'In104 Nonlinear Algebraic Equations $ Nonlinear Algebraic Equations Let f = f (x) be a nonlinear function 6f -x then the problem of nding x = x∗ such that f (x∗) = 0, is referred to as a nonlinear algebraic equation. & 1 % 'In104 Nonlinear Algebraic Equations $ The linear case ( ) = αx + β, where α and β are constants, is very simple: f x 6f ( ) = 0, implies that f x∗ -x r −β/α = −β/α. The case of a second order polynomial is also quite simple. Given f (x) = αx + βx + γ, for constants α, β and γ , we have that x∗ 2 6f implies ( ) = 0, f x∗ x∗ & = r −β ± 2 q r 4 β 2 − αγ 2α -x . % 'In104 Nonlinear Algebraic Equations $ For more complicated functions, it may be very hard to derive analytical formulas for the solution. Also, it may be hard to know whether there is a point x∗ such that f (x∗) = 0 6 - & 3 % 'In104 Nonlinear Algebraic Equations $ Basic observation If f is a continuous function and there are two points a < b such that f (a) · f (b) < 0, then there is a point x∗ ∈ (a, b) such that f (x∗) = 0 6f s a t x∗ b -x s & 4 % 'In104 Nonlinear Algebraic Equations $ How can we compute x∗ numerically? & • The bisection method • The secant method • Newton's method 5 % 'In104 Nonlinear Algebraic Equations $ The Bisection Method Given a smooth function f = f (x) and two values a < b such that f (a) · f (b) < 0. We assume that there is only one point x∗ in (a, b) such that f (x∗) = 0 Dene 1 c = (a + b). 2 6f s a cs s b -x s If f (a) · f (c) < 0 then put b = c. If f (a) · f (c) > 0 then put a = c. 6 & % 'In104 Nonlinear Algebraic Equations $ Repeat the process until |f (c)| is suciently small. Algorithm Given a, b such that f (a) · f (b) < 0 and given a tolerance . Dene c = (a + b). 1 2 while |f (c)| > do if f (a) · f (c) < 0 then b = c else a = c c end & = (a + b) 1 2 7 % 'In104 $ Nonlinear Algebraic Equations Example ( ) = 2 + x − ex, f x n 1 2 3 4 5. . 10. . 15 & c 1.5000 0.7500 1.1250 1.3125 1.2188 .. 1.1455 .. 1.1461 a () −0.9817 0.6330 0.0448 −0.4030 −0.1642 .. 0.0015 .. 1.0 · 10− = 0, b =3 f c f 6 s -x 4 8 % 'In104 Nonlinear Algebraic Equations $ Newton's method 6 f , , , , , , x x s -x s s 1 0 Given a smooth function f (x) and an initial guess x . We want to nd a value h such that f (x + h) ≈ 0. The Taylor series of f around x = x is given by f (x + h) = f (x ) + hf 0 (x ) + O(h ). 0 0 0 0 & 0 0 9 2 % 'In104 Nonlinear Algebraic Equations $ Since we have f (x + h) ≈ f (x ) + hf 0(x ), we can nd h such that f (x + h) ≈ 0 by solving f (x ) + hf 0(x ) = 0, thus h = −f (x )/f 0(x ). We dene f (x ) x =x +h=x − 0 . f (x ) And similarly f (xk ) xk = xk − f 0(x ) , k = 0, 1, . . . k 0 0 0 0 0 0 0 1 0 0 0 0 0 +1 & 10 % 'In104 Nonlinear Algebraic Equations $ Algorithm Given an initial approximation x and a tolerance . 0 k =0 while |f (xk)| > do = xk − ff0 xxkk = +1 xk+1 k k end & ( ) ( ) 11 % 'In104 Nonlinear Algebraic Equations $ Example ( ) = 2 + x − ex, f x x0 = 2, = 10− 8 ( ) 0 2.0 −3.4 1 1.4696 −8.8 · 10− 2 1.2073 −1.4 · 10− 3 1.1488 −5.6 · 10− 4 1.146198 −1.1 · 10− 5 1.146193 −3.9 · 10− k xk f xk 01 01 03 05 11 & 12 % 'In104 Nonlinear Algebraic Equations $ The Secant Method 6f , , , , , , , , , , , , , , , , -x x x x s s 2 () 1 0 f x Given x , dene the linear function f (x ) − f (x ) v (x) = f (x ) − (x − x) 0 , x1 1 1 x1 − x0 Note that and 0 1 ( ) = f (x ) v x0 0 ( ) = f (x ). v x1 & 1 13 % 'In104 Nonlinear Algebraic Equations $ Since v approximates f , we get an approximate solution of f (x∗) = 0, by solving v (x) = 0. The solution of this equation is given by f (x ) x=x − . 1 f (x1)−f (x0 ) x1−x0 1 Hence we dene x =x 1 − 2 or x2 & =x ( ) f x1 f (x1)−f (x0 ) x1−x0 . x ( ) f (xx ) − − f (x ) 1 − f x1 14 1 1 0 0 % 'In104 Nonlinear Algebraic Equations $ Algorithm Given two initial approximations x and x and a tolerance . 