🔢 mathematics

Projective systems and bounds on the length of codes of non-zero defect

This paper establishes new bounds on the lengths of linear codes with fixed Singleton defect using a projective system framework, unifying existing results, addressing gaps regarding dual code properties, and proposing conjectures on the non-existence of length-maximal codes for dimensions $k \ge 5$ .

Original authors: Tim L. Alderson, Zhipeng Zhang

Published 2026-06-29

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Tim L. Alderson, Zhipeng Zhang

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Imagine you are an architect trying to build the longest possible bridge using a specific set of building blocks. In the world of mathematics, these "bridges" are error-correcting codes—special arrangements of data that allow computers to fix mistakes when information gets scrambled during transmission.

The paper you're asking about is like a blueprint analysis. The authors, Tim Alderson and Zhipeng Zhang, are trying to figure out the absolute maximum length these bridges can be before they become unstable or impossible to build. They are looking at a specific type of bridge that isn't quite "perfect" (mathematically speaking, it has a small "defect"), but is still very strong.

Here is a breakdown of their work using simple analogies:

1. The Building Blocks: Projective Systems

Usually, mathematicians think of these codes as lists of numbers. But these authors decided to look at them as geometric shapes.

The Analogy: Imagine a giant 3D space filled with dots. A code is just a specific collection of these dots.
The Rule: If you draw a flat sheet (a "hyperplane") through this space, it can only cut through a certain number of dots. If it cuts through too many, the code is "broken."
The Goal: They want to pack as many dots as possible into this space without breaking the rules.

2. The "Defect" (The Imperfection)

In the ideal world, there are "perfect" codes (called MDS codes) that are as long as mathematically possible.

The Analogy: Think of a perfect code as a bridge that uses every single available block perfectly.
The Reality: Sometimes, you can't build a perfect bridge. You have to settle for one that is slightly shorter or slightly weaker. This paper focuses on codes that are one or two steps away from perfection. They call this gap the "defect" ( $s$ ).
The Question: If we allow a small defect, how much longer can our code get? Is there a limit?

3. The Main Findings: The "Speed Bumps"

The authors derived several rules (bounds) that act like speed bumps, telling us exactly how long a code can be before it hits a wall.

The "Too Long" Problem: They found that if you try to make a code too long, it stops behaving nicely. Specifically, if a code is long enough, it must be "projective."
- The Analogy: Imagine you are stacking marbles. If you stack too many, you are forced to stop stacking them in a messy pile (where marbles overlap) and start arranging them in a neat, single-layer grid. The math proves that long codes must be neat and non-overlapping.
The "Dual" Relationship: Every code has a "twin" or "shadow" called a dual code. The authors found that if your code is long enough, its twin is also a very specific, strong type of code. It's like saying, "If you build a bridge this long, the shadow it casts on the ground must also be a perfect bridge."
The "Short" Reality for High Dimensions: The paper suggests that for very complex, high-dimensional codes (dimension 5 or higher), you simply cannot build the "longest possible" versions if the number system you are using is large enough.
- The Analogy: It's like trying to build a skyscraper with a specific type of weak brick. You can build a 3-story or 4-story building, but if you try to build a 5th story, the math says it will collapse. The authors suspect that for dimensions 5 and up, the "perfectly long" versions simply don't exist.

4. The "Gap" They Filled

There was a missing piece in previous research. Mathematicians knew that if a code was really long, its twin was strong. But they didn't have a clear rule for when that happened.

The Fix: The authors provided a clear "if-then" rule. They said, "If your code is longer than [Number X], then its twin is guaranteed to be strong." This connects two previously separate ideas.

5. The Big Guess (Conjectures)

Based on their calculations and computer simulations, the authors make a bold guess:

The Guess: For any large enough system, you will never find a "length-maximal" code (the absolute longest possible) if the code is complex enough (dimension 5 or higher).
The Evidence: They checked many examples and found that whenever they tried to build these long, complex codes, they either couldn't build them at all, or they weren't actually the longest possible.

Summary

In short, this paper uses geometry to map out the limits of data storage and transmission. It tells us:

There is a limit: You can't make these codes infinitely long; there are hard mathematical ceilings.
Neatness is required: Long codes must be arranged in very specific, non-overlapping patterns.
High complexity is rare: The "perfectly long" codes likely stop existing once you get to a certain level of complexity (dimension 5).

The authors didn't invent a new type of code for your phone or a new medical scanner; instead, they drew a more accurate map of the mathematical landscape, showing us exactly where the "land" ends and the "ocean" begins for these specific types of data bridges.

Technical Summary: Projective Systems and Bounds on the Length of Codes of Non-Zero Defect

Problem Statement
The paper addresses the fundamental problem in coding theory of determining the maximum length, denoted $n$ , of a linear $[n, k, d]_q$ code with a fixed Singleton defect $s = n - k + 1 - d > 0$ . While the case of zero defect ( $s=0$ , Maximum Distance Separable or MDS codes) is well-studied under the "Main Conjecture on MDS codes," the bounds for codes with non-zero defect (Almost-MDS or AMDS, and generally $A^s$ MDS codes) are less unified. Specifically, the authors investigate the parameters $m_s(k, q)$ (the maximum length of a non-degenerate $A^s$ MDS code) and $m_s^t(k, q)$ (the maximum length where the dual code is additionally $A^t$ MDS). A central gap identified in the literature concerns the conditions under which an $A^s$ MDS code is necessarily dual to an $A^s$ MDS code (i.e., an $N^s$ MDS code), particularly for codes of sufficient length.

