DIFFERENTIAL EQUATIONS. A principal model of physical phenomena.

Transcription

1 DIFFERENTIAL EQUATIONS A principal model of physical phenomena. The ordinary differential equation: The initial value: y = f(x, y) y(x 0 )=Y 0 Find a solution Y (x) onsomeintervalx 0 x b. Together these two conditions constitute an initial value problem. We will study methods for solving systems of first order equations, but we begin with a single equation. Many of the crucial ideas in the numerical analysis arise from properties of the original equation.

2 SPECIAL CASES 1. y (x) =λy(x)+b(x), x x 0 General solution: Y (x) =ce λx + x x 0 e λ(x t) b(t)dt with c arbitrary. With y(x 0 )=Y 0, Y (x) =Y 0 e λ(x x x 0) + e λ(x t) b(t)dt x 0 2. y (x) =ay 2 General solution: Y (x) = 1 ax + c, With y(x 0 )=Y 0 0,use c arbitrary c = ax 0 1 Y 0

3 3. y (x) = [y(x)] 2 + y(x) General solution: Y (x) = 1 1+ce x 4. Separable equations: y (x) =g(y(x))h(x) General solution: Write 1 dy g(y) dx = h(x) Let z = y(x), dz= y (x)dx. Evaluate the integrals in dz g(z) = h(x)dx Replace z by Y (x) andsolvefory (x), if possible.

4 DIRECTION FIELDS At each point (x, y) at which the function f is defined, evaluate it to get f(x, y). Then draw in a small line segement at this point with slope f(x, y). With enough of these, we have a picture of how the solutions of the differential equation behave. y = f(x, y) Consider the differential equation y = y +2cosx We can draw direction fields by hand by the method described in the book; or we can use the Matlab program I am providing.

5 SOLVABILITY THEORY Consider whether there is a function Y (x) whichsatisfies y = f(x, y), x x 0, y(x 0 )=Y 0 (1) Assume there is some open set D containing (x 0,Y 0 ) for which: 1. If two points (x, y) and(x, z) are contained in D, then the line segment joining them is also contained in D. 2. f(x, y) is continuous for all points (x, y) contained in D. 3. f(x, y)/ y is continuous for all points (x, y) contained in D. Then there is an interval [c, d] containing x 0 and there is a unique function Y (x) definedon[c, d] whichsatisfies (1), with the graph of Y (x) contained in D.

6 x Figure 1: Direction field for y = y +2cosx

7 THE LIPSCHITZ CONDITION The preceding condition on the partial derivative of f is an easy way to specify the following condition is satisfied, and it is the condition which is really needed. The Lipschitz condition: There is a non-negative constant K for which f(x, y) f(x, z) K y z for all points (x, y), (x, z) in the region D. In practice, we can use K = max (x,y) D f(x, y) The Lipschitz condition returns throughout our treatment of both the theory of differential equations and the theory of the numerical methods for their solution. y

8 EXAMPLE Let α>0 be a given constant, and consider solving y = 2x α 2y2, x 0, y(0) = 1 Then the partial derivative is f y (x, y) = 4xy α 2 and f y (0, 1) = 0. Thus f y (x, y) issmallfor(x, y) near to (0, 1), and it is continuous for all (x, y). Choose D = {(x, y) : x 1, y B} for some B>0. Then there is a solution Y (x) on some interval [c, d] containing x 0 =0. Howbigis [c, d]? In this case, α2 Y (x) = α 2 x 2, α <x<α If α is small, then the interval is small.

9 IMPROVED SOLVABILITY THEORY Assume there is a Lipschitz constant K for which f satisfies f(x, y) f(x, z) K y z for all (x, y), (x, z) satisfying x 0 x b, Then the initial value problem <y,z< y = f(x, y), x 0 x b, y(x 0 )=Y 0 has a solution Y (x) on the entire interval x 0 x b. Example: Considery = y+g(x)withg(x) continuous for all x. Then y = y + g(x), y(x 0 )=Y 0 has a solution Y (x) has a unique continuous solution for <x<.

