t < t2, and let t3 and t4 be two such consecutive zeros of g. Then interchanging f and g and using the same arguments we can show that f must have at least one zero t5 in the open interval t3 < t < t4. Then t1 < t5 < t2 and so t1 and t2 would not be consecutive zeros of f, which is a contradiction to our assumption concerning t1 and t2. Thus g has precisely one zero in the open interval t1 < t < t2.
Figure
We may restate Theorem 6.4 in the following form: The zeros of one of two real linearly independent solutions of Equation (1) separate the zeros of the other solution (see figure).
Example 2
We have already observed that the equation
283
d2x +x=0 dt 2
is of the type (1) in above theorem. The functions f and g defined, respectively, by f(t) = sin t and g(t) = cos t are linearly independent solutions of this equation. Between any two consecutive zeros of one of these two linearly independent solutions, there is precisely one zero of the other solution. We know that zeros of f(t) are t = n, n = 0, 1, 2, . and zeros of g(t) are t = (2n+1) each other.
d dx P(t ) + Q1 (t ) x = 0 dt dt
2. Let 2 be a real solution of
(1)
d dx P t ( ) + Q2 (t ) x = 0 dt dt
3.
(2)
Let P has a continuous derivative and be such that P(t) > 0 and let Q1 and Q2 be continuous and such that Q2(t) > Q1(t).
Conclusion
If t1 and t2 are successive zeros of 1 on [a, b], then 2 has at least one zero at some point of the open interval t1 < t < t2.
Proof
Assume that 2 does not have a zero on the open interval t1 < t < t2. Then without loss in generality we can assume that 1(t) > 0 and 2(t) > 0 on t1 < t < t2. By hypothesis, we have
d [ P(t )1(t )] + Q1 (t )1 (t ) = 0 dt
(3)
284
d [ P(t )2 (t )] + Q2 (t )2 (t ) = 0 dt
(4)
for all t[a, b]. Multiply (3) by 2(t) and (4) by 1(t) and subtract to obtain
2 (t )
Since
(5)
2 (t )
t2
(6)
t1
t2
(7)
By hypothesis, P(t2) > 0. Since 1(t2) = 0 and 1(t) > 0 on t1 < t < t2, we have '1(t2) < 0. Since 2(t) > 0 on t1 < t < t2, we have 2(t2) 0. Thus P(t2) '1(t2)2(t2) 0. Similarly we have P(t1) '1(t1)2(t1) 0. Thus, the left member of (7) is not positive. But by hypothesis Q2(t) Q1(t) > 0 on t1 t t2, and so the right member of (7) is positive. Thus the assumption that 2 does not have a zero on the open interval t1 < t < t2, leads to a contradiction, and so 2 has a zero at some point of this open interval.
285
Hence the proof of the theorem. As a particular case of importance, suppose that the hypotheses of Theorem 6.5 are satisfied and that t1 is a zero of both 1 and 2. Then if t2 and are the next zeros of 1 and 2, respectively, we must have < t2.
Figure
Example 3
Consider the equation
d 2x + A2 x = 0 2 dt
and
d 2x + B2 x = 0 2 dt
where A and B are constants such that B > A > 0. The functions 1 and 2 defined respectively by 1(t) = sin At and 2(t) = sin Bt are real solutions of these respective equations. Consecutive zeros of sin At are
n and A
(n + 1) A
(n = 0, 1, 2,....)
Then by theorem 6.5, we are assured that sin Bt has at least one zero n such that
286
n (n + 1) < n < A A
(n = 0, 1, 2,....)
In particular, t = 0 is zero of both sin At and sin Bt. The next zero of sin At is /A, while the next zero of sin Bt is /B; and clearly /B < /A. This verifies the results of comparison Theorem.
Examples: (i)
for t 0, A and B are constants, then f(t) is non-oscillatory. (ii) Let f(t) = sin t, t 0.
t0
is called non-oscillatory if every solution u = u(t) of it, is non-oscillatory. Otherwise, differential equation is called oscillatory.
Example 1
u" + u = 0 is oscillatory.
Its general solutions is u(t) = A cos t + B sin t, t 0. W.l.o.g., we can assume that both A and B are non-zero constants, otherwise, u(t) is trivially oscillatory. In that case, u(t) has a zero at t = n + tan-1 (A/B), for n = 0, 1, 2, 3, ..
287
So, this equation is oscillatory.
Example 2
Its general solution is u(t) = Aet + Be-t, A & B are constants. This solution is nonoscillatory. Hence, this equation is non-oscillatory.
Definition Let f(t) and g(t) be two real valued functions defined and continuous in
some interval [a, b]. Then f(t) is said to oscillate more rapidly than g(t) if the number of zeros of f(t) in [a, b] exceed the number of zeros of g(t) in [a, b] by more than one.
Example 3
Then zeros of f(t) are only half as far apart as the zeros of g(t). So f(t) oscillates more rapidly than g(t) in the interval [0, 4].
