🔢 mathematics

Cactus barriers

This paper utilizes cactus varieties and scheme-theoretic methods in algebraic geometry to explain the precise geometric reasons behind the inherent limitations of determinantal methods in proving lower bounds for the border rank of tensors.

Original authors: Jarosław Buczyński

Published 2026-02-13

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Jarosław Buczyński

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

The Big Picture: Trying to Build a House with Bricks

Imagine you are an architect trying to build a complex, beautiful house (a tensor or a complicated mathematical object). You have a pile of simple, standard bricks (simple states or vectors). Your goal is to figure out the minimum number of bricks needed to build that specific house.

In mathematics, this minimum number is called the Rank. If you can build the house with 10 bricks, its rank is 10. If you need 1,000, it's very complex.

For decades, mathematicians have been trying to prove that certain houses are extremely complex (requiring millions of bricks). To do this, they use a tool called a Linear Rank Method. Think of this method as a "magic ruler" or a "detector."

The Magic Ruler (Linear Rank Methods)

The "magic ruler" works like this:

You look at your pile of simple bricks.
You check how the ruler reacts to them. Let's say the ruler always shows a "reading" of 5 when pointed at a single brick.
You then point the ruler at your complex house.
If the ruler shows a reading of 50, you can logically conclude: "Since one brick gives a 5, and the house gives a 50, the house must be made of at least 10 bricks."

This is a powerful way to prove a house is complex without actually taking it apart.

The Problem: The "Cactus" Wall

The paper argues that this "magic ruler" has a hard limit. No matter how clever you make the ruler, it will never be able to prove that a house is more complex than a certain point.

The author calls this limit the "Cactus Barrier."

Why "Cactus"? In math, there is a concept called a Cactus Variety. Imagine a cactus. It has a main stem, but it also has little spiky offshoots.

Standard Bricks: These are like smooth, perfect spheres.
Cactus Bricks: These are like the cactus. They are "fuzzy" or "thick" at the edges. They represent not just a single point, but a tiny, squishy blob of points stuck together.

The paper proves that the "magic ruler" is actually measuring the complexity of these Cactus Bricks, not the standard ones.

The Analogy: The "Fuzzy" Limit

Here is the core problem explained simply:

The Goal: You want to prove a house needs 1,000 standard bricks.
The Ruler's Limit: The ruler can only detect "fuzzy" complexity.
The Cactus Trick: It turns out you can build that same house using only 100 "Cactus Bricks" (those fuzzy, thick blobs).
The Result: When you use the ruler, it sees the "fuzziness" of the 100 Cactus Bricks. It says, "Okay, this house is complex, but it only looks like it needs 100 units of effort."

Even if the house actually needs 1,000 standard bricks, the ruler gets stuck at 100. It cannot see past the "fuzziness" of the Cactus Bricks.

Why This Matters

For a long time, mathematicians hoped to use these rulers to prove that certain problems (like multiplying huge matrices or simulating quantum computers) are incredibly hard. They wanted to prove, "You need a billion steps to do this!"

However, this paper says: "Stop. You are hitting a wall."

The paper explains why the wall exists. It's not because the math is broken; it's because the tools we are using (the linear rank methods) are fundamentally blind to the difference between "standard bricks" and "cactus bricks."

The Barrier: The paper calculates exactly where this wall is. For example, in a specific 3D space, the wall is at 6m - 4. No matter how you tweak your ruler, you can never prove a complexity higher than that number using this method.
The Gap: For large, complex problems, the difference between what the ruler can prove (e.g., 100) and the actual complexity (e.g., 1,000,000) is massive.

The "Smoothness" Secret

The paper also explains a subtle geometric trick.

If you try to build a house with "perfectly smooth" bricks, the ruler works fine.
But if you allow the bricks to be slightly "cracked" or "fuzzy" (mathematically called non-smooth schemes), the ruler gets confused. It sees the cracks as extra complexity, but it can't tell if those cracks are real or just an illusion.

The author shows that the "Cactus Barrier" is the point where the "fuzziness" takes over, and the ruler stops being useful for proving high complexity.

The Takeaway

We hit a wall: Current methods for proving how hard a problem is have a hard ceiling.
The reason: These methods are actually measuring "fuzzy" versions of the problem (Cactus varieties) rather than the real, sharp problem.
What's next? To break this barrier, mathematicians need to invent entirely new tools that can distinguish between "fuzzy" complexity and "real" complexity. They need to find a way to "smooth out" the cactus and see the real bricks underneath.

In short: We can't use the old ruler to measure the new height. We need a new tool.

1. Problem Statement

The paper addresses a fundamental limitation in algebraic complexity theory: the inability of linear rank methods to prove strong lower bounds for the border rank of tensors (or polynomials).

Context: In tensor decomposition, the goal is to express a tensor $F$ as a sum of simple tensors (rank) or as a limit of such sums (border rank). Linear rank methods involve constructing a matrix $M$ of linear forms such that the rank of $M(F)$ provides a lower bound on the rank of $F$ .
The Barrier: It is known that for tensors in $\mathbb{C}^m \otimes \mathbb{C}^m \otimes \mathbb{C}^m$ , linear rank methods cannot prove a border rank lower bound exceeding $6m - 4$ . This is significantly lower than the expected generic rank (which scales with $m^2$ ).
The Question: What is the precise geometric reason for this barrier, and does it apply to other settings beyond tensors?

