Nonlinear equations:
Newton’s method
MATH2070: Numerical Methods in Scientific Computing I
Location: http://people.sc.fsu.edu/∼jburkardt/classes/math2070 2019/nonlinear newton/nonlinear newton.pdf
Even in zero visibility, a skier knows which way is downhill!
Rootfinding (Newton version)
Given f(x), with derivative f
0
(x), and a starting point x, estimate a root x
∗
for which f(x
∗
) = 0?
1 Newton’s linear model for f(x)
Suppose that we are looking for a root x
∗
of f(x) = 0, and that f (x) is at least twice continuously differen-
tiable. Then, at a point x sufficiently close to x
∗
, we can write:
f(x∗) = f(x) + f
0
(x) ∗ (x
∗
− x) + O(x
∗
− x)
2
Noting that f(x
∗
) = 0, and dropping the O(x
∗
− x)
2
terms, we arrive at Newton’s estimate for the location
of the root:
x∗ ≈ x −
f(x)
f
0
(x)
Thus, to estimate the root of a function, we try to find a starting point that we hope is already close, and
we must be prepared to evaluate the derivative of the function as part of the iteration.
We hope that, once the iteration is close to the root, the norm of the function will decrease on every step,
and the size of the step |x
i
− x
i−1
| will decrease quadratically, as it becomes similar to the size of the actual
(but unknowable) error |x
∗
− x
i−1
|.
2 Newton’s method pseudocode
The pseudocode for Newton’s method is very similar to that for previous algorithms. However, we only pass
in a single starting value x, and along with the function f(x) we now must include the derivative function
fp(x).
1