Riccati Equations
Before we give the formal definition of Riccati equations, a little introduction may be helpful. Consider the first order differential equation
dy = f ( x, y ). dx
If we approximate f(x,y), while x is kept constant, we will get f(x, y) = P(x) + Q(x)y + R(x)y2 + . If we stop at y, we will get a linear equation. Riccati looked at the approximation to the second degree: he considered equations of the type
dy = P ( x) + Q( x) y + R ( x) y 2 . dx
Such type of equations bear his name, Riccati equations. They are nonlinear and do not fall under the category of any of the classical equations. In order to solve a Riccati equation, one will need a particular solution. Without knowing at least one solution, there is absolutely no chance to find any solutions to such an equation. Indeed, let y1 be a particular solution of
288
dy = P ( x) + Q ( x) y + R ( x) y 2 . dx
Consider the new function z defined by
z=
1 y y1
Then
dz = - (Q( x) + 2 y1 R ( x)) z R ( x) dx
which is a linear equation satisfied by the new function z. Once it is solved, we go back to y via the relation
y = y1 +
1 z
Example 1
dy = 2 y + y 2 , dx
knowing that y1 = 2 is a particular solution.
Answer
y1 is indeed a solution.
Consider the new function z defined by
z=
Then we have
1 y2
y = 2+
which implies
1 z
y'= -
z' z2
z' 1 1 2 = 2 2 + + 2 + z z z
289
Then easy calculations give Hence
z' 3 1 = + . z2 z z2
z=
Therefore, we have
1/ 3e3 x + C 1 = + Ce 3 x 3x 3 e
y = 2+
1 1 - + Ce 3 x 3
Note If one remembers the equation satisfied by z, then the solutions may be found
a bit faster. Indeed in this example, we have P(x) = -2, Q(x) = -1, and R(x) = 1. Hence the linear equation satisfied by the new function z, is
dz = (Q( x) + 2 y1 R ( x)) z R ( x) = (1 + 4) z 1 = 3 z 1 dx
Example 2
Answer Check that sin(x) is indeed a particular solution of the given differential
equation. We also recognize that the equation is of Riccati type. Set
z=
which gives
1 y sin( x)
290
1 y = sin( x) + , z Hence
y ' = cos( x)
Substituting into the equation gives
z' z2
This is the linear equation satisfied by z. The integrating factor is u ( x) = e The general solution is cos( x ) dx
sin( x )
= e In (cos( x )) =
z=
1/ 2 sec 2 ( x)dx + C u ( x)
y = sin( x) +
1 1 sin( x) + C cos( x) 2
291
The initial condition y(0) = -1 implies IVP is y = sin( x) + 1 1 sin( x) cos( x) 2 . 1 = 1 or C= -1. Therefore, the solution to the C
Relationship between Riccati Equation and linear differential equation of second order
The importance of Riccati Equation in theory of differential equations is due to the following relationship between it and the general linear differential equation of second order. Consider the general Riccati Equation dy + Q ( x) y + R ( x) y 2 = P ( x) dx Let us make the transformation y= 1 du u ' = Ru dx Ru (2) (1)
The resulting equation is the following linear differential equation of second order R d 2u du ( R ' QR) PR 2u = 0 2 dx dx (3)
Conversely, there corresponds to the general homogenous linear differential equation of second order a Riccati equation. Given the equation A( x) We make the transformation du = ( Ry )u dx and obtain the following Riccati equation (5) d 2u du + B( x) + C ( x)u = 0 2 dx dx (4)
292
dy R ' B C + + y + Ry 2 = . dx R A AR Comparing (6) with (1), we get Q( x) = R '( x) B( x) + , R( x) A( x) P( x) = C ( x ) . A( x) R( x) (7) (6)
Since R(x) is an arbitrary function, we can determine it so that Q(x) is zero. In this case, (6) assumes the simpler form dy C + Ry 2 = , dx AR where B ( x) R(x) = exp dx . A( x) (8)
293
We do this by introducing the ``phase'' and the ``radius'' of a solution u(x). This is done in three steps.
A)
First apply the Prfer substitution P(x)u'(x) = r(x) cos (x); u(x) = r(x) sin (x) (2)
to the quantities in (1). For this, we introduce the new dependent variables r and as defined by the formulae r 2 = u 2 + P 2 (u ') 2 ; = arctan u . Pu' (3)
Figure The Poincar phase plane of the second order linear differential equation is
spanned by the amplitude u and its derivative u' (multiplied by the positive coefficient P). A solution to the differential equation is represented by an x-parametrized curve. The (Prfer) phase is the polar angle . (Without loss of generality one may always assume that u(x) is real. Indeed, if u(x) were a complex solution, then it would differ from a real one by a mere complex constant.) A solution u(x) can thus be pictured in this Poincar plane as a curve parametrized by the independent variable x. The transformation (Pu', u) (r,) is non-singular for all r 0. Furthermore, we always have r > 0 for any non-trivial solutions. Because if r(x) = 0, i.e., u(x) = 0 and u'(x) = 0 for some particular x then by the uniqueness theorem for second order linear o.d.e. u(x) = 0 x, i.e., we have the trivial solution.
