tsang

Upload: schelanko

Post on 14-Apr-2018

219 views

Category:

Documents


0 download

TRANSCRIPT

  • 7/27/2019 Tsang

    1/5

    Wei Tsang

    Mat 351

    Prof PrestonMay 7, 2006

    Newtons Method for Complex PolynomialMany important questions in mathematics involve the solution of equations or

    systems of equations. Unfortunately, many of them cannot be solved by algebraic

    manipulation. We therefore need to find a way to solve the equations numerically, in thehope of making accuracy as much as possible. There is a simple technique of elementary

    calculus known as Newtons method. Newtons method is a method for iteratively

    approximating the root of a function f(x) =0 using the derivative (under appropriate

    hypothesis). In other words, we want to find a value x* such that f(x*) =0, and weproceed it by giving an educated guess and then refining our guess over and over again.

    If the function is quite simple, we may be able to do some algebra to find a root x

    exactly. For instance, if f(x) is a quadratic polynomial, we can use the quadratic formula.

    If f(x) is a degree 3 or 4 polynomial, there are messier formulas which work as well. Butif f(x) is a higher degree polynomial or an even more complicated function, there is no

    analog to the quadratic formula, i.e. there is no systematic process to algebraicallydetermine the roots exactly. We have to approximate, and Newton's method is just one

    way of doing this.

    We will first start our discussion on Newton's method over the real number line, i.e.

    when x is always real and so is f(x). Later then we will see that the situation becomesmuch more interesting when we enlarge our scope, letting x and f(x) be complex numbers.

    For simplicity, we will use the function f(z) =z3-1 in this case. Of course, we all know

    that z=1 is the only real root, and there are also two other complex rootsz= -1/2 + (3

    1/2/2) i = -0.5 + 0.866 i and z = -1/2 - (3

    1/2/2) i = -0.5 - 0.866 i. (1)

    , which can be calculated by formula to the k-th root of a complex number, and so we do

    not really need Newton's method in this example. However, since many physical scienceequation has no real solution, we introduce complex numbers and are much more

    interested in how a given complex number z converging to the root of the equation, for

    example, f(z)=z3-1 under the sequence given by Newtons method. We will later see that

    even in this simple case we already get pictures which are quite complicated enough than

    on the real number line.

    The Newton's method formula

    The idea behind Newton's method is simple. Suppose that a mapping f: [a, b]R is a

    differentiable mapping and f(x) changes sign, then there is a unique x* in the closedinterval [a, b] such that f(x*)=0. The existence of root x* and the convergence to the root

    are guaranteed by Contraction Mapping Theorem. So if we take a starting "guess" for a

    root locally close to the x* and call it x0, draw the tangent line to f(x) at x0, and figure outwhere this tangent line crosses the x-axis. This is the next guess x1 for a root. If f(x) were

    equal to its tangent line, x1 would actually be a root of f(x), of course. In symbol, by

    linear approximation,

    f(x*)-f(x0 )= f '(x0 ) (x*-x0 )

  • 7/27/2019 Tsang

    2/5

    Then x*= x1 - f(x0 )/ f '(x0 )

    In general, this is not the case. However, the tangent line approximates f(x) for x near

    x0, and so we hope that x1 is not "too far off" from a root of f(x), in particular hopefully

    closer than x0. Repeating the process with x1 instead of x0, we will get the nextapproximation x2, and so on. Thus each xn is determined by f(xn-1) and the slope of the

    tangent line at xn, which is f '(xn-1). The appropriate formula is

    xn = xn-1 - f(xn-1) / f '(xn-1) (2)

    Demonstration of Newtons method:

    Examples for f(x) =x3-1 over the real numbers

    In this case, formula (2) becomes

    xn = xn-1 - (xn-13

    -1) / 3xn-12

    . (3)Let's start with the guess x0=0.5.Iterating formula (3) gives the sequence

    shown at right. We are getting closer and

    closer to the root x=1, so things are

    working well. In fact, you can see fromthe picture that any guess x0>1 will

    quickly converge to the root x=1. In fact,if you randomly pick almost any x0,chances are xn will converge to x=1.

    Notice, however, that if x0 is close to 0,

    then the tangent line there is almost flatand x1 will be blasted way out to theright on the x axis. This is not much of a

    problem, though, since x2 will be pushed

    rapidly back, as the graph shows forx1=1.7 (approx.).

  • 7/27/2019 Tsang

    3/5

    If we choose x0=0, we are in trouble,

    since the tangent is horizontal, f '(x0) =0and formula (*) crashes. Furthermore, if

    we choose a different x0 which

    eventually "lands" on xn=0, we have the

    same problem. The graph at right showsthis in red for x0=-1.434, which crashes

    in 2 steps in this manner. There is an

    infinite sequence of "bad" negativepoints like this. Note that if we choose an

    initial guess very close to such a bad

    point, xn will eventually become close-but-not-equal-to 0, and then xn+1 will get

    blasted way out to the right and from

    there eventually converge to x=1. Thecloser we began to a bad point, the more

    steps it will take to get "reasonablyclose" to x=1. An example of this is

    shown in blue.

    Using complex numbers

    The question may arise what is going to happen if we now allow x (and f(x)) to becomplex numbers rather than just real numbers. So now our domain of a function is on a

    complex plane. Can we get the other complex roots? This question makes perfect sense

    since a polynomial of degree n has n roots, which is the Fundamental Theorem ofAlgebra. It is also possible to define the derivative over complex numbers just as for real

    numbers, so formula (2) makes sense. It turns out that polynomials are all complex-

    differentiable, and the complex-derivative is (surprise!) the same as the real-derivative.So formula (3) still holds, even though the geometry is much more complicated. (In part

    since x is now in some sense 2-dimensional and likewise f(x), so we need to think in 4

    dimensions!)

    We change notation and call the variable z instead of x, and write x and y for its realand imaginary parts, so that z = x + i y. Starting with some (complex) guess z0, we can

    apply the formula and get a sequence zn, which hopefully converges to a (complex) root.

  • 7/27/2019 Tsang

    4/5

    Here is what happens for three values of z0 close to 0.5 + 0.5 i.

    The picture shows that one choice (blue) converges to the real root, but another (green)converges to one of the other complex roots. But if we choose precisely z0 = 0.539 +

    0.417 i (determined with some effort!), then z1= 0 .539 - 0.417 i, and z2=z0! So we getinto an infinite cycle back and forth between these two values and do not converge to any

    root. This is a different sort of "bad point" than the ones we found on the negative real

    axis, when f '(x)=0.

    Incidentally, the grey shading in the graph represents |f(z)|. The dark squares are nearthe roots, when |f(z)| is 0, and the squares are progressively lighter as |f(z)| gets bigger,

    and hence f(z) gets "further from 0". So the darken square is the complex root of the

    equation f(z)=0.Recall that the complex-valued polynomial of degree n has n roots. From the graph

    above, we might question which root is being converged if we guess a z0 somewherebetween two roots. The question leads to an interesting phenomenon called Newton basin,which is fractal.

    Using the same function f(z)=z3- 1to give a fractal, since the polynomial has the three

    toots of unity as its roots. The three roots are 1, -0.5+0.866i, and -0.5-0.866i.

  • 7/27/2019 Tsang

    5/5

    We see the three white small circles are the three roots of the f(z)=0. To answer to whichroot is converge, notice that if we pick z0 in the pink region, the sequence defined by

    Newtons method would converge to 1, likewise, z0 would converge to -0.5+0.866i if it is

    in blue and to -0.5-0.866i if in light blue. (See reference)

    Reference:

    http://aleph0.clarku.edu/~djoyce/newton/examples.html