10 STABILITY The concept of stability refers in a loose sense to what happens to the solution Y (x) ofaninitialvalue problem if we make a small change in the data, which includes both the differential equation and the initial value. If small changes in the data lead to large changes in the solution, then we say the initial value problem is unstable ; whereas if small changes in the data lead to small changes in the solution, we call the problem stable.

11 EXAMPLE Consider solving y = 100y 101e x, y(0) = 1 (2) This has a solution of Y (x) =e x. Now consider the perturbed problem y = 100y 101e x, y(0) = 1 + ɛ where ɛ is some small number. The solution of this is Y ɛ (x) =e x + ɛe 100x,and Y ɛ (x) Y (x) =ɛe 100x Thus Y ɛ (x) Y (x) increases very rapidly as x increases, and we say (2) is an unstable problem. We must now define these concepts with a bit more care.

12 STABILITY Consider the initial value problem y = f(x, y), x 0 x 0 b, y(x 0 )=Y 0 and denote its solution by Y (x). problem Now consider the y = f(x, y), x 0 x b, y(x 0 )=Y 0 + ɛ and denote its solution by Y (x; ɛ). We want to study the behaviour of Z(x; ɛ) Y (x; ɛ) Y (x) to see how it changes as x increases. To do this, we first derive an equation for Z(x; ɛ).

13 Substitute the solutions into the above equations: Y (x) =f(x, Y (x)), Y(x 0 )=Y 0 (3) Y (x; ɛ) =f(x, Y (x; ɛ)), Y(x 0 ; ɛ) =Y 0 + ɛ (4) both for x 0 x b. Subtracting (3) from (4), we have Z (x; ɛ) =f(x, Y (x; ɛ)) f(x, Y (x)), Z(x 0 ; ɛ) =ɛ We can apply the mean-value theorem to the right side to obtain the perturbation equation Z (x; ɛ) = = f(x, ζ) [Y (x; ɛ) Y (x)], Z(x 0 ; ɛ) =ɛ y f(x, ζ) Z(x; ɛ) y (5) where ζ ζ(x; ɛ) is some unknown number between Y (x) andy (x; ɛ).

14 In the case ɛ is small, and with x close to x 0,wehave ζ Y (x) and Z (x; ɛ). = or even f(x, Y (x)) Z(x; ɛ), Z(x 0 ; ɛ) =ɛ y Z (x; ɛ) =. f(x, Y 0) Z(x; ɛ), Z(x 0 ; ɛ) =ɛ y Note that both of these has the form Z (x). = a(x)z(x), x x 0 ; Z(x 0 )=ɛ a simple linear differential equation.

15 EXAMPLE Consider solving In this y = a(x)+cosy, y(0) = 0 f(x, y) =a(x)+cosy f(x, y) = sin y y The equation (5) becomes Z (x; ɛ) = sin (ζ(x; ɛ)) Z(x; ɛ), x 0; Z(0,ɛ)=ɛ with ζ(x; ɛ) some unknown number between Y (x) and Y (x; ɛ). This can be solved as [ x ] Z(x; ɛ) =ɛ exp sin ζ(t; ɛ) dt Since 1 sin (ζ(t; ɛ)) 1, we have Z(x; ɛ) ɛ e x, x 0 0

16 In general, the equation Z (x; ɛ) = can be solved to get f(x, ζ) Z(x; ɛ), x x 0 ; Z(x 0 ; ɛ) =ɛ y [ x f(t, ζ(t; ɛ)) Z(x; ɛ) =ɛ exp dt, x x 0 (6) x 0 y We can use this to derive a variety of results, depending on the assumptions we make about the partial derivative f y (x, y) f(x, y) y ]

17 For example, suppose f(x, y) y 0, x 0 x b, <y< Then using the perturbation equation (6), we have because x Z(x; ɛ) ɛ x 0 f(t, ζ(t; ɛ)) y dt 0 e negative quantity 1 Example: Consider the equation Then y = y 3 + g(x) f y (x, y) = 3y 2 0 for all (x, y). For this equation, the perturbed solution satisfies Y (x; ɛ) Y (x) ɛ, x x 0

18 If all we know is that K max x 0 x b <y< then we can say only that Z(x; ɛ) ɛ exp [ x x 0 [ x ɛ exp Kdt x 0 = ɛ e K(x x 0) f(x, y) y < f(t, ζ(t; ɛ)) y dt ] ] (7) This might increase quite rapidly if K is a large number, or even moderately large.

