SYSTEMS OF NONLINEAR EQUATIONS

Transcription

1 SYSTEMS OF NONLINEAR EQUATIONS Widely used in the mathematical modeling of real world phenomena. We introduce some numerical methods for their solution. For better intuition, we examine systems of two nonlinear equations and numerical methods for their solution. We then generalize to systems of an arbitrary order. The Problem: Consider solving a system of two nonlinear equations f (x, y) = 0 g(x, y) = 0 (1)

2 Example Consider solving the system f (x, y) x 2 + 4y 2 9 = 0 g(x, y) 18y 14x = 0 (2) z y x 0 2 Figure : Graph of z = x 2 + 4y 2 9, along with z = 0 on that surface

3 To visualize the points (x, y) that satisfy simultaneously both equations in (1), look at the intersection of the zero curves of the functions f and g. The next figure illustrates this. The zero curves intersect at four points, each of which corresponds to a solution of the system (2). For example, there is a solution near (1, 1). 18y 14x 2 +45=0 3 y x 1 2 x 2 +4y 2 9=0 3 Figure : Graphs of f (x, y) = 0 and g(x, y) = 0

4 TANGENT PLANE APPROXIMATION The graph of z = f (x, y) is a surface in xyz-space. For (x, y) (x 0, y 0 ), we approximate f (x, y) by a plane that is tangent to the surface at the point (x 0, y 0, f (x 0, y 0 )). The equation of this plane is z = p(x, y) with p(x, y) f (x 0, y 0 ) + (x x 0 )f x (x 0, y 0 ) + (y y 0 )f y (x 0, y 0 ), (3) f x (x, y) = f (x, y), f y (x, y) = x f (x, y), y the partial derivatives of f with respect to x and y, respectively. If (x, y) (x 0, y 0 ), then f (x, y) p(x, y) For functions of two variables, p(x, y) is the linear Taylor polynomial approximation to f (x, y).

5 NEWTON S METHOD Let α = (ξ, η) denote a solution of the system f (x, y) = 0 g(x, y) = 0 Let (x 0, y 0 ) (ξ, η) be an initial guess at the solution. Approximate the surface z = f (x, y) with the tangent plane at (x 0, y 0, f (x 0, y 0 )). If f (x 0, y 0 ) is sufficiently close to zero, then the zero curve of p(x, y) will be an approximation of the zero curve of f (x, y) for those points (x, y) near (x 0, y 0 ). Because the graph of z = p(x, y) is a plane, its zero curve is simply a straight line.

6 Example Consider f (x, y) x 2 + 4y 2 9. Then f x (x, y) = 2x, f y (x, y) = 8y. At (x 0, y 0 ) = (1, 1), f (x 0, y 0 ) = 4, f x (x 0, y 0 ) = 2, f y (x 0, y 0 ) = 8 At (1, 1, 4) the tangent plane to the surface z = f (x, y) is z = p(x, y) 4 + 2(x 1) 8(y + 1) y 1 2 x 1 (x 0,y 0 ) f(x,y)=0 p(x,y)=0 2 Figure : Zero curves of f (x, y) and p(x, y) for (x, y) near (x 0, y 0 )

7 The tangent plane to the surface z = g(x, y) at (x 0, y 0, g(x 0, y 0 )) has the equation z = q(x, y) with q(x, y) g(x 0, y 0 ) + (x x 0 )g x (x 0, y 0 ) + (y y 0 )g y (x 0, y 0 ) Recall that the solution α = (ξ, η) to f (x, y) = 0 g(x, y) = 0 is the intersection of the zero curves of z = f (x, y) and z = g(x, y). Approximate these zero curves by those of the tangent planes z = p(x, y) and z = q(x, y). The intersection of these latter zero curves gives an approximate solution to the above nonlinear system. Denote by (x 1, y 1 ) the solution to p(x, y) = 0 q(x, y) = 0

8 Example Return to the equations f (x, y) x 2 + 4y 2 9 = 0 g(x, y) 18y 14x = 0 with (x 0, y 0 ) = (1, 1). y 1 2 x g(x,y)=0 1 (x 0,y 0 ) q(x,y)=0 f(x,y)=0 p(x,y)=0 (x 1,y 1 ) 2 Figure : Use of zero curves of tangent plane approximations