294
B)
Second, obtain a system of first order o.d.e. which is equivalent to the given differential equation (1). (i) Differentiate the relation cot = Pu' u u instead. This yields the same result.) Pu'
= -Qor
d 1 = Q(x) sin 2 + cos 2 F(x,) dx P(x) This is Prfer's differential equation for the phase, the Prfer phase. (ii) Differentiate the relation r2 = u2 + (Pu')2 and obtain r dr = uu' + (Pu') (Pu') ' dx = u Pu' - Pu' Qu P r sin r cos - r cos Q r sin P
(4)
= or
dr 1 1 = - Q(x) r sin 2 . dx 2 P(x) This is Prfer's differential equation for the amplitude.
(5)
295
C)
Third, solve the system of Prfer equations (4) and (5). Doing so is equivalent
to solving the originally given equation (1). Any solution to the Prfer system determines a unique solution to the equation (1), and conversely. This system is called the Prfer system associated with the self adjoint differential equation (1). Of the two Prfer equations (4) and (5), the one for the phase (x) is obviously much more important, it determines the qualitative, e.g. oscillatory, behaviour of u(x). The feature which makes the phase equation so singularly attractive is that it is a first order equation which also is independent of the amplitude r(x). The amplitude r(x) has no influence whatsoever on the phase function (x). Once (x) is known from (4), the Prfer amplitude function r(x) is determined by integrating (5). One obtains r(x) = K exp where K = r(a) is the initial amplitude.
Note Each solution to the Prfer system (4) and (5), depends on two constants:
the initial amplitude K = r(a), the initial phase = (a). 2. Changing the constant K just multiplies the solution u(x) by a constant factor.
Thus the zeros of u(x) can be located by studying only the phase differential equation, d = F ( x, ) . dx Vibrations, oscillations, wiggles, rotations and undulations are all characterized by a changing phase. If the independent variable is the time, then this time, the measure of that aspect of change which permits an enumeration of states, manifests itself physically by the advance of the phase of an oscillating system.
Summary The phase of a system is the most direct way of characterizing its
oscillatory nature. For a linear 2nd order o.d.e., this means the Prfer phase (x), which obeys the first order differential equation.
296
d 1 = Q( x) sin 2 + cos 2 F ( x, ) dx P( x) It is obtained from the second order equation d d P( x) + Q( x) u ( x) = 0 dx dx
These equations make it clear that the zeros and the oscillatory behavior of u(x) are controlled by the phase function (x).
(3)
Therefore, the first equation of Prfer system becomes d = x cos 2 + x sin 2 = x . dx Solving it, we obtain (x) = x2 + C, 2 (5) (4)
where C is an arbitrary constant. Then, the second differential equation of the Prfer system gives dr 1 = [ x x ] r sin 2 = 0 dx 2
297
Integrating r(x) = constant = r(a) (let) (6)
Thus, we conclude from the Prfer transformation, that the solution u(x) = r sin, now becomes
x2 u ( x) = r (a ) sin + C . 2
(7)
Then, from (7) we get x2 u (a) sin + C 2 . u ( x) = 2 a sin + C 2 as the solution of given equation (1).
Lagrange Identity
Consider the pair of differential equations. d du p (t ) + q(t )u (t ) = f (t ), dt dt d dv p (t ) + q (t )v(t ) = g (t ), dt dt (1)
(2)
where f = f(t) and g = g(t) are continuous functions on interval I. Multiplying the second relation (2) by u(t), first (1) by v(t) and subtracting the results, we obtain d du dv p(t ) u v = gu fv. dt dt dt The relation (3) is called the Lagrange identity. Its integrated form
[ p (u v ' u ' v]ta = ( gu fv)ds ,
a t
(3)
(4)
298
where [a, t] I, is called Greens formula.
Summary
Some basic results of Sturm theory are presented for the self adjoint form of second order equation. Sturm separation and fundamental comparison theorems are proved, with illustrations. Riccati equations are solved and a relationship between the general linear differential equation of second order and Riccati equation is defined. In concluding part of the chapter, phase-plane method is developed to find out the solution of Sturm-Liouville equation, using the Prffer substitution.
Keywords
Sturm theory, Zeros of solutions, Abels formula, Riccati equation, Prffer transformation.
299
Written by: Sunita Pannu Vetted by: Kapil Kumar
Chapter-7
Introduction
In this chapter, we shall consider special kind of boundary value problems known as Sturm Liouville problems. These problems arise naturally, for instance, when separation of variables is applied to the wave equation, the potential equation or the diffusion equation. The study of these types of problems will introduce us to several important concepts including characteristic values, characteristic functions, orthogonality and orthonormality of functions, which are very useful in many applied problems.
Boundary Value Problems The problems that involve both a differential equation
and one or more supplementary conditions, which the solution of given differential equation must satisfy, are called boundary value problems. If all the associated supplementary conditions relate to one x-value, the problem is called an initial-value problem (or one point boundary value problem). If the conditions relate to two different x-values, the problem is called a two-point boundary value problem (or simply a boundary value problem).