2. Methodology

The author employs algebraic geometry, specifically scheme-theoretic methods, to formalize and generalize the problem.

Shift to Projective Geometry: The setting moves from vector spaces $W$ to projective spaces $\mathbb{P}(W)$ .
Introduction of Cactus Varieties: The core innovation is the use of cactus varieties ( $K_r(X)$ $K_{r} (X)$ ). Unlike secant varieties ( $\sigma_r(X)$ $σ_{r} (X)$ ), which are defined by spans of $r$ $r$ distinct points, cactus varieties are defined by the linear spans of finite subschemes of degree $r$ $r$ .
- Secant Variety: $\sigma_r(X) = \overline{\bigcup \langle x_1, \dots, x_r \rangle}$ where $x_i \in X$ .
- Cactus Variety: $K_r(X) = \overline{\bigcup \langle R \rangle}$ where $R \subset X$ is a finite subscheme of degree $r$ .
Rank Definitions: The paper defines:
- $X$ -rank ( $r_X$ ) and $X$ -border rank ( $br_X$ ) based on secant varieties.
- $X$ -cactus rank ( $cr_X$ ) and $X$ -border cactus rank ( $bcr_X$ ) based on cactus varieties.
Deformation Theory: The proofs rely heavily on the properties of finite subschemes and their deformations. Specifically, the author uses the fact that on a smooth variety, any finite subscheme can be deformed into an open dense subset (Proposition 2.1).
Vector Bundle Analysis: The paper analyzes the pullback of natural vector bundle maps associated with matrices of linear forms to show how the rank of the matrix behaves on the linear spans of finite schemes.

3. Key Contributions and Results

A. The Cactus Barrier Theorem (Theorem 1.3 & Corollary 4.5)

The central result establishes a hard limit on linear rank methods:

Statement: If a variety $X$ is contained in the locus where a matrix $M$ has rank $\le k$ (i.e., $X \subset \{ \text{rk } M \le k \}$ ), then the cactus variety $K_r(X)$ is contained in the locus where the matrix has rank $\le k \cdot r$ .
Implication: If the generic cactus rank of $X$ is $g$ (meaning $K_g(X) = \mathbb{P}(W)$ ), then for any point $F \in \mathbb{P}(W)$ , the matrix rank satisfies $\text{rk } M(F) \le k \cdot g$ .
Conclusion: Linear rank methods can never detect a border rank higher than the generic border cactus rank. If the border cactus rank is significantly lower than the border rank (which is often the case), linear methods are fundamentally incapable of proving the true lower bound.

B. Generalization to Smooth Loci

The paper proves that the barrier applies even if $X$ is not closed, provided we look at its smooth locus $X_0$ .

Theorem 3.2: The generic cactus variety of $X$ is equal to the cactus variety of its smooth locus $X_0$ .
Theorem 4.2: If $X$ is a smooth variety contained in $\{ \text{rk } M \le k \}$ , then $K_r(X) \subset \{ \text{rk } M \le k \cdot r \}$ . This confirms that the barrier is robust for smooth varieties (the standard case in tensor problems).

C. Quantitative Example: The $6m-4$ Barrier

The paper explains the specific bound $6m-4$ for $\mathbb{C}^m \otimes \mathbb{C}^m \otimes \mathbb{C}^m$ :

For the Segre embedding $X = \mathbb{P}^{m-1} \times \mathbb{P}^{m-1} \times \mathbb{P}^{m-1}$ , the generic cactus rank is known to be $6m-4$ .
Since $K_{6m-4}(X)$ fills the ambient space, no linear rank method can prove a border rank lower bound greater than $6m-4$ .
This recovers and geometrically explains previous asymptotic barriers found by [EGOW18] and [GMOW19].

D. Scope of the Barrier

The author argues that this barrier likely extends beyond simple linear rank methods:

Other methods like flag conditions and Weak Border Apolarity also seem to be bounded by cactus ranks.
The paper suggests that breaking this barrier requires methods that can distinguish between smoothable and non-smoothable finite schemes (deformation theory).

4. Significance

Geometric Explanation: The paper provides the first rigorous geometric explanation for why certain lower bound techniques fail. It shifts the focus from "finding a better matrix" to understanding the geometry of finite subschemes.
Limitation of Current Techniques: It formally proves that a vast class of existing lower bound techniques (linear rank methods, apolarity-based methods) are intrinsically limited by the cactus rank, not the border rank.
New Research Directions:
- Barrier Breaking: To prove super-linear lower bounds (e.g., for the matrix multiplication tensor), one must develop methods that can detect the difference between the border rank and the border cactus rank. This requires solving problems related to the smoothability of finite schemes.
- Grassmann Cactus Varieties: The paper introduces Grassmann cactus varieties as a potential tool for refining these barriers, suggesting that the gap between known lower bounds and true tensor ranks might be narrowed by studying these more general structures.
Unification: It unifies various disparate results in complexity theory (tensor rank, Waring rank, secant varieties) under the single framework of cactus varieties and scheme theory.

Summary

Jarosław Buczyński's paper demonstrates that linear rank methods are fundamentally blind to the difference between border rank and border cactus rank. By utilizing the geometry of cactus varieties (spans of finite schemes), the author proves that these methods cannot yield lower bounds exceeding the generic cactus rank. This explains the persistent $6m-4$ barrier in tensor rank problems and suggests that future breakthroughs in algebraic complexity must rely on deformation theory and the study of non-smoothable schemes.