[Paper Review] Descartes' Rule of Signs for Polynomial Systems supported on Circuits

Frédéric Bihan, Alicia Dickenstein|arXiv (Cornell University)|Jan 21, 2016

Polynomial and algebraic computation9 references3 citations

TL;DR

This paper establishes a multivariate generalization of Descartes' Rule of Signs for polynomial systems supported on circuits—specifically, systems with $ n+2 $ monomials in $ n $ variables. It provides a sharp upper bound on the number of positive real solutions using the sign variation of a coefficient sequence derived from the system’s exponent configuration and coefficients, generalizing the classical univariate rule via Gale duality and oriented matroid theory.

ABSTRACT

We give a multivariate version of Descartes' rule of signs to bound the number of positive real roots of a system of polynomial equations in n variables with n+2 monomials, in terms of the sign variation of a sequence associated both to the exponent vectors and the given coefficients. We show that our bound is sharp and is related to the signature of the circuit.

Motivation & Objective

• To extend Descartes' Rule of Signs to multivariate polynomial systems with $ n+2 $ monomials in $ n $ variables, a setting known as circuits.
• To establish a sharp upper bound on the number of positive real solutions in terms of the sign variation of a coefficient sequence derived from the system’s exponent vectors and coefficients.
• To generalize classical univariate Descartes' rule by incorporating Gale duality and oriented matroid theory to analyze the structure of sparse polynomial systems.
• To prove that the bound is both sharp and congruent modulo 2, mirroring the classical rule’s parity condition.
• To characterize when the number of positive solutions is finite and to provide necessary and sufficient conditions for this finiteness.

Proposed method

• Define the system using a coefficient matrix $ C imes (n imes (n+2)) $ and exponent matrix $ A imes ((n+1) imes (n+2)) $, where $ A $ encodes the monomial exponents.
• Use Gale duality to reduce the multivariate problem to analyzing the number of real roots of a univariate rational function derived from the system.
• Introduce an ordering on the coefficient vectors based on the signs of maximal minors of $ C $, which induces a sequence of coefficients whose sign variations bound the number of positive solutions.
• Apply a generalized Descartes' rule for spaces of analytic functions to bound the number of real roots of the univariate rational function.
• Utilize the concept of a circuit (affine dependence among exponent vectors) to define a canonical affine relation $ oldsymbol{ u} $, and compute the sign variation $ \mathrm{sgnvar}(s_\alpha) $ of the associated coefficient sequence $ s_\alpha $.
• Prove that the number of positive real solutions $ n_{\mathcal{A}}(C) $ is bounded above by $ \mathrm{sgnvar}(s_\alpha) $, and that this bound is sharp under the condition $ \mathbf{0} \in \mathcal{C}^\circ $.

Experimental results

Research questions

• RQ1 Can Descartes' Rule of Signs be generalized to multivariate polynomial systems with $ n+2 $ monomials in $ n $ variables, i.e., systems supported on circuits?
• RQ2 What is the sharp upper bound on the number of positive real solutions of such systems in terms of the coefficients and exponent vectors?
• RQ3 How does the sign variation of a coefficient sequence derived from the system’s structure relate to the number of positive real solutions?
• RQ4 Is the bound congruent modulo 2 with the actual number of positive solutions, as in the univariate case?
• RQ5 Under what conditions is the number of positive real solutions finite, and when can the bound be achieved?

Key findings

• The number of positive real solutions $ n_{\mathcal{A}}(C) $ of a system supported on a circuit is bounded above by the sign variation $ \mathrm{sgnvar}(s_\alpha) $ of a coefficient sequence derived from the system’s exponent configuration and coefficients.
• The bound $ \mathrm{sgnvar}(s_\alpha) $ is sharp: there exist systems where $ n_{\mathcal{A}}(C) = \mathrm{sgnvar}(s_\alpha) $, even achieving the maximum possible value of $ n+1 $ for certain circuits.
• The number of positive real solutions satisfies a congruence modulo 2 with the sign variation, generalizing the classical univariate rule.
• The bound is independent of the choice of representative in the linear equivalence class of the coefficient matrix $ C $, and depends only on the sign pattern of the maximal minors and the circuit structure.
• The condition $ \mathbf{0} \in \mathcal{C}^\circ $, where $ \mathcal{C}^\circ $ is the positive cone generated by the coefficient columns, is necessary and sufficient for the existence of at least one positive solution.
• The bound is not always equal to the normalized volume $ \mathrm{vol}_{\mathbb{Z}}(\mathcal{A}) $, and in some cases (e.g., $ n=4 $, $ \boldsymbol{\nu} = (1,-1,3,-3,1,-1) $), the maximum bound $ n+1 $ is not achievable, showing the bound is not always sharp in terms of volume.