Example 1
Solve the equation d2y + y =0, dx 2 y(1) = 3, y'(1) = -4 .
300
This problem consists of finding a solution of the given differential equation which assumes the value 3 at x = 1 and whose first derivative assumes the value -4 at x = 1. Both these conditions relate to one x-value, namely x = 1. Thus, this is an initial value problem and has a unique solution.
Example 2
Solve the equation d2y + y =0, dx 2 y(0) = 1, y =5 2
In this problem, we again want a solution of the same differential equation, but this time, the solution must assume the value 1 at x = 0 and 5 at x = conditions relate to the two different x values, 0 and boundary value problem and also has a unique solution.
. This is a (two-point)
Sturm-Liouville Equation
In mathematics and its applications, a classical Sturm Liouville equation, named after Jacques Charles Francois Sturm (1803-1855) and Joseph Liouville (18091882), is a real second-order linear differential equation of the form d dy p ( x) + [ q ( x) + r ( x) ] y = 0 dx dx (1)
where y is a function of the free variable x. Here the functions p(x) > 0 has a continuous derivative, q(x) and r(x) > 0 are specified at the outset, and in the simplest of cases are continuous on the finite closed interval [a, b]. In addition, the function y is typically required to satisfy some boundary conditions at a and b. The function r(x), is called the weight or density function. The value of is not specified in the equation; finding the values of for which there exists a non-trivial solution of (1) satisfying the boundary conditions is part of the problem called the Sturm-Liouville problem (S L).
301
Such value of when they exist are called the eigenvalues of the boundary value problem defined by (1) and the prescribed set of boundary conditions. The corresponding solutions (for such ) are the eigenfunctions of this problem.
A.
Liouville problem.
Definition
We consider a boundary value problem which consists of 1. A second order homogeneous linear differential equation of the form d dy p( x) + [q( x) + r ( x)] y = 0 dx dx (1)
where p, q and r are real functions such that p has a continuous derivative, q and r are continuous, and p(x) > 0 and r(x) > 0 for all x on a real interval a x b and is a parameter independent of x; and 2. Two supplementary conditions A1y(a) + A2y'(a) = 0, B1y(b) + B2y'(b) = 0, (2)
where A1, A2, B1 and B2 are real constants such that A1 and A2 are not both zero and B1 and B2 are not both zero. This type of boundary-value problem is called a Sturm-Liouville problem (or Sturm-Liouville system).
Note Two important special cases are those in which the supplementary conditions
(2) are either of the form y(a) = 0, or of the form y'(a) = 0, y'(b) = 0 (4) y(b) = 0 (3)
302
Example 3
The boundary-value problem d2y + y = 0 dx 2 y(0) = 0, y() = 0 (1) (2)
is a Sturm-Liouville problem. The differential equation (1) may be written d dy 1. + [ 0 + .1] y = 0 dx dx where p(x) = 1, q(x) = 0, and r(x) = 1. The supplementary conditions (2) are of the special form. Let us see what is involved in solving Sturm-Liouville problem. We must find a function f which satisfies the given differential equation and the two supplementary conditions. Clearly one solution of any problem of this type is the trivial solution such that (x) = 0 for all values of x. But, this trivial solution is not very useful. We shall search for nontrivial solutions of the problem. That is, we shall attempt to find functions, not identically zero, which satisfy the given differential equation and the two conditions. We shall see that the existence of such nontrivial solutions depends upon the value of the parameter in the differential equation.
Example 4
Find nontrivial solutions of the Sturm-Liouville problem d2y + y = 0 dx 2 y(0) = 0, y() = 0 (1) (2)
Solution
Consider separately the three cases = 0, < 0 and > 0. In each case we shall first find the general solution of the differential equation. We shall then attempt to determine the two arbitrary constants in this solution so that the supplementary conditions (2) are also satisfied.
303
Case 1 = 0. In this case the differential equation (1) reduces to
d2y = 0, dx 2 so the general solution is y = c1 + c2 x. (3)
We now apply the conditions (2) to the solution (3). We find that in order for the solution (3) to satisfy the conditions (2), we must have c1 = c2 = 0. But then the solution (3) becomes the solution y such that y(x) = 0 for all values of x. Thus if the parameter = 0, the only solution of the given problem is the trivial solution.
Case 2 < 0. The auxiliary equation of the differential equation (1) is m2 + = 0 and
has the roots Denoting
. Since in this case < 0, these roots are real and unequal.
by , we see that for < 0 the general solution of (1) is of the form
y = c1ex + c2e-x .
(4)
We now apply the conditions (2) to the solution (4). Applying the first condition y(0) = 0, we obtain c1 + c 2 = 0 Applying the second condition y() = 0, we find that c1e + c2e- = 0 (6) (5)
Thus for the solution (4) to satisfy the conditions (2), the constants c1 and c2 must satisfy the system of equations (5) and (6). Obviously c1 = c2 = 0 is the solution of this system; but these values of c1 and c2 would only give the trivial solution of the given problem. We must therefore seek nonzero values of c1 and c2 which satisfy (5) and (6). This system has nonzero solutions only if the determinant of coefficients is zero. Therefore, we must have
1 1 =0
But < 0 in this case. Thus there are no non-trivial solutions of the given problem in the case < 0.