9 Calculating (x 1, y 1 ). To find the intersection of the zero curves of the tangent planes, we must solve the linear system f (x 0, y 0 ) + (x x 0 )f x (x 0, y 0 ) + (y y 0 )f y (x 0, y 0 ) = 0 g(x 0, y 0 ) + (x x 0 )g x (x 0, y 0 ) + (y y 0 )g y (x 0, y 0 ) = 0 The solution is denoted by (x 1, y 1 ). In matrix form, [ ] [ ] [ ] fx (x 0, y 0 ) f y (x 0, y 0 ) x x0 f (x0, y = 0 ) g x (x 0, y 0 ) g y (x 0, y 0 ) y y 0 g(x 0, y 0 ) It is actually computed as follows. Define δ x and δ y to be the solution of the linear system [ ] [ ] [ ] fx (x 0, y 0 ) f y (x 0, y 0 ) δx f (x0, y = 0 ) g x (x 0, y 0 ) g y (x 0, y 0 ) δ y g(x 0, y 0 ) [ ] [ ] [ ] x1 x0 δx = + y 1 Usually the point (x 1, y 1 ) is closer to the solution α than is the original point (x 0, y 0 ). y 0 δ y

10 Continue this process, using (x 1, y 1 ) as a new initial guess. Obtain an improved estimate (x 2, y 2 ). This iteration process is continued until a solution with sufficient accuracy is obtained. The general iteration is given by [ ] [ ] [ ] fx (x k, y k ) f y (x k, y k ) δx,k f (xk, y k ) = g x (x k, y k ) g y (x k, y k ) δ y,k g(x k, y k ) ] ] ] (4) [ xk+1 y k+1 = [ xk y k + [ δx,k δ y,k This is Newton s method for solving f (x, y) = 0 g(x, y) = 0, k = 0, 1,... Many numerical methods for solving nonlinear systems are variations on Newton s method.

11 Example Consider again the system f (x, y) x 2 + 4y 2 9 = 0 g(x, y) 18y 14x = 0 Newton s method (4) becomes [ ] [ ] [ 2xk 8y k δx,k xk 2 = + 4y k 2 9 ] 28x k 18 δ y,k 18y k 14xk ] ] ] (5) [ xk+1 y k+1 = [ xk y k + [ δx,k δ y,k Choose (x 0, y 0 ) = (1, 1). The resulting Newton iterates are given in Table 1, along with Error = α (x k, y k ) max{ ξ x k, η y k }

12 Table : Newton iterates for the system (5) k x k y k Error E E E E E E 16 The final iterate is accurate to the precision of the computer arithmetic.

13 AN ALTERNATIVE NOTATION We introduce a more general notation for the preceding work. The problem to be solved is F 1 (x 1, x 2 ) = 0 F 2 (x 1, x 2 ) = 0 Introduce [ ] [ ] F 1 F 1 x1 F1 (x 1, x 2 ) x =, F (x) =, F (x) = x 1 x 2 x 2 F 2 (x 1, x 2 ) F 2 F 2 x 1 x 2 F (x) is called the Frechet derivative of F (x). It is a generalization to higher dimensions of the ordinary derivative of a function of one variable. The system (6) can now be written as (6) F (x) = 0 (7) A solution of this equation will be denoted by α.

14 Newton s method becomes for k = 0, 1,.... F (x (k) )δ (k) = F (x (k) ) x (k+1) = x (k) + δ (k) (8) A shorter and mathematically equivalent form: for k = 0, 1,.... [ 1 x (k+1) = x (k) F (x )] (k) F (x (k) ) (9) This last formula is often used in discussing and analyzing Newton s method for nonlinear systems. But (8) is used for practical computations, since it is usually less expensive to solve a linear system than to find the inverse of the coefficient matrix. Note the analogy of (9) with Newton s method for a single equation.

15 THE GENERAL NEWTON METHOD Consider the system of n nonlinear equations F 1 (x 1,..., x n ) = 0. F n (x 1,..., x n ) = 0 (10) Define x = x 1. x n, F (x) = F 1 (x 1,..., x n ). F n (x 1,..., x n ), F (x) = F 1 x 1. F n x 1 F 1 x n. F n x n The nonlinear system (10) can be written as F (x) = 0 Its solution is denoted by α R n.

16 Newton s method is F (x (k) )δ (k) = F (x (k) ) x (k+1) = x (k) + δ (k), k = 0, 1,... (11) Alternatively, as before, [ 1 x (k+1) = x (k) F (x )] (k) F (x (k) ), k = 0, 1,... (12) This formula is often used in theoretical discussions of Newton s method for nonlinear systems. But (11) is used for practical computations, since it is usually less expensive to solve a linear system than to find the inverse of the coefficient matrix. CONVERGENCE. Under suitable hypotheses, it can be shown that there is a c > 0 for which α x (k+1) α c x (k) 2, k = 0, 1,... α x (k) max α i x (k) 1 i n Newton s method is quadratically convergent. i