[Paper Review] Linear systems in P^3 with low degrees and low multiplicities

Marcin Dumnicki|ArXiv.org|Oct 12, 2008

Coding theory and cryptography2 references3 citations

TL;DR

This paper proves that linear systems of hypersurfaces in $\mathbb{P}^3$ of degree $d$ with double, triple, and quadruple points in general position are non-special for $14 \leq d \leq 40$, resolving the final open cases in the classification of low-degree linear systems with bounded multiplicities. Using a glueing theorem from prior work and optimized computational algorithms, the author drastically reduces the number of cases requiring matrix rank computation—down to just 22 for $d=40$—and confirms non-specialty via systematic verification over finite fields.

ABSTRACT

We prove that the linear system of hypersurfaces in P^3 of degree d, 14 <= d <= 40, with double, triple and quadruple points in general position are non-special. This solves the cases that have not been completed in a paper by E. Ballico and M.C. Brambilla.

Motivation & Objective

• To complete the classification of non-special linear systems in $\mathbb{P}^3$ with multiplicities at most 4, filling the gap left by prior work.
• To resolve the unresolved cases for degrees $14 \leq d \leq 40$ in the system $\mathcal{L}_3(d; 4^{\times x}, 3^{\times y}, 2^{\times z})$.
• To significantly reduce the computational burden of verifying non-specialty by applying a glueing theorem to merge points of lower multiplicity into higher ones.
• To implement an optimized algorithm that minimizes the number of matrix rank computations required for verification.

Proposed method

• Apply the glueing theorem from [2] to replace combinations of $2^\times 5$ points with a single $4$-point, and $4^{\times a}, 3^{\times b}$ with a $10$-point when $2a + b = 22$.
• Use the virtual dimension invariance under glueing to ensure that non-specialty of the glued system implies non-specialty of the original.
• Implement Algorithm B in FreePascal (or similar systems), iterating over reduced ranges of $q, x, y, z$ with constraints $2x + y \leq 21$ and $z \leq 4$.
• Limit the number of $10$-points considered per degree by choosing optimal values (e.g., 1 for $d=13$–$18$, 5 for $d=19$, etc.) to minimize cases.
• Leverage linear automorphisms to fix four base points at $(1:0:0:0)$, $(0:1:0:0)$, $(0:0:1:0)$, $(0:0:0:1)$, reducing matrix size by up to 880 rows/columns.
• Use the splitting theorem from [2] to reduce verification of large systems to smaller, more manageable subsystems, e.g., splitting $\mathcal{L}_3(40;10^{\times 56},4^{\times 2})$ into $\mathcal{L}_3(39;10^{\times 52},4^{\times 2})$ and $\mathcal{L}_3(40;40,10^{\times 4})$.

Experimental results

Research questions

• RQ1 Are linear systems $\mathcal{L}_3(d; 4^{\times x}, 3^{\times y}, 2^{\times z})$ non-special for all $14 \leq d \leq 40$ and non-negative integers $x, y, z$?
• RQ2 Can the number of required matrix rank computations for verifying non-specialty be drastically reduced using glueing techniques?
• RQ3 Does the glueing of $2^\times 5$ points into a single $4$-point preserve non-specialty when virtual dimension remains unchanged?
• RQ4 Can the splitting theorem be effectively used to decompose large systems into smaller, computationally feasible components?

Key findings

• The linear system $\mathcal{L}_3(d; 4^{\times x}, 3^{\times y}, 2^{\times z})$ is non-special for all $14 \leq d \leq 40$, completing the classification for degrees up to 40 with multiplicities at most 4.
• The number of cases requiring direct matrix rank computation was reduced from 2,294,011 for $d=40$ to just 22 using the glueing method.
• The system $\mathcal{L}_3(3; 2^{\times 5})$ is non-special with virtual dimension $-1$, and $\mathcal{L}_3(9; 4^{\times a}, 3^{\times b})$ is non-special when $2a + b = 22$, both with virtual dimension $-1$.
• The glueing operation $2^{\times 5} \to 4$ and $4^{\times a}, 3^{\times b} \to 10$ preserves virtual dimension and allows reduction of computational complexity.
• For $d \geq 22$, the algorithm efficiently verifies non-specialty by considering only $q$ such that $220q + 20x + 10y + 4z$ is within $N-4$ to $N+20$, where $N = \binom{d+3}{3}$.
• The use of fundamental base points and the splitting theorem further reduces matrix size and computation time, enabling verification on standard hardware.