[Paper Review] On Lazard's Valuation and CAD Construction

Scott McCallum, Hoon Hong|arXiv (Cornell University)|Jan 26, 2015

Polynomial and algebraic computation11 references3 citations

TL;DR

This paper rigorously investigates Lazard's valuation of multivariate polynomials at a point, establishing foundational properties and linking valuation-invariance to order-invariance. It proves Lazard's main CAD construction claim for n=3 under a stronger hypothesis, offering a corrected, limited framework for his 1990 improved projection method in cylindrical algebraic decomposition.

ABSTRACT

In 1990 Lazard proposed an improved projection operation for cylindrical algebraic decomposition (CAD). For the proof he introduced a certain notion of valuation of a multivariate Puiseux series at a point. However a gap in one of the key supporting results for the improved projection was subsequently noticed. In this report we study a more limited but rigorous concept of Lazard's valuation: namely, we study Lazard's valuation of a multivariate polynomial at a point. We prove some basic properties of the limited Lazard valuation and identify some relationships between valuation-invariance and order-invariance.

Motivation & Objective

• To address a gap in Lazard's 1990 claim regarding improved projection for cylindrical algebraic decomposition (CAD) by focusing on a more rigorous, limited concept of valuation.
• To establish foundational properties of Lazard's valuation for multivariate polynomials over a field, particularly at a point in real or complex space.
• To clarify the relationship between valuation-invariance and order-invariance in the context of polynomial behavior near a point.
• To provide a stepping stone toward validating Lazard's broader claim for CAD construction, especially for non-well-oriented polynomial sets.
• To lay groundwork for extending the theory to Puiseux series and rational functions, despite convergence challenges.

Proposed method

• Defining Lazard's valuation of a multivariate polynomial f at a point a as the lexicographically least multi-index (v₁,…,vₙ) such that the corresponding partial derivative of f does not vanish at a.
• Using lexicographic order on ℕⁿ to ensure uniqueness and minimality in the valuation definition.
• Proving basic properties such as vₐ(f) = (0,…,0) iff f(a) ≠ 0, and extending the valuation to rational functions via vₐ(f/g) = vₐ(f) − vₐ(g).
• Demonstrating that valuation-invariance implies order-invariance under certain conditions, and showing that order-invariance is locally bounded.
• Establishing that for n=3, Lazard’s main claim about valuation-invariance and CAD construction holds under a slightly stronger hypothesis.
• Analyzing the challenges in extending the valuation to multivariate Puiseux series due to restricted convergence regions.

Experimental results

Research questions

• RQ1 Can Lazard’s valuation for multivariate polynomials be rigorously defined and its basic properties proven?
• RQ2 What is the relationship between valuation-invariance and order-invariance for polynomials at a point?
• RQ3 Does Lazard’s main claim about CAD construction via valuation-invariance hold for n=3 under a stronger hypothesis?
• RQ4 What are the fundamental obstacles to extending Lazard’s valuation to multivariate Puiseux series?
• RQ5 Can the theory of Lazard’s valuation be extended meaningfully to rational functions and power series rings?

Key findings

• Lazard’s valuation of a multivariate polynomial at a point is well-defined as the lexicographically least multi-index corresponding to a non-vanishing partial derivative.
• The valuation vₐ(f) = (0,…,0) if and only if f(a) ≠ 0, providing a direct link to non-vanishing at the point.
• Valuation-invariance implies order-invariance in a neighborhood of the point, and order-invariance is locally bounded, supporting CAD construction.
• Lazard’s main claim regarding valuation-invariance and CAD construction is proven for the case n=3 under a slightly stronger hypothesis than originally stated.
• Extending the valuation to multivariate Puiseux series is hindered by convergence issues, as such series may converge only in slender regions around the point.
• The theory of Lazard’s valuation for polynomials provides a solid foundation for further exploration, though extending it to more general series remains challenging.