0 k 1 =1 while |f (xk)| > do = xk − f (xk ) f k =k+1 xk+1 xk −xk−1 (xk )−f (xk−1 ) end & 15 % 'In104 Nonlinear Algebraic Equations $ Example ( ) = 2 + x − ex, f x x1 = 2, = 10− 8 ( ) 1 2.0 −3.38907 2 1.076750 1.41632 · 10− 3 1.113782 6.79253 · 10− 4 1.147913 −3.69598 · 10− 5 1.146152 8.87754 · 10− 6 1.14619317 1.11843 · 10− 7 1.14619322 −3.39107 · 10− k xk f xk 01 02 03 05 07 12 & 16 % 'In104 Nonlinear Algebraic Equations $ Application I Recall the method of least squares for computing a constant approximation to a set of data (ti, pi), i = 0, 1, . . . , n − 1. Minimize ( )= R α n− X1 i=0 (α − pi) 2 A minimum is computed by dierentiating R, ( )=2 R0 α n− X1 i=0 (α − pi), and solving the equation R0(α) = 0. The solution is given by n− X 1 α= p. (1) 1 & n i=0 17 i % 'In104 Nonlinear Algebraic Equations $ Since (1) is linear, the equation is very easy to solve. But suppose that we want to minimize ( )= R α n− X1 i=0 (α − pi) 4 . Then dierentiation gives ( )=4 R0 α n− X1 i=0 (α − pi) 3 . And we want to solve R0(α) = 0. This is a nonlinear equation. & 18 % 'In104 Nonlinear Algebraic Equations $ Example In this example, we use the Norsk Hydro stock prizes in the time period 1.8.97 to 14.8.97 as data points. We need a termination criteria for the dierent methods. We stop the iterations when |f (xk )| ≤ , and we choose = 5.0 · 10− 8 R(α) 4 6 x 10 5.5 5 4.5 4 3.5 404 & 404.5 405 405.5 406 406.5 Plot of R(α) = 407 Pn−1 19 i=0 407.5 408 408.5 (α − pi)4. 409 % 'In104 Nonlinear Algebraic Equations R’(α) 4 2 $ x 10 1.5 1 0.5 0 −0.5 −1 −1.5 −2 404 404.5 405 405.5 406 406.5 407 407.5 408 408.5 409 P 1 Plot of R0(α) = 4 n− (α − pi)3. We see from the gure that i=0 we have only one solution to the equation R0(α) = 0. & n The Bisection Method c f (c) 1 405.0 2 407.5 3 406.25 4 406.875 5 .. . 406.5625 .. . 20 .. . 406.342516 .. . 36 406.342525 20 −8.88 · 103 7.02 · 103 −5.81 · 102 3 3.28 · 10 3 1.37 · 10 .. . −5.93 · 10−2 .. . −3.188 · 10−9 % 'In104 Nonlinear Algebraic Equations $ The Secant Method k xk 1 415.0267 2 411.1726 3 407.8734 4 406.39415 5 406.340587 6 406.342528 7 406.3425254307 8 406.3425254305 f (xk ) 4 6.46 · 10 2.98 · 104 9.23 · 103 2 3.22 · 10 −1.21 · 101 −2 2.10 · 10 −6 1.37 · 10 −11 1.05 · 10 Newton's Method k & xk 1 413.53 2 408.34 3 406.31 4 406.3424 5 406.34252540304 6 406.34252540305 21 f (xk ) 4 4.89 · 10 1.20 · 104 −2.17 · 102 −2.55 · 10−1 −3.52 · 10−7 −11 1.06 · 10 % 'In104 Nonlinear Algebraic Equations $ P 425 420 415 410 405 400 395 390 0 1 2 3 4 5 6 7 8 9 t P n−1 The constant approximations R(α) = ( α − pi )2 (upper i =0 P 1 line) and R(α) = n− (α − pi)4 (lower line) i=0 & 22 % 'In104 Nonlinear Algebraic Equations $ Example II We now compare minimizing ( )= R1 α n− X1 i=0 (α − pi) 2 , which is a constant approximation using the method of least squares, to minimizing the function ( )= R2 α n− X1 i=0 (α − pi) 4 , for a set of data points (ti, pi). The sum in the rst case involves the square of the dierence between the unknown constant α and the data points pi. In the second case, this power is increased to four. & 23 % 'In104 Nonlinear Algebraic Equations $ In order to minimize R (α), the second method will place α such that the big dierences decrease compared to the rst method. 2 445 440 435 430 P 425 420 415 410 405 400 395 390 0 1 2 3 4 5 6 7 8 9 t P n−1 The constant approximations R2 (α) = (α − pi)4 (upper i=0 Pn−1 line) and R1(α) = i=0 (α − pi)2 (lower line) & 24 % 'In104 Nonlinear Algebraic Equations $ Application II When is a stock prize at its highest? p 425 420 r 415 410 405 400 395 390 0 1 2 3 4 5 6 t∗ 7 8 9 7 8 9 t And when does it turn? p 425 420 415 410 r 405 400 395 390 & 0 1 2 3 t∗ 4 25 5 6 t % 'In104 Nonlinear Algebraic Equations $ We rst compute an approximation by a fourth order polynomial, u(t), to the discrete data points (ti, pi), i = 0, . . . , n − 1: Find α, β, γ, δ and φ such that ( )= R α, β, γ, δ, φ = n− X1 i=0 n− X1 i=0 (u(ti) − pi) 2 (α + βti + γti + δti + φti − pi) 2 3 4 2 , is minimized. After the above approximation u(t) is calculated, we dierentiate it once and solve u0(t∗) = 0. (2) Since (2) is a nonlinear equation, we use the methods discussed in this lesson to solve it. & 26 % 'In104 Nonlinear Algebraic Equations $ When we compute the turning point for the function u, we dierentiate once more. Again, the desired time t∗ is found by solving u00(t∗) = 0, with either the bisection method, the secant method or Newton's method. & 27 %