Tue Sep 27 20:15:45 CEST 2011
The data files are now put in a zip and a
tar.gz file.
Fri 28 Feb 21:37:46 CET 2014
Site moved from own hosting to github
pages.
We consider non-zero polynomials [;F(X_1, \ldots , X_m) \in \mathbb{F}[X_1, \ldots , X_m]\setminus \{0\};] where [;\mathbb{F};] is any field.
Consider a point [;\vec{a}=(a_1, \ldots , a_m)\in \mathbb{F}^m;]. Let [;J_t \subseteq \mathbb{F}[X_1, \ldots , X_m];] be the ideal generated by
We say that [;\vec{a};] is a zero of [;F;] of multiplicity [;r;] if [;\mathbb{F}(X_1, \ldots , X_m) \in J_r \backslash J_{r+1};]. Equivalently, multiplicity can be defined using the Hasse derivative.
Let [;\prec;] be the pure lexicographic ordering with [;X_m \prec \cdots \prec X_1;]. Assume the only thing we know about [;F(X_1, \ldots , X_m);] is its leading monomial [;X_1^{i_1}\cdots X_m^{i_m};] (taken of course with respect to [;\prec;]). Let [;S_1, \ldots , S_m \subseteq \mathbb{F};] be finite sets, say [;s_1= |S_1|, \ldots , s_m=|S_m|;]. We would like to know what is the maximal attainable number of zeros in [;S_1 \times \cdots \times S_m;] of multiplicity at least [;r;] for such a polynomial. The function [;D(i_1, \ldots ,i_m,r,s_1, \ldots ,s_m);] from [3] upper bounds this number. Similarly, the function [;H(i_1, \ldots , i_m,r,s_1, \ldots , s_m);] from [4] serve as a lower bound. Below we list values of [;D(i_1, \ldots , i_m,r,q, \ldots ,q);] and [;H(i_1, \ldots , i_m,r,q, \ldots ,q);] for a number of cases. The upper bound [;D(i_1,\ldots,i_m,r,q,\ldots,q);] is an improvement to the upper bound
Here, the first expression comes from a corollary to the generalized Schwartz-Zippel bound [1] [2] and [;q^m;] is the number of points in the ensemble under consideration.
When [;m=2;], we can put the lower and upper bounds in a table. When [;m;] is larger, we consider tables of tables as we will see later.
For instance for [;m=2;], [;r=2;] and [;s_1=s_2=q=3;], the table looks a follows.
[;i_1=0;] | [;i_1=1;] | [;i_1=2;] | [;i_1=3;] | [;i_1=4;] | [;i_1=5;] | |
[;i_2=0;] | 0-0 | 0-0 | 3-3 | 3-3 | 6-6 | 6-6 |
[;i_2=1;] | 0-0 | 1-1 | 3-4 | 4-5 | 6-7 | 7-7 |
[;i_2=2;] | 3-3 | 3-3 | 5-5 | 6-7 | 7-8 | 8-8 |
[;i_2=3;] | 3-3 | 4-4 | 6-6 | |||
[;i_2=4;] | 6-6 | 6-6 | 7-7 | |||
[;i_2=5;] | 6-6 | 7-7 | 8-8 |
Each entry [;(i_1,i_2);] contains "[;H(i_1,i_2,3,3,3);] - [;D(i_1,i_2,2,3,3);]".
In addition to this table, we also find a table which compares the upper bound calculated by [;D;] and the upper bound from (2.1). Here, each cell contains
[;i_1=0;] | [;i_1=1;] | [;i_1=2;] | [;i_1=3;] | [;i_1=4;] | [;i_1=5;] | |
[;i_2=0;] | 0/0 | 0/1 | 3/0 | 3/1 | 6/0 | 6/1 |
[;i_2=1;] | 0/1 | 1/2 | 4/0 | 5/1 | 7/0 | 7/2 |
[;i_2=2;] | 3/0 | 3/1 | 5/1 | 7/0 | 8/1 | 8/1 |
[;i_2=3;] | 3/1 | 4/2 | 6/1 | |||
[;i_2=4;] | 6/0 | 6/1 | 7/2 | |||
[;i_2=5;] | 6/1 | 7/2 | 8/1 |
As an example where [;m>2;], let [;m=4;], [;r=2;] and [;s_1=s_2=s_3=s_4=q=2;]. The table consists of 16 matrices of size [;4 \times 4;]. We here only list the upper left part.
