next up previous contents
Next: The Modified Newton Method Up: Nonlinear Equations Previous: Fixed point iterations   Contents

The Newton Method for Systems

The Newton method is useful for solving nonlinear systems of equations of the form
(16.5)

There are independent variables ; in vector notation, these variables are the entries of the vector variable


The system is defined by functions (each is a function of the variables ). In vector notation,


Therefore, the system of equations ([*]) can be written compactly as


We want to find a solution of the system, i.e. a set of values such that all the equations are simultaneously satisfied


Newton's method builds a sequence of points


which converge to the true solution of the problem


In one dimension, the sequence is built using Newton's formula

(16.6)

For -dimensional problems, this formula generalizes to

(16.7)

The place of the derivative is now taken by the Jacobian matrix defined as


Let


The Newton formula ([*]) can be written as

(16.8)

This is a system of linear equations; is a matrix, and and are -dimensional vectors.

One step of the Newton method (say, the step) proceeds as follows:

0
We have available, and want to compute the next iterate, ;
1
Evaluate the vector function (we evaluate each component function individually);
2
Evaluate the derivative matrix (each entry is a function of and need to be evaluated individually);
3
Solve the system ([*]). For this, we need to
3.1
compute the LU decomposition (with pivoting) of the matrix ;
3.2
apply the back-substitution algorithm to the right hand side , to obtain the solution ;
This step cannot be performed if is singular; if this happens we need to terminate the computations.
4
Compute the next iterate as .

Note that one Newton step is very expensive. We have to evaluate functions (the entries of ) and we need to calculate the LU decomposition of each step.


next up previous contents
Next: The Modified Newton Method Up: Nonlinear Equations Previous: Fixed point iterations   Contents
Adrian Sandu 2001-08-26