Methodology
The authors adopt the framework of projective systems, as advocated by Tsfasman and Vlăduţ, representing linear codes as finite multisets of points in the projective space $PG(k-1, q)$. This geometric perspective allows the translation of coding theoretic parameters into properties of point sets, arcs, and flats.

Geometric Translation: An $[n, k, d]_q$ code is viewed as a multiset $G$ of $n$ points in $PG(k-1, q)$ such that no hyperplane contains more than $n-d$ points. The Singleton defect $s$ corresponds to the maximum number of points in a hyperplane being $k+s-1$ .
Quotient Geometry (Shortening): The paper utilizes a geometric form of code shortening. By taking the quotient of the ambient space $PG(k-1, q)$ by a flat $\Lambda$ , the authors derive bounds on the original code's length based on the properties of the resulting lower-dimensional code. This technique is used to inductively reduce dimension and relate bounds across different $k$ .
Combinatorial Counting: The authors employ counting arguments involving pencils of hyperplanes and flats to derive integrality conditions (via binomial coefficients) that must be satisfied by length-maximal codes.

Key Contributions and Results

Unified Bounds on Code Length:
The paper derives a series of bounds for $m_s(k, q)$ and $m_s^t(k, q)$ that subsume or improve upon existing results (e.g., those by De Boer, Dodunekov, Landjev, and Faldum & Willems).
- General Bound: For $k \ge 3$ , $m_s(k, q) \le (s+1)(q+1) + k - 2$ .
- Refined Bounds: Under specific conditions (e.g., $0 < s < q-2$ and $(s+2, q) \neq (2^e, 2^h)$ ), the bound is tightened to $m_s(k, q) \le (s+1)(q+1) + k - 4$ .
- Dual Code Constraints: The authors establish that if a code has sufficient length ( $n > s(q+1) + k - 1$ ), its dual must be an AMDS code ( $t=1$ ). This addresses a gap in previous literature where sufficient conditions for $N^s$ MDS codes were vacuous.
Projectivity and Structure:
The paper provides sufficient conditions under which long codes must be projective (i.e., no repeated coordinates).
- If $n > s(q+1) + k - 1$ , the code is projective.
- If $n$ exceeds certain thresholds, the code must meet the Griesmer bound.
- It is shown that for $k \ge 3$ , codes of sufficient length are necessarily projective and dual to an AMDS code.
Non-Existence of Length-Maximal Codes for High Dimensions:
The authors investigate the parameter $\kappa(s, q)$ , defined as the maximum dimension $k$ for which a length-maximal $A^s$ MDS code exists.
- They provide explicit values and bounds for $\kappa(s, q)$ for various $s$ and $q$ (summarized in Tables 2 and 3).
- Conjectures: Based on computational evidence and theoretical bounds, the authors propose two conjectures:
  - Weak Conjecture: For $q > 3$ , linear length-maximal codes of dimension $k \ge 5$ do not exist ( $\kappa(s, q) \le 4$ ).
  - Strong Conjecture: For $q > 3$ , $\kappa(s, q)$ is exactly 2, 3, or 4 depending on the parity of $q$ and the divisibility of $s+2$ by $q$ . Specifically, length-maximal codes of dimension 5 or higher do not exist.
Conditions for $N^s$ MDS Codes:
The paper establishes sufficient conditions on $n$ and $k$ under which an $A^s$ MDS code is necessarily $N^s$ MDS (dual is also $A^s$ MDS). Specifically, if $1 < s < q-1$ , $k > (s-1)(q+1)-1$ , and $n > s(q+1) + k - 3$ , then the code is $N^s$ MDS.

Significance and Claims
The authors claim that their geometric approach unifies and extends a range of existing results, providing tighter bounds that often subsume previous literature. The primary significance lies in:

Unification: Demonstrating that the projective system framework effectively handles codes of non-zero defect, similar to how it handles MDS codes.
Structural Insight: Showing that "long" codes of non-zero defect are structurally constrained: they must be projective, meet the Griesmer bound, and have AMDS duals.
Gap Filling: Providing non-vacuous sufficient conditions for a code to be $N^s$ MDS, correcting a limitation in prior works by Tong and Liao & Liao.
Conjectural Limits: Proposing that the landscape of length-maximal codes is finite and small in dimension for $q > 3$ , suggesting that the search for such codes is effectively limited to dimensions $k \le 4$ .

The paper concludes by noting that the only known examples of linear length-maximal codes in dimensions $k > 4$ are the dual of the perfect ternary Golay code ( $[11, 5, 6]_3$ ) and the extended ternary Golay code ( $[12, 6, 6]_3$ ), supporting the conjecture that no others exist for $q > 3$ .