19 WHEN ARE WE IN TROUBLE? If the partial derivative f y (x, y) evaluated at y = Y (x), namely f y (x, Y (x)) is large, then the perturbation [ x f(t, ζ(t; ɛ)) Z(x; ɛ) =ɛ exp dt, x x 0 (8) x 0 y will grow rapidly as x increases. In the problem y = λy + g(x), y(0) = Y 0 (9) we have f y (x, y) =λ, and Z(x; ɛ) =ɛ exp [ x x 0 λdt ] ] = ɛe λ(x x 0) If λ<0, the perturbation becomes smaller with increasing x. But if λ>0, then the perturbation increases as x increases; and if λ is large in magnitude, then the perturbation increases rapidly.

20 GENERAL DISCUSSION AND NOTATION The words stable and unstable are not sufficiently descriptive of what happens in practice. In y = λy + g(x), y(0) = Y 0 (10) as λ increases, λ > 0, the perturbation becomes worse. In Chapter 1, we introduced the idea of condition number. Small condition numbers were associated with well-conditioned problems, and large condition numbers were associated with ill-conditioned problems. We could introduce condition numbers for our initial value problem y = f(x, y), x 0 x b, y(x 0 )=Y 0 by using our perturbation equation for Z(x; ɛ). This is seldom done in practice. Instead we note that some problems are more stable or well-conditioned than others. With (10), we look at the size of λ. Withageneral problem, we look at the size of f y (x, y) forthe values of y around Y (x).

21 HIGHER ORDER EQUATIONS Consider the third order equation y (x) 2y y 2 =cosx, x x 0 The initial conditions for this are specifications of y(x 0 ), y (x 0 ), y (x 0 ). To rewrite this as a system of first order equations, introduce y 1 = y, y 2 = y, y 3 = y Then these new functions satisfy the first order system y 1 = y 2, y 1 (x 0 )=y(x 0 ) y 2 = y 3, y 2 (x 0 )=y (x 0 ) y 3 = cosx +2y 2y 2 1, y 3(x 0 )=y (x 0 ) We do other higher order equations in a similar manner.

22 EXAMPLE Consider the vector differential equation mr (t) = mmg r(t) 3 r(t), r(t) =x(t)i + y(t)j Then we rewrite this as a system of first order equations by introducing y 1 = x, y 2 = x y 3 = y, y 4 = y This leads to the first order system with r = ( y y2 3)1 2. y 1 = y 2 y 2 = MG r 3 y 1 y 3 = y 4 y 4 = MG r 3 y 3

23 FIRST ORDER SYSTEMS The general form of a system of m first order differential equations is y 1 =. f 1 (x, y 1,..., y m ), y 1 (x 0 )=Y 1,0. y m = f m (x, y 1,..., y m ), y m (x 0 )=Y m,0 We can make this look like our earlier single equation by introducing matrix-vector notation. Let y = y 1. y m f(x, y) =, Y 0 = Y 1,0. Y m,0 f 1 (x, y 1,..., y m ). f m (x, y 1,..., y m ) Then the linear system can be written as y = f(x, y), x x 0 ; y(x 0 )=Y 0

24 The earlier role of f y (x, y) is played by the matrix f y (x, y) = f 1 y 1. f m y 1 f 1 y m. f m y m We will return to this when needing it in a later section. The analogue of the first order linear equation is y = A(x)y + g(x), x x 0 ; y(x 0 )=Y 0 with A(x) amatrixoforderm m.