Professional Documents
Culture Documents
A. G. Walton
Theorem 1
Let g be continuous on [a, b] and let g(x) [a, b] for all x [a, b]. Then g has a xed point in [a, b]. Suppose further that |g (x)| k < 1 for all x (a, b). Then g has a unique xed point p in [a, b].
M1M2(Comp) algebraic.pdf
A. G. Walton
Proof
(i) Existence If g(a) = a or g(b) = b the existence of a xed point is clear. Suppose not. Since g(x) [a, b] for all x [a, b] then g(a) > a and g(b) < b. Dene h(x) = g(x) x. Then h is continuous on [a, b], h(a) > 0, h(b) < 0. Thus by the intermediate value theorem there exists a p (a, b) such that h(p) = 0. p is therefore a xed point of g. (ii) Uniqueness Suppose we have two xed points p and q in [a, b] with p = q. Then |p q| = |g(p) g(q)| = |p q| |g ( )| by the mean value theorem with (p, q). Since |g ( )| < 1 we have |p q| < |p q| , which is a contradiction. Hence we have uniqueness. To nd the xed point of g(x) we choose an initial approximation p0 and dene a sequence pn by pn = g(pn1 ), n = 1, 2, 3, . . . This procedure is known as xed point or functional iteration. Lets see xed point iteration in action (xedpoint.m).
Theorem 2
Let g be continuous in [a, b] and suppose g(x) [a, b] for all x [a, b]. Suppose further that |g (x)| k < 1 for all x (a, b). Then if p0 is any number in [a, b] the sequence dened by pn = g(pn1 ), converges to the unique xed point p in [a, b]. n1
Proof
Suppose that a pn1 b. Then we have a g(pn1 ) b and hence a pn b. Since a p0 b it follows by induction that all successive iterates pn remain in [a, b]. Now suppose the exact solution is p = , i.e. g() = . Then pn+1 = g() g(pn ) = ( pn )g (cn ), for some cn (, pn ) using the Mean-Value theorem (see theorems.pdf). Since |g (cn )| k it follows that | pn+1 | k | pn | and hence | pn | k n | p0 | . The right hand side tends to zero as n (since k < 1) and so we have pn as n , as required. How do we choose g(x)? For some choices of g the scheme may not converge! One way of choosing g is via Newtons (or Newton-Raphson) method. Newtons method is derived as follows:
Newtons Method
Suppose that the function f is twice continuously differentiable on [a, b]. We wish to nd p such that f (p) = 0. Let x0 be an approximation to p such that f (x0 ) = 0 A Taylor series expansion about p gives (p x0 )2 f ( ), 2 where (p, x0 ). Here we have used the Lagrange form of the remainder for Taylor series (see theorems.pdf). Newtons method arises by assuming that if |p x0 | is small then (p x0 )2 f ( )/2 can be neglected. So we are then left with 0 = f (p) f (x0 ) + (p x0 )f (x0 ). 0 = f (p) = f (x0 ) + (p x0 )f (x0 ) + 2 and |x0 p| is small.
M1M2(Comp) algebraic.pdf f (x0 ) . f (x0 ) Applying this result successively gives the Newton-Raphson method f (xn ) xn+1 = xn . NEWTONS METHOD f (xn ) Note that this is of the form f (x) . xn+1 = g(xn ), with g(x) = x f (x) p x0 Solving this equation for p we have
A. G. Walton
Geometrical interpretation:
At the current value xn nd the tangent to the curve. Extend this until it cuts the x-axis - this is the value xn+1 . Continue procedure.
Disadvantages
May not converge. Evaluation of f (x) may be expensive. Can be slow if f (x) has a multiple root, i.e f (p) = f (p) = 0. If roots are complex numbers then we need a complex initial guess.
Secant method
A method which does not require the evaluation of the derivative f (x) is the secant method. In this we make the approximation f (xn ) f (xn1 ) . f (xn ) = xn xn1 Substituting into Newtons method we have f (xn )(xn xn1 ) . SECANT METHOD xn+1 = xn f (xn ) f (xn1 ) Note that this method needs two initial approximations x0 and x1 . It requires less work than Newton since we do not need to compute f (x).
Geometrical interpretation
Fit a straight line through the last two values (xn , f (xn )), (xn1 , f (xn1 )). Then xn+1 is where this line crosses the x-axis.
M1M2(Comp) algebraic.pdf
A. G. Walton
x y
, F =
f g
(qn+1 qn ) = F (qn ).
Here F/q is a matrix (the Jacobian) consisting of partial derivatives. F = q f /x f /y g/x g/y .
at each step of the iteration. We need initial guesses for x and y to start this off. Lets see how we could program this in Matlab (newton2d.m).
M1M2(Comp) algebraic.pdf
A. G. Walton
Suppose we change the coefcient 210 to 210 107 and leave all other coefcients unchanged. Lets see in Matlab what happens to the roots (see polynomial.m). Matlab shows us that deation is sometimes not an accurate process.