304
Case 3 > 0. Since > 0 here, the roots of the auxiliary equation of (1) are
the conjugate complex numbers . Thus in this case the general solution of (1) is of the form y = c1 sin x + c2 cos x. (7)
We now apply the condition (2) to this general solution. Applying the first condition y(0) = 0, we obtain c1sin 0 + c2 cos 0 = 0 and hence c2 = 0. Applying the second condition y() = 0, we find that c1 sin Since c2 = 0, this reduces to c1 sin
+ c2 cos = 0.
= 0.
(8)
We can set c1 = 0 or we can set sin = 0. However, if we set c1 = 0, then (since c2 = 0 also) the solution (7) reduces immediately to the unwanted trivial solution. Thus to obtain a nontrivial solution we can not set c1 = 0 but rather we must set sin
= 0.
(9)
If k > 0, then sin k = 0 only if k is a positive integer n = 1, 2, 3. Thus in order to satisfy (9) we must have
differential equation (9) has a nontrivial solution of the form (7) satisfying the conditions (2), we must have = n2 , where n = 1, 2, 3, .. (10)
The parameter in (2) must be a member of the infinite sequence 1, 4, 9, 16, , n2,.........
Summary
not have a nontrivial solution; if > 0, a nontrivial solution can exist only if is one of the values given by (10). We now note that if is one of the values given by (10) , then the problem does have nontrivial solutions. From (7) we see that nontrivial solutions corresponding to = n2 (n = 1, 2, 3, . ) are given by
305
y = cn sin nx (n = 1, 2, 3, .) (11)
where cn (n = 1, 2, 3, . ) are arbitrary nonzero constants. That is, the functions defined by c1 sin x, c2 sin 2x, c3 sin 3x,.., where c1, c2, c3,... are arbitrary nonzero constants, are nontrivial solutions of the given problem.
B.
Liouville problem depends upon the value of the parameter in the differential equation of the problem. Those values of the parameter for which nontrivial solutions do exist, as well as the corresponding nontrivial solutions themselves, are singled out by the following definition:
Definition
Consider the Sturm-Liouville problem A, consisting of the differential equation (1) and the supplementary conditions (2). The values of the parameter for which there exist nontrivial solutions of the problem are called the characteristic values of the problem. The corresponding nontrivial solutions themselves are called the characteristic functions of the problem. The characteristic values are also called eigenvalues; and the characteristic functions are also called eigenfunctions.
Example 5
Consider again the Sturm-Liouville problem d2y + y = 0, dx 2 y(0) = 0, y() = 0 (1) (2)
In example (4) we found that the values of in (1) for which there exist nontrivial solutions of this problem are the values = n2 , where n = 1, 2, 3 (3)
These are the characteristic values of the problem under consideration. The characteristic functions of the problem are the corresponding nontrivial solutions y = cn sin nx (n = 1, 2, 3), (4)
306
where cn (n = 1, 2, 3) are arbitrary nonzero constants.
Example 6
Find the characteristic values and characteristic functions of the SturmLiouville problem d dy x + y = 0, dx x dx y'(1) = 0, y'(e2) = 0 (1) (2)
Solution
differential equation (1) reduces to d dy x = 0. dx dx The general solution of this differential equation is y = C ln |x| + C0, where C and C0 are arbitrary constants. If we apply the conditions (2) to this general solution, we find that both of them require that C = 0. Thus for = 0 we obtain the solutions y = C0, where C0 is an arbitrary constant. These are nontrivial solutions for all choices of C0 0. Thus = 0 is a characteristic value and the corresponding characteristic functions are given by y = C0, where C0 is an arbitrary nonzero constant. If > 0, we see that for x 0 this equation is equivalent to the Cauchy-Euler equation x2 d2y dy +x + y = 0. 2 dx dx (3)
Letting x = et, Equation (3) transforms into d2y + y = 0. dt 2 Since > 0, the general solution of (4) is of the form
y = c1 sin t+ c 2 cos t.
(4)
307
Thus for > 0 and x > 0, the general solution of (1) may be written
y = c1 sin ( ln x) + c 2 cos ( ln x).
