I7 Logo
Chair for Foundations of Software Reliability and Theoretical Computer Science
Informatik Logo TUM Logo
Publications - Convergence Thresholds of Newton's Method for Monotone Polynomial Equations


Javier Esparza, Stefan Kiefer, and Michael Luttenberger. Convergence thresholds of Newton's method for monotone polynomial equations. Technical report, Technische Universität München, Institut für Informatik, October 2007.


Monotone systems of polynomial equations (MSPEs) are systems of fixed-point equations X_1 = f_1(X_1, ..., X_n), ..., X_n = f_n(X_1, ..., X_n) where each f_i is a polynomial with positive real coefficients. The question of computing the least non-negative solution of a given MSPE X = f(X) arises naturally in the analysis of stochastic models like stochastic context-free grammars, probabilistic pushdown automata, and back-button processes. Etessami and Yannakakis have recently adapted Newton's iterative method to MSPEs. In a previous paper we have proved the existence for strongly connected MSPEs (scMSPEs) of a threshold k_f such that after k_f iterations of Newton's method each new iteration computes at least 1 new bit of the solution. However, the proof was purely existential. In this paper we give an upper bound for k_f as a function of the maximal and minimal components of f. Using this result we show that k_f is at most single exponential resp. linear for scMSPEs derived from probabilistic pushdown automata resp. from back-button processes. Further, we prove the existence of a threshold for arbitrary MSPEs after which each new iteration computes at least 1/(w 2^h) new bits of the solution, where w, h are the width and height of the DAG of strongly connected components.

Suggested BibTeX entry:

    author = {Javier Esparza and Stefan Kiefer and Michael Luttenberger},
    institution = {Technische Universit\"{a}t M\"{u}nchen, Institut f\"{u}r Informatik},
    month = {October},
    title = {Convergence Thresholds of {N}ewton's Method for Monotone Polynomial Equations},
    year = {2007}

GZipped PostScript (206 kB)
PDF (287 kB)
Conference version