    Next: Implementation Up: Rouche theorem Previous: Rouche theorem   Contents

Mathematical background

Let a system of equations in unknowns: We will denote by the matrix of the derivatives of order of with respect to the variable. Let us consider a point and define as  and . If is strictly lower than , then has a single root in a ball centered at with radius and the Newton scheme with initial guess will converge to the solution.

The most difficult part for using this theorem is to determine . For algebraic equations it is easy to determine a value , that we will call the order of Rouche theorem, such that and consequently may be obtained by computing for all in and taking as the Sup of all .

For non algebraic finding requires an analysis of the system.

Rouche theorem may be more efficient than Moore or Kantorovitvh theorems. For example when combined with a polynomial deflation (see section 5.9.6) it allows one to solve Wilkinson polynomial of order up to 18 with the C++ arithmetic on a PC, while stand solving procedure fails for order 13.    Next: Implementation Up: Rouche theorem Previous: Rouche theorem   Contents
Jean-Pierre Merlet 2012-12-20