(5)
We now apply the supplementary conditions (2). From (5) we find that c c dy = 1 cos ( ln x) 2 sin ( ln x) dx x x for x > 0. Applying the first condition y'(1) = 0 of (2) to (6), we have
c1 cos ( ln 1) c2 sin ( ln 1) = 0
(6)
or c1
(7)
Since c1 = 0, the choice c2 = 0 would lead to the trivial solution. We must have sin 2 = 0 and hence 2 = n , where n = 1, 2, 3. Thus in order to satisfy the second condition (2) nontrivially we must have
n2 (n = 1, 2,3......) 4
(8)
Corresponding to these values of we obtain for x > 0, the nontrivial solutions n ln x y = cn cos (n = 1, 2,3......), 2 where the cn(n = 1, 2, 3.) are arbitrary nonzero constants. Thus the values (9)
= 0, , 1, , 4,
1 4
9 4
25 n2 ,......., ,...., 4 4
308
given by (8) for n 0, are the characteristic values of the given problem. The functions. ln x 3ln x c0 , c1 cos , c2 cos(ln x), c3 cos ,......, 2 2 given by (9) for n 0, where c0, c1, c2, c3 are arbitrary nonzero constants, are the corresponding characteristic functions. We observe that in each of these problems the infinite set of characteristic values can be arranged in a monotonic increasing sequence 1 < 2 < 3 < such that n + as n + . For example, the characteristic values of the problem of Example 4 can be arranged in the monotonic increasing sequence 1 < 4 < 9 < 16 < . (10)
such that n = n2 + as n + . We also note that in each problem there is a one-parameter family of characteristic functions corresponding to each characteristic value, and any two characteristic functions corresponding to the same characteristic value are merely nonzero constant multiples of each other. For example, in the problem of Example 4, the one-parameter family of characteristic functions corresponding to the characteristic value n2 is cn sin nx, where cn 0 is the parameter.
fixed ) of a SLBV problem consisting of a differential equation. d dy p( x) + [q( x) + r ( x)] y = 0 dx dx and the boundary conditions y(a) = y(b) = 0. (2) Then any solution of (1) can be expressed as linear combination of y1 and y2. That is y(x, ) = A y1 (x, ) + B y2 (x, ). (3) (1)
309
The constants A and B are determined by the fact that y(x, ) in (3) also satisfies the boundary conditions (2), which leads to A y1 (a, ) + B y2 (a, ) = 0 A y1 (b, ) + B y2 (b, ) = 0 or in the form of matrix equation
y1 (a, ) y (b, ) 1
(4)
y2 ( a , ) y2 (b, )
A B
0 = 0
(5)
Now, condition for the existence of non-trivial solution of (4) is that the determinant of the matrix of coefficients in (5) vanishes. Otherwise, the only solution is A = B =0, which yields the trivial solution y(x) = 0. Thus
y1 (a, ) y1 (b, ) y2 ( a , ) y2 (b, )
= y1 (a, ) y2 (b, ) y1 (b, ) y2 (a, ) = 0 . (6)
Now, y1 (x, ) and y2 (x, ) being analytic functions of , the determinant itself is an analytic function of . Therefore, by the theory of complex-valued functions, the zeros of the determinant must be isolated. [The zeros of an analytic function are isolated]. Since the zeros of the determinant correspond to solutions of the SLBVP, we thus conclude that the eigenvalues of (1) and (2) are discrete. This completes the proof.
Exercises
Find the characteristic values and characteristic functions of each of the following Sturm-Liouville problems. 1. d2y + y = 0, dx 2 d2y + y = 0, dx 2 d2y + y = 0, dx 2 y(0) = 0, y = = 0. 2 y(L) = 0, where L > 0
2.
y(0) = 0,
3.
y(0) = 0,
y() y '() = 0
310
4. d dy x + y = 0, dx x dx y(1) = 0, y(e) = 0.
5.
y(0) = 0,
y(1) = 0.
Definition
Example 7
f ( x) g ( x) ( x)dx = 0.
The functions sin x and sin 2x are orthogonal with respect to the weight function having the constant value 1 on the interval 0 x , for
Definition
Let {n} n = 1, 2, 3, . be an infinite set of functions defined on the interval a x b. The set {n} is called an orthogonal system with respect to the weight function on a x b, if every two distinct functions of the set are orthogonal with respect to on a x b. That is, the set {n} is orthogonal with respect to on a x b if
Example 8
m ( x) n ( x) ( x)dx = 0
for m n.
Consider the infinite set of functions {n}, where n(x) = sin nx (n = 1, 2, 3, ..) on the interval 0 x . The set {n} is an orthogonal system with respect to the weight function having the constant value 1 on the interval 0 x , for
311
for m n.
Note The weight function (x) is not always equal to 1. B Orthogonality of Characteristic Functions
We now state and prove a basic theorem concerning the orthogonality of characteristic functions of a Sturm-Liouville problem, also know as Sturm-Liouville Theorem.
where p, q and r are real functions such that p has a continuous derivative, q and r are continuous, and p(x) > 0 and r(x) > 0 for all x on a real interval a x b; and is a parameter independent of x ; and 2. The conditions A1 y(a) + A2y'(a) = 0 B1 y(b) + B2y'(b) = 0, (2)
where A1, A2, B1 and B2 are real constants such that A1 and A2 are not both zero and B1 and B2 are not both zero. Let m and n be any two distinct characteristic values of this problem. Let m be a characteristic function corresponding to m and let n be a characteristic function corresponding to n.
