Minimal Border Rank Tensors

• Minimal border rank tensors are concise tensors achieving the minimal border rank dictated by dimensional constraints, often serving as structure tensors for smoothable algebras.
• They are algebraically characterized by binding properties and 111-algebras, enabling classification via module-theoretic invariants in low-dimensional cases.
• These tensors play a crucial role in algebraic complexity, informing fast matrix multiplication methods and offering insights into tensor degeneration phenomena.

Minimal border rank tensors are tensors whose border rank attains the smallest possible value dictated by dimensionality constraints in the relevant ambient space. This notion plays a foundational role in both algebraic complexity theory and the geometry of secant varieties, with implications ranging from the study of efficient matrix multiplication to the theory of homogeneous forms and their associated algebraic schemes.

1. Definitions and Structural Invariants

Let $V_1, V_2, V_3$ be finite-dimensional vector spaces over an algebraically closed field $k$, typically $\mathbb{C}$. Given a tensor $T \in V_1 \otimes V_2 \otimes V_3$, the rank $R(T)$ is the least $r$ such that $T$ can be written as a sum of $r$ pure tensors. The border rank $\underline{R}(T)$ is the smallest $r$ such that $T$ lies in the Zariski closure of tensors of rank at most $r$. Conciseness is imposed if the tensor does not lie in a proper subtensor space, equivalently if all contraction maps $V_i^* \to V_j \otimes V_k$ ($i \neq j \neq k$) are injective.

Minimal border rank refers to concise tensors $T \in k^n \otimes k^n \otimes k^n$ with $\underline{R}(T)=n$—the sharp lower bound dictated by dimension. This category encompasses structure tensors of smoothable commutative algebras of dimension $n$ (Bläser et al., 2016). The notion extends to the symmetric setup, where for $F \in S^d V^*$, minimal border rank is defined by $\underline{r}(F) = \dim V$ (Huang et al., 2019, Buczyńska et al., 2019).

Key auxiliary notions:

• Smoothable rank: Minimal $r$ for which $T$ lies in the span of a scheme that is a flat limit of $r$ distinct points (i.e., smoothable scheme).
• Cactus rank: Minimal $r$ for which $T$ lies in the span of a subscheme (possibly non-smoothable) of length $r$.

For concise minimal border rank tensors, one always has $$\underline{r}(T) \leq \mathrm{cr}(T) \leq \mathrm{sr}(T)$$ (Huang et al., 2019).

2. Algebraic Characterization and Classification

A tensor $T$ of minimal border rank is characterized as follows:

• If $T$ is “binding” (i.e., contractions in two directions yield full rank), then $T$ is equivalent to the structure tensor of a commutative, associative, smoothable algebra of dimension $n$ (Bläser et al., 2016).
• For $$T \in \mathbb{C}^m \otimes \mathbb{C}^m \otimes \mathbb{C}^m$$, every concise minimal border rank tensor for $m \leq 5$ is classified via module-theoretic methods: $1_*$-generic tensors correspond to modules over polynomial rings with module-theoretic invariants encapsulating the Strassen/End-closed algebra (Jagiełła et al., 2024, Jelisiejew et al., 2022).

The 111-algebra (or “triple intersection algebra”) emerges as a robust invariant: for concise $T$, it is the commutative $m$-dimensional algebra of endomorphisms $(X, Y, Z)$ acting simultaneously on the three factors such that $X.T = Y.T = Z.T$ (Jelisiejew et al., 2022). For $m\leq 5$, the zero locus of the 111-equations cuts out the concise minimal border rank locus set-theoretically.

In the symmetric setting, the apolar algebra of $F$—the quotient by its annihilator—has dimension $n$ if and only if $F$ is concise of minimal border rank, and is smoothable if and only if $F$ is “tame”; otherwise, $F$ is “wild” (Huang et al., 2019).

3. Geometric and Tensorial Constructions

The minimal border rank locus is of central importance, as every tensor of border rank $\leq m$ is a restriction of a minimal border rank tensor (Jagiełła et al., 2024). The following families exemplify the phenomenon:

• Structure tensors of smoothable algebras: The Coppersmith-Winograd tensor $T_{CW}$ is minimal border rank, as is any structure tensor of an algebra smoothable to $k^n$ (Bläser et al., 2016, Homs et al., 2021).
• Explicit families: The Alexeev–Forbes–Tsimerman tensors $T_{2^k}$ have border rank $2^k$ and (much higher) ordinary rank $2^{k+1}-1$ (Landsberg, 2012).
• Highest weight vector tensors and new smoothable or monomial algebra constructions provide further minimal border rank tensors with implications for the value $\omega$ in the complexity of matrix multiplication (Homs et al., 2021).

