2
$\begingroup$

Macaulay Resultants help identify common root in $\mathbb P^{n-1}(\mathbb K)$ of $n$ homogeneous polynomials in $n$ variables when $\mathbb K$ is algebraically closed. Is it possible that some type of multivariate resultants give all solutions $(x_0,x_1,\dots,x_{n-1},x_n)\not=(0,0,\dots,0,0)$ with $x_0=1$ and $\prod_{i=0}^nx_i\neq0$ of $r\geq n+1$ degree $d$ homogeneous polynomials in polynomial ring $\mathbb Z[x_0,x_1,\dots,x_{n-1},x_n]$ $$f_1(x_0,x_1,\dots,x_n)=0$$ $$f_2(x_0,x_1,\dots,x_n)=0$$ $$\vdots$$ $$f_r(x_0,x_1,\dots,x_n)=0$$ or is there different technique that applies (we assume coefficient matrix is rank $r$)?

The method could be exponential in $n,d$ (note Macaulay Resultants are exponential in $n,d$ and I only seek a method that is compatible in complexity with algebraically closed case). It only needs to depend polylogarithmically in $q=\binom{n+d}d\max_j\|f_j\|_\infty$ (it is trivial to find a method with $O(\ell^{nd})$ complexity where $\ell=\|(x_0,x_1,\dots,x_n)\|_\infty$ is infinity norm of solutions sought and it is assumed $\ell<q$ holds but I seek something $O((\log q)^{nd})$ which is a much relaxed framework). If convenient we may think of working in $\mathbb Z_m[x_0,x_1,\dots,x_{n-1},x_n]$ with $m\gg\binom{n+d}d\max_j\|f_j\|_\infty$ a prime.

Undecidability is not an issue since roots are bounded in $\infty$ norm and I only seek a slightly cleverer algorithm to trivial one. It also does not seem amenable to simple reduction from factoring such as $x_1x_2-Nx_0^2=0$ as it does not look like amenable to producing two additional homogeneous equations so that $x_0=1$ and product of coordinates of roots $\neq0$ holds.

$\endgroup$
8
  • 1
    $\begingroup$ If you have one homogeneous polynomial in two variables, say $f(x,y)=xy$, then findind the non-trivial roots mod $m$ amounts to factor the integer $m$. Maybe you could clarify your setting and what you wish to know. $\endgroup$ Commented Oct 3, 2018 at 13:17
  • $\begingroup$ It is reasonable to assume $m=p$ is prime to start with. In this case classical elimination theory (with resultants) will give you the common roots in $\overline{\mathbb{F}_p}$ (I'm not sure the most efficient method however). Regarding the equation $xy=0$, you can always artificially add variables e.g. consider $xy=xz=yz=0$, but I assume that is not what you intend. $\endgroup$ Commented Oct 3, 2018 at 13:42
  • $\begingroup$ @FrançoisBrunault No definitely not my intent. Being a computer scientist I like to look at finite things and not infinite objects like $\overline{\mathbb F_p}$ and I think $\mathbb Z_m$ is right non-trivial object. $\endgroup$
    – Turbo
    Commented Oct 3, 2018 at 13:44
  • $\begingroup$ $(x,y,z)=(p,q,0) $ is a common root. Regarding elimination theory, if one is careful enough it does provide the solutions in finite extensions e.g. finite fields $\mathbb{F}_p[x]/(T)$ where T is an irreducible polynomial. I now understand your concern is also about efficiency. It makes sense in Z/mZ since this is a finite object. $\endgroup$ Commented Oct 3, 2018 at 14:43
  • $\begingroup$ @FrançoisBrunault Oh I see then. Then I guess $O((\log m)^n)$ is unlikely if $m$ is non-prime. How does getting solutions in finite extensions help here? Is there illustrative example? $\endgroup$
    – Turbo
    Commented Oct 3, 2018 at 15:01

0

You must log in to answer this question.