Conclusion
Proof
312
function corresponding to n, the function n satisfies the differential equation (1) with = n. Thus, we have d ( x)] + [q( x) + m r ( x)]m ( x) = 0, [ p ( x)m dx d ( x)] + [q( x) + n r ( x)]n ( x) = 0, [ p( x)n dx (3)
(4)
for all x such that a x b. Multiplying both sides of (3) by n(x) and both sides of (4) by m(x) and then subtracting the results we obtain
n ( x)
m ( x ) n ( x ) r ( x ) dx = m ( x)
b a
n ( x )
b a
d ( x)] dx [ p( x)m dx
Integrating by parts, the right member of (5) becomes ( x ) p ( x)n ( x)m ( x ) dx n ( x) p( x)n ( x) m ( x ) p ( x ) n
b a
b a
b a
or ( x ) n ( x)m ( x) { p ( x) m ( x)n } a .
b
313
(a ) n ( a ) m (a )]. - p (a )[m ( a ) n
(6)
Since m and n are characteristic functions of the problem, they satisfy the supplementary conditions (2) of the problem. If A2 = B2 = 0 in (2), these conditions reduce to y(a) = 0, y(b) = 0. Then in this case m(a) = 0, m(b) = 0, n(a) = 0, and n(b) = 0, and so the right member of (6) is equal to zero. If A2 = 0 but B2 0 in (2), these conditions reduce to y(a) = 0, y(b) + y '(b) = 0, where = B1/B2. Then the second bracket in the right member of (6) is again equal to zero. Also, the first bracket in this member may be written as
(b)]m ( b ) [m ( b ) + m (b)]n ( b ) , [n ( b ) + n
and so it is also equal to zero. Thus in this case the right member of (6) is equal to zero. Similarly, if either A2 0, B2 = 0 or A2 0, B2 0 in (2), then the right member of (6) is equal to zero (prove it). Thus in all cases the right member of (6) is equal to zero and so
( m n )
m ( x ) n ( x ) r ( x ) dx = 0.
a x b.
m ( x ) n ( x ) r ( x ) dx = 0.
This proves that eigen functions m and n are orthogonal with respect to r on
Remark Let {n} be the infinite set of characteristic values of a Sturm Liouville
problem, arranged in a monotonic increasing sequence 1 < 2 < 3 < .. For each n = 1, 2, 3, let n be one of the characteristic functions corresponding to the characteristic value n. Then above Theorem implies at once that the infinite set of characteristic functions 1, 2, 3, .. is an orthogonal system with respect to the weight function r on a x b.
314
Example 9
Consider again the Sturm-Liouville problem d2y + y = 0, dx 2 y(0) = 0, y() = 0 (1) (2)
which we have already investigated. Corresponding to each characteristic value n = n2 (n = 1, 2, 3), we found the characteristic functions cn sin nx (n = 1, 2, 3) where cn(n = 1, 2, 3) are arbitrary nonzero constants. Let {n} denotes the infinite set of characteristic functions for which cn = 1 (n = 1, 2, 3). That is n(x) = sin nx (0 x ; n = 1, 2, 3.).
Then by the above Theorem, the infinite set {n} is an orthogonal system with respect to the weight function r, where r(x) = 1 for all x, on the interval 0 x . That is
(3)
for m = 1, 2, 3, ; n = 1, 2, 3, ; m n.
Theorem 7.3 Prove that the eigen values of a SLBVP are real Proof Let n be an eigen value corresponding to the eigen function n(x) of the given
SLBVP. Then, by definition d d p(x) n + [ q(x) + n r(x) ]n ( x) = 0 dx dx and A1 n (a) + A2 'n (a) = 0 B1 n (b) + B2 'n (b) = 0 (2) (1)
We know that p(x), q(x) and r(x) are real valued functions of x over the interval [a, b]. So, taking the complex conjugate of (1) and (2), we obtain
dn d q(x) + n r(x) n ( x ) = 0 p(x) + dx dx
(3)
and
(a ) = 0 A1 n ( a ) + A 2 n
315
( b) = 0 , B1 n ( b ) + B2 n
(4)
where A1, B1, A2, B2 are real constants. Thus, n is also an eigen function, corresponding to an eigen value n of the same SLBVP. So from Sturm Liouville Theorem, it follows that
( n n )
r ( x ) ( x ) ( x ) dx = 0
b a
n n
or
( n n )
r ( x ) | ( x ) |
b a
n
dx = 0.
(5)
Since, r(x) > 0 and |n(x)| 0, being a nontrivial solution, so we must have ( n n ) = 0 or n = n
Exercises
Verify the validity of the conclusion of S-L Theorem for the characteristic functions of the following Sturn Liouville problems. 1. d2y + y = 0, dx 2 d dy x + y = 0, dx x dx y(0) = 0, y(/2) = 0. y'(e2) = 0.
2.
y'(1) = 0,
Orthonormal Systems
Definition
A function f is called normalized with respect to the weight function on the interval a x b if and only if
Example 10
f ( x ) ( x ) dx = 1.
