What Is the Probability That a Random Symmetric Tensor Is Close to Rank-One?

Alberto Cazzaniga, Antonio Lerario, Andrea Rosana

Here is a summary of the paper tailored for a technically literate audience:

TL;DR

This paper provides an exact mathematical formula for the probability that a random real symmetric tensor is "close" to a rank-one tensor. By translating this algebraic problem into a differential geometry question about the volume of a "tube" around the Veronese variety, the authors use Weyl's tube formula and Random Matrix Theory to completely solve it, yielding insights relevant to quantum entanglement and tensor approximation.

What Problem Does This Paper Address?

In machine learning, signal processing, and quantum computing, researchers frequently use tensors (multi-dimensional arrays). Because high-order tensors are incredibly complex and memory-intensive, it is common practice to approximate them using rank-one tensors—the simplest possible building blocks, formed by the outer product of vectors. In quantum mechanics, for example, a rank-one tensor represents a "separable" (non-entangled) state, and the distance a state is from rank-one defines its "geometric measure of entanglement."

The authors ask a fundamental statistical question: If you draw a real symmetric tensor at random from a natural Gaussian distribution, what is the exact probability that it lies within a specific distance $\delta$ of some rank-one tensor?

Methods

The authors tackle this by translating the algebraic problem into a problem of differential geometry:

  1. Geometric Translation: They map the space of symmetric tensors to the space of homogeneous polynomials endowed with the Bombieri-Weyl (Kostlan) probability distribution. Under this mapping, the set of all rank-one tensors corresponds to a well-known geometric object called the spherical Veronese variety.
  2. Weyl's Tube Formula: Asking if a tensor is within a distance $\delta$ of a rank-one tensor is geometrically equivalent to asking for the volume of a "tubular neighborhood" of radius $\epsilon$ (where $\epsilon = \arcsin(\delta)$) around the Veronese variety on a sphere. To compute this, they use Weyl's tube formula.
  3. Random Matrix Theory (RMT): To actually evaluate Weyl's formula, one needs to know the "curvature coefficients" of the Veronese variety. The authors brilliantly prove that the second fundamental form (which dictates curvature) of this variety can be explicitly described using Gaussian Orthogonal Ensemble (GOE) matrices. This reduces complex geometric integrals to computing the expected values of random matrix determinants.

Datasets

Not discussed in the paper. (This is a theoretical mathematics paper).

Key Insights and Ideas

  • The Random Matrix Connection: The most surprising mathematical insight is the direct link between the extrinsic geometry of the Veronese variety and GOE random matrices. By proving that the Weingarten operator (which encodes curvature) behaves like a GOE matrix, the authors bridge algebraic geometry and random matrix theory.
  • The "Reach" Guarantees Unique Approximations: The authors explicitly calculate the reach of the Veronese variety—the maximum radius a tube can have before it starts overlapping itself. Geometrically, this is the manifold's bottleneck width. Practically, it means that any tensor closer to rank-one than this specific reach distance is mathematically guaranteed to have a unique best rank-one approximation.
  • Symmetry Simplifies Everything: Because the Veronese variety is an orbit under the orthogonal group, its geometry looks exactly the same from every point. The authors only had to compute the curvature at a single, simple coordinate point to understand the entire space.

Key Results

  • Exact Probability Formula: The authors provide a massive, exact closed-form integral formula (Theorem E) for the probability of a tensor being $\delta$-close to rank-one.
  • The Exact Reach: They found that the reach $\rho$ of the spherical Veronese variety is exactly $\pi/4$ for tensor degrees $2 \le d \le 5$, and $\frac{1}{\sqrt{2}}\sqrt{1 + \frac{1}{d-1}}$ for degrees $d \ge 6$.
  • Simplified Rational Normal Curves: For the specific case of rational normal curves, the intimidating probability formula simplifies beautifully to just $\sqrt{d} \delta^{d-1}$.
  • Matrix Case ($d=2$): When applied to standard symmetric matrices, the authors provide a closed-form formula for the probability that the Frobenius norm of a matrix is bounded by its smallest singular value, a novel and highly specific result for Random Matrix Theory.

Limitations and Future Work

  • Limitations: While mathematically exact, the general probability formula (Theorem E) is computationally "unpleasant," relying on complex combinations of Gamma functions and integrals. Furthermore, Weyl's tube formula is only valid for distances smaller than the "reach" of the manifold; the formula does not apply to tensors that are very far from rank-one.
  • Future Work: The authors propose generalizing this framework to higher-rank tensor approximations. Geometrically, this would involve looking at the volumes of tubular neighborhoods around secant varieties to the Veronese variety, which represent tensors of rank 2, 3, etc.

Summary updated

Abstract

We address the general problem of estimating the probability that a real symmetric tensor is close to rank-one tensors. Using Weyl's tube formula, we turn this question into a differential geometric one involving the study of metric invariants of the real Veronese variety. More precisely, we give an explicit formula for its reach and curvature coefficients with respect to the Bombieri-Weyl metric. These results are obtained using techniques from Random Matrix theory and an explicit description of the second fundamental form of the Veronese variety in terms of GOE matrices. Our findings give a complete solution to the original problem. In the case of rational normal curves it leads to a simple formula describing explicitly exponential decay with respect to the degree of the tensor.
BibTeX
@article{cazzaniga2024what,
  author = {Alberto Cazzaniga and Antonio Lerario and Andrea Rosana},
  title = {{What Is the Probability That a Random Symmetric Tensor Is Close to Rank-One?}},
  journal = {SIAM J. Appl. Algebra Geom.},
  year = {2024},
  doi = {10.1137/23m1551675},
}
Figure