Newton's method with deflation for isolated singularities of polynomial systems (Q2503291): Difference between revisions

Let \(F({\mathbf x})={\mathbf 0}\) be a polynomial system of \(N\) equations in \(n\) unknowns \({\mathbf x} \in {\mathbb C}^n.\) We are interested in \({\mathbf x}^{\star}\), an isolated solution of \(F({\mathbf x})={\mathbf 0}\). Newton's method generates a sequence of approximations \({\mathbf x}_k\) for \({\mathbf x}^{\star}\). If \({\mathbf x}^{\star}\) is nonsingular, then the sequence converges quadratically (i.e: \(| | {\mathbf x}_k-{\mathbf x}_{k+1}| | =O(| | {\mathbf x}_{k-1}-{\mathbf x}_{k}| | ^2)\)) to \({\mathbf x}^{\star}\). If \({\mathbf x}^{\star}\) is singular, the convergence slows down and gets lost when \({\mathbf x}_k \approx {\mathbf x}^{\star}\). The authors present a modification of Newton's method to restore quadratic convergence of a sequence converging to an isolated singular root without imposing extra requirements of precision on the polynomial system \(F({\mathbf x}).\) The symbolic-numeric method developed produces a new polynomial system which has the original multiple solution as a regular root. Using standard bases, it is shown that the number of deflation stages is bounded by the multiplicity of the isolated root. This modified deflation method works in general, is numerically stable, relatively simple to implement, and a prelimimary implementation performs quite well on a wide class of examples.