[;i_3=0;] | [;i_3=1;] | |||||||||||
[;i_1=0;] | [;i_1=1;] | [;i_1=2;] | [;i_1=3;] | [;i_1=0;] | [;i_1=1;] | [;i_1=2;] | [;i_1=3;] | |||||
[;i_4=0;] | [;i_2=0;] | 0-0 | 0-0 | 8-8 | 8-8 | [;i_4=0;] | [;i_2=0;] | 0-0 | 4-4 | 8-12 | 12-12 | |
[;i_2=1;] | 0-0 | 4-4 | 8-12 | 12-12 | [;i_2=1;] | 4-4 | 8-8 | 12-14 | 14-14 | |||
[;i_2=2;] | 8-8 | 8-8 | [;i_2=2;] | 8-12 | 12-12 | |||||||
[;i_2=3;] | 8-8 | 12-12 | [;i_2=3;] | 12-12 | 14-14 | |||||||
[;i_3=0;] | [;i_3=1;] | |||||||||||
[;i_1=0;] | [;i_1=1;] | [;i_1=2;] | [;i_1=3;] | [;i_1=0;] | [;i_1=1;] | [;i_1=2;] | [;i_1=3;] | |||||
[;i_4=1;] | [;i_2=0;] | 0-0 | 4-4 | 8-12 | 12-12 | [;i_4=1;] | [;i_2=0;] | 4-4 | 8-8 | 12-14 | 14-14 | |
[;i_2=1;] | 4-4 | 8-8 | 12-14 | 14-14 | [;i_2=1;] | 8-8 | 11-11 | 14-15 | 15-15 | |||
[;i_2=2;] | 8-12 | 12-12 | [;i_2=2;] | 12-14 | 14-14 | |||||||
[;i_2=3;] | 12-12 | 14-14 | [;i_2=3;] | 14-14 | 15-15 |
Proposition 16 in [4] upper bounds [;D(i_1,i_2,r,s_1,s_2);] by closed formulas.
We provide tables comparing the true value of [;D;] and the closed formulas. For instance for [;m=2;], [;r=2;] and [;s_1=s_2=q=3;], the table looks a follows. Each entry [;i_1,i_2);] contains
where [;CF;] is the closed formula from Proposition 19.
[;i_1=0;] | [;i_1=1;] | [;i_1=2;] | [;i_1=3;] | [;i_1=4;] | [;i_1=5;] | |
[;i_2=0;] | 0/0 | 0/1 | 3/0 | 3/0 | 6/0 | 6/0 |
[;i_2=1;] | 0/1 | 1/1 | 4/0 | 5/0 | 7/0 | 7/0 |
[;i_2=2;] | 3/0 | 3/1 | 5/0 | 7/0 | 8/0 | 8/0 |
[;i_2=3;] | 3/1 | 4/1 | 6/0 | |||
[;i_2=4;] | 6/0 | 6/0 | 7/0 | |||
[;i_2=5;] | 6/1 | 7/0 | 8/0 |
A program written in Haskell was used to
generate the numbers. The source code is available as a
plain .hs file and as a
pretty printed file.
Further, you can find the file at
github. For a commented
version, please refer to the thesis of C. Thomsen.
The Haskell functions for [;D;], [;H;],
(2.1) and the closed formula are called
d
, h
, sz'
, (sz
) and
cf
, respectively, and are programmed such that
In this section you find tables containing the bounds as explained in Section 3.
Download files as zip or tar.gz.
For each combination of [;m;], [;r;] and [;q;], there are two or three files:
Thanks to Dr Douglas R. Woodall for creating the original version of LaTeXMathML.js and thanks to Jeff Knisley for updating LaTeXMathML.js. Thanks to Avital Oliver and helpers for creating TeXtheWorld and thanks to CodeCogs for rendering math symbols nicely.