For $m \leq 5$, every indecomposable minimal border rank tensor degenerates to the curvilinear algebra tensor (Jagiełła et al., 2024). The degeneration order of these tensors is completely described by module invariants and the Białynicki-Birula decomposition.

4. Wild Versus Tame, Vanishing Hessian, and Border VSPs

The distinction between “tame” and “wild” is ecosystem-defining in this context:

• Wild forms/tensors: Smoothable rank’s strict separation from border rank. In the concise, minimal border rank case, wildness is equivalent to 1-degeneracy (failure of any contraction to give a full-rank matrix), and in symmetric forms to vanishing of the Hessian (Jelisiejew et al., 2022, Huang et al., 2019).
• Vanishing Hessian criterion (symmetric case): For concise minimal border rank $F$ of degree $d \geq 3$, $\mathrm{Hess}(F)=0$ if and only if $F$ is wild (i.e., smoothable rank or cactus rank exceeds dimension), but this equivalence fails for non-minimal border rank (Huang et al., 2019).
• Border varieties of sums of powers (border VSPs): Minimal border rank tensors may have border-VSPs with distinctive geometry, e.g., a projective space for the $G_d$ family, or reducible for wild cubic series $F_n$ (Huang et al., 2019).

In small tensor spaces, all minimal border rank tensors for $m \leq 4$ are tame; all 1-degenerate concise minimal border rank tensors are wild for $m=5$, and have been classified up to isomorphism (exactly five classes) (Jagiełła et al., 2024, Jelisiejew et al., 2022).

5. Border Apolarity, Hilbert Schemes, and Geometric Constructions

Minimal border rank tensors are deeply intertwined with multigraded Hilbert schemes and border apolarity:

• Border apolarity theorem: $\underline{r}(F) \leq r$ if and only if there is a homogeneous ideal $I$ in the slip ($\mathrm{Slip}_{r,X}$ component of the Hilbert scheme) with $I \subseteq \mathrm{Ann}(F)$. The corresponding border variety of sums of powers is $\mathrm{bVSP}(F, r)$ (Buczyńska et al., 2019).
• Hilbert function lower bounds: Minimal border rank for concise forms forces ideals with at least $n-1$ minimal generators in high degree, and the analogue for higher Segre/Veronese products persists in the multigraded setting (Buczyńska et al., 2019).
• Border VSP geometry: VSPs for minimal border rank wild forms may be reducible and carry combinatorially rich components, as shown for certain cubic and higher-degree wild examples (Huang et al., 2019).

These features are crucial in the study of the border Comon's conjecture, which asserts that for concise symmetric tensors of minimal border rank, the symmetric border rank coincides with the border rank; this conjecture has been proved for several large families, including all sharp and tame tensors and those with $n \le d+1$ (Mańdziuk et al., 2024).

6. Applications, Algorithmic Relevance, and Lower Bound Mechanisms

Minimal border rank tensors are central in the search for efficient matrix multiplication algorithms:

• Every building block in fast matrix multiplication leverages minimal border rank tensors (notably the Coppersmith-Winograd construction) (Bläser et al., 2016).
• For explicit lower bounds relevant to geometric complexity theory, Young flattenings and non-commutative rank methods yield equations vanishing on the locus of tensors of border rank below $2m-4,2m-5$ (depending on $m$), with explicit criteria for nontriviality (Derksen et al., 2016, Landsberg, 2012, Gondi, 25 Aug 2025).
• The module-theoretic and 111-algebra methods are employed for the explicit classification and detection of wildness, and their failure modes elucidate irreversibility barriers in degeneration diagrams for small $m$ (Jelisiejew et al., 2022, Jagiełła et al., 2024).

Recent advances underscore:

• Uniformity results for the realization of points of given border rank in infinite rank regimes and transfer of border rank bounds via GL-variety arguments (Bik et al., 2023).
• Complete classification for $m \leq 5$ and the absence of non-smoothable (wild) concise minimal border rank tensors for $m \leq 4$ (Jagiełła et al., 2024).

7. Open Problems and Outlook

Active areas and open conjectures include:

• Full classification of minimal border rank tensors beyond $m=6$ (where infinite moduli appear) (Jagiełła et al., 2024).
• Comprehensive understanding of border VSPs and their reducibility for wild forms in higher dimension/degrees (Huang et al., 2019).
• Sharp lower bounds for border rank in diverse families of GL(V)-invariant tensors and generalization to higher Schur functor types, with Kempf collapsing providing candidate non-minimal examples (Gondi, 25 Aug 2025).
• Characterization and detection of the wild/tame boundary, potential obstructions to the border Comon's conjecture in larger parameter regimes (Mańdziuk et al., 2024).

Minimal border rank tensors serve as the nexus point for tensor classification, border rank geometry, algebraic degeneration, and complexity lower bounds, providing a unifying structural framework for advances throughout multilinear algebra, complexity, and algebraic geometry.