Univariate polynomials with few real zeroes
Benoit Bertrand
Frederic Bihan
Frank Sottile

    This page archives and discusses data from experiments conducted at MSRI in the spring of 2004 investigating the maximum number of real zeroes of a univariate polynomial of the form
(§) f(z) (g(z)) p z k+l   -   h(z),

where p, k, and l are positive integers, and f, g, and h are univariate polynomials of degree k.

    Note that a polynomial of the form (§) has l + 2k + kp complex zeroes. These data suggest that the maximum number of real roots for a univariate polynomial of the form (§) is bounded by 3k+c, where c is some constant. In fact, we have a construction of such a polynomial with 4k+2 (or 4k+1 depending on the parity of l) real zeroes.
    This experimentation is part of a research project we engaged in at the MSRI as part of the program on Topological Aspects of Real Algebraic Geometry. It will result in a publication: "Polynomial systems with few real zeroes".


    Such a polynomial is the univariate eliminant of a polynomial system whose support is the near circuit which consists of the four integer points (0,0,0), (1,0,0), (0,1,0), and (1,p,q) in R3, and the segment (0,0,0), (0,0,-1), ..., (0,0,-k). When p and q are relatively prime, the first four vectors are the vertices of an empty simplex. The first four points and (0,0,-k) are in convex position if q-pk-k>0.

    We apply the affine transformation z |---> z + k(1-x-y), and then set l:=q-pk-k. When l>0, this point configuration lies in the positive octant. It is the column vectors of the matrix

000...0 101
000...0 01p
012...k 00l
and a polynomial system with this support is equivalent to the following system
x=f(z)
y=g(z)
xy pz lz k =h(z)
Substituting  f(z) for x and g(z) for y gives the univariate eliminant   f(z)(g(z)) p z k+l - h(z). This polynomial makes sense if k+l >=0. If k+l<0, then we multiply this Laurent polynomial by z-k-l to turn it into a polynomial.

The polytope when (kpl)  =  (3,2,5).

    The tables below summarize the results of our experiments, which involved over 40 computers at MSRI, and a few at UMASS, and some at Texas A&M University in the spring, summer, and autumn of 2004. In all, we generated over 620 million different polynomials and used over 206 million seconds (6.55 years) of computer time. For every value of k, p, and l below, we generated at least a million polynomials of the form (§) and determined their numbers of real solutions. These data were tabulated and the maximum number of real solutions observed is recorded in the table below. Data files recording the frequencies of the different numbers of real solutions are linked to the corresponding entries in the table. A summary of these data and the description of the polynomials is found here. When p=1, we have proven that the maximum number of real solutions is 3k+2, (or 3k+1, if l-k is odd). Red entries are those where the observed maximum number of observed real solutions is known to be the actual maximum.
Observed maximum number of real solutions to a polynomial of the form (§)
Values for l
 k   p  -3-2 -10 1 23 45 67 89 1011  p   k 
2 1 ---- ---- 7 8 7 8 7 8 7 8 7 8 7 1 2
2 -------- 9 8 9 8 9 10 9 10 9 10 9 2
3 7 6 7 8 7 8 7 8 7 8 9 8 9 8 9 3
4 7 6 7 8 9 8 9 8 9 8 9 8 9 8 9 4
5 7 6 7 8 7 8 7 8 7 8 7 8 7 8 7 5
6 9 6 7 8 9 8 9 8 9 8 9 8 9 8 9 6
3 1 ---- ---- 10 11 10 11 10 11 10 11 10 11 10 1 3
2 ---- 11 12 11 12 11 12 11 12 11 12 11 12 11 2
3 8 9 10 9 10 9 10 9 10 9 10 9 8 9 8 3
4 9 10 11 12 11 12 11 12 11 12 11 12 4
5 8 11 10 11 10 11 10 11 10 11 10 5
4 1 ---- ---- 13 14 13 14 13 14 13 14 13 14 13 1 4
2 13 14 13 14 13 14 15 14 15 14 15 14 13 14 13 2
3 13 14 11 10 11 10 11 10 11 10 11 10 11 3
4 13 14 15 14 13 14 13 14 13 14 15 4
5 1 ---- ---- 16 17 16 17 16 17 16 17 16 17 16 1 5
2 17 16 15 14 15 14 15 14 15 14 15 14 2
3 14 13 12 13 12 13 12 13 12 11 12 3
6 1 ---- ---- 19 20 19 20 19 20 19 20 19 20 19 1 6
2 17 16 17 16 17 16 17 16 17 16 19 18 2
3 15 14 13 12 13 12 13 12 13 12 13 3
 k   p  -3-2 -10 1 23 45 67 89 1011  p   k 
Values for l

Based upon work supported by the National Science Foundation grants CAREER DMS-0134860 and DMS-9810361 (funding the MSRI), and the Clay Mathematical Foundation.
Modified since 15 December 2004 by Frank Sottile