next up
Next: 2.iii.b. Fewnomial Upper Bounds
Up: 2.iii. Real Solutions to Sparse Polynomial Systems

2.iii.a. Constructive Lower Bounds

By the example in Section 2.ii.a, if P is the convex hull of v0, v1, ..., vn, a lattice simplex with no interior lattice points, then a general system (2.1) is equivalent to the system of binomials

y1d1 = a1,   y2d2 = a2,   ...,   yndn = an  . (2.5)

with each ai non zero. Here, d1, d3, ..., dn are the invariant factors of the matrix whose ith column is vi - v0 and d1d3...dn is the normalized volume of P. Following Sturmfels [Stu1], let e(P) be the number of these invariant factors which are even. If e(P)=0, so that d is odd, then P is an odd cell.

Proposition 2.5   The polynomial system (2.5) has 2e(P) real solutions if ai > 0 whenever di is even, and no real solutions otherwise. In particular, if P is an odd cell, then there is at least one real solution.

Theorem 2.6 (Sturmfels [Stu1, Corollary 2.3]) The maximum number r(P) of real solutions to a sparse polynomial system (2.1) with common Newton polytope P is at least the the number of odd cells in any regular triangulation Pw of P.

Proof. In the limit as t approaches zero, the lifted system (2.4) given by a regular triangulation Pw of P becomes the disjunction of facial subsystems of (2.1), one for each facet F of Pw. Thus the number of real solutions in the limit is a constructive lower bound for r(P). By Proposition 2.5, the number of odd cells in Pw is a lower bound on the number of real solutions to the facial subsystem given by the facets of Pw. Q.E.D.

Proposition 2.5 also gives an upper bound on the limiting number of real solutions r of the lifted system (2.4) as t approaches zero (this is due to Sturmfels [Stu1, Theorem 2.2].)

r is at most 2e(F)   ,
(2.6)

the sum over all facets F of the regular triangulation Pw of P. More sophisticated accounting of the possible signs of the coefficients of facial subsystems improves this bound. This accounting is accomplished in [PS] and [IR], leading to a combinatorial upper bound for such limiting systems. Itenberg and Roy [IR] show there is a system (2.1) for which this upper bound is attained, and thus this combinatorial upper bound for limiting systems is a lower bound for r(P), the maximum number of real solutions to a system with common Newton polytope P.

Itenberg and Roy then conjectured that this combinatorial bound was in fact the global bound, that is, they conjectured that the maximal number of real solutions occurs in limiting systems. They also gave a similar bound for r+(P), the number of solutions with positive coordinates. This was too optimistic, for Li and Wang [LW] found a remarkably simple counterexample to this conjecture of Itenberg and Roy

y - x - 1   =   200 - 100y3 + 900x3 - x3y3   =   0 .

This system has 3 positive solutions

(0.317659, 1.317659),   (.659995, 1.659995),   and   (8.12058, 9.12058) ,

but the combinatorial upper bound is 2. Thus systems with the maximal number of real solutions cannot in general be constructed with these limiting techniques.

We still have more questions than answers. Among the questions are:

  1. Improve this lower bound for r(P) (or for r+(P) of Itenberg and Roy.
  2. Find new general methods to construct systems with many real solutions.
  3. Determine which polytopes P have the property that r(P) < n!VolP, that is, not all the solutions can be real.


next up
Next: 2.iii.b. Fewnomial Upper Bounds
Up: 2.iii. Real Solutions to Sparse Polynomial Systems