2
The function f ( x ) = 2/ sin x is normalized with respect to the weight function having the constant value 1 on the interval 0 x , for
316
Definition
2 2 sin x (1) dx =
sin 2 x dx =
2 . = 1. 2
Let {n} (n = 1, 2, 3, ......) be an infinite set of functions defined on the interval a x b. The set {n} is called an orthonormal system with respect to the weight function on a x b if (1) (2) It is an orthogonal system with respect to on a x b. Every function of the system is normalized with respect to on a x b. That is, the set {n} is orthonormal with respect to on a x b if
a m ( x ) n ( x ) ( x ) dx =
b
0 1
for m n, for m = n.
Example 11
Consider the infinite set of functions {n}, where n(x) = 2/ sin nx (n = 1, 2, 3, ......) on the interval 0 x . The set {n} is an orthogonal system with respect to the weight function having the constant value 1 on the interval 0 x , for
Further, every function of the system is normalized with respect to this weight function on 0 x , for
2 sin nx (1) dx = 1.
Thus the set {n} is an orthonormal system with respect to the weight function having the constant value 1 on 0 x .
Remark Consider the Sturm-Liouville problem. Let {n} be the infinite set of
characteristic values of this problem, such that 1 < 2 < 3 < ....... If n (n = 1, 2, 3, ......) is one of the characteristic functions corresponding to the characteristic value n,
317
then we know from S-L Theorem that the infinite set of characteristic functions 1, 2, 3, ....... is an orthogonal system with respect to the weight function r on a x b. But this set of characteristic functions is not necessarily orthonormal with respect to r on a x b. Now, if n is one of the characteristic functions corresponding to n, then knn, where kn is an arbitrary nonzero constant, is also a characteristic function corresponding to n. Thus from the given set of characteristic functions 1, 2, 3, ....... we can form a set of "new" characteristic functions k11, k22, k33, ....... and this "new" set is also orthogonal with respect to r on a x b. Now if we can choose the constants k1, k2, k3, ........ in such a way that every characteristic function of the "new" set is also normalized with respect to r on a x b, then the "new" set of characteristic functions k11, k22, k33, ....... will be an orthonormal system with respect to r on a x b. The constants k1, k2, k3, ........ can indeed be chosen so that the set k11, k22, k33, ....... is orthonormal. As the function r in the given differential equation is such that r(x) > 0 for all x on the interval a x b. Also by definition no characteristic function n (n = 1, 2. 3,...) is identically zero on a x b. Therefore
[ (x)]
b a n
r(x) dx = K n > 0
2
(n = 1, 2, 3, ....), (n = 1, 2, 3, ....).
1 n (x) r(x) dx = 1 a Kn
b
1 1 , K1
1 2 , K2
1 3 ,....... K3
is an orthonormal set with respect to r on a x b. Thus, from a given set of orthogonal characteristic functions 1, 2, 3, ......., we can always form the set of orthonormal characteristic functions k11, k22, k33, ....... where kn = 1 = Kn 1 (n = 1, 2, 3......)
[n (x)] r(x) dx
2
318
Example 12
The Sturm-Liouville problem of example 9 has the set of orthogonal characteristic functions {n}, where n(x) = cn sin nx (n = 1, 2, 3,...... ; 0 x ) and cn (n = 1, 2, 3....) are nonzero constants. We now form the sequence of orthonormal characteristic functions {knn}, where kn is defined as above. We have Kn =
( cn sin nx ) (1) dx =
2
c2 n , 2
kn = 1 knn(x) = c n
1 1 = cn Kn
2 , 2 sin nx
2 ( c n sin nx ) =
(n = 1, 2, 3.....)
Thus the Sturm-Liouville problem under consideration has the set of orthonormal characteristic functions {n}, where n(x) = 2/ sin nx (n = 1, 2, 3,
......; 0 x ). We see that this is the set of orthonormal functions considered in Example 10.
Summary
In the course of this chapter, Sturm-Liouville problems are studied and nontrivial solutions of such problems are found. Characteristic values of a SLBVP are found to be real and discrete. A basic theorem known as Sturm-Liouville theorem concerning the orthogonality of characteristic functions is proved and its validity is verified for the characteristic functions of a SLBVP.
Keywords
Boundary value problems, Characteristic values, Characteristic functions, Orthogonality, Orthonormality.
319
References/Suggested readings
1. E.A. Coddington and N. Levinson. Theory of Ordinary Differential Equations, McGraw Hill, NY, 1955. 2. G. Birkhoff and G.C., Rota. Ordinary Differential Equations, John Wiley and Sons inc., NY, 1978. 3. 4. S.L. Ross. Differential Equations, John Wiley and Sons inc., NY, 1984 W.E. Boyce and R.C. Diprima. Elementary Differential Equations and Boundary Value Problems, John Wiley and sons Inc. NY, 1986. 5. 6. P. Hartman. Ordinary Differential Equations, John Wiley & Sons, NY, 1964. D. Somasundaram. Ordinary Differential Equations, Narosa Publishing House, Chennai, 2008. 7. G.F. Simmons. Differential Equations with Applications and Historical Notes, McGraw Hill, 1991. 8. I.N. Sneddon. Elements of Partial Differential Equations, McGraw Hill, Inc., Japan, 1984.