Matthias Christandl,Koen Hoeberechts,Harold Nieuwboer,Péter Vrana,Jeroen Zuiddam
Abstract:Asymptotic tensor rank is notoriously difficult to determine. Indeed, determining its value for the $2\times 2$ matrix multiplication tensor would determine the matrix multiplication exponent, a long-standing open problem. On the other hand, Strassen's asymptotic rank conjecture makes the bold claim that asymptotic tensor rank equals the largest dimension of the tensor and is thus as easy to compute as matrix rank. Despite tremendous interest, much is still unknown about the structural and computational properties of asymptotic rank; for instance whether it is computable.
We prove that asymptotic tensor rank is "computable from above", that is, for any real number $r$ there is an (efficient) algorithm that determines, given a tensor $T$, if the asymptotic tensor rank of $T$ is at most $r$. The algorithm has a simple structure; it consists of evaluating a finite list of polynomials on the tensor. Indeed, we prove that the sublevel sets of asymptotic rank are Zariski-closed (just like matrix rank). While we do not exhibit these polynomials explicitly, their mere existence has strong implications on the structure of asymptotic rank.
As one such implication, we find that the values that asymptotic tensor rank takes, on all tensors, is a well-ordered set. In other words, any non-increasing sequence of asymptotic ranks stabilizes ("discreteness from above"). In particular, for the matrix multiplication exponent (which is an asymptotic rank) there is no sequence of exponents of bilinear maps that approximates it arbitrarily closely from above without being eventually constant. In other words, any upper bound on the matrix multiplication exponent that is close enough, will "snap" to it. Previously such discreteness results were only known for finite fields or for other tensor parameters (e.g., asymptotic slice rank). We obtain them for infinite fields like the complex numbers.
What problem does this paper attempt to address?
The main problems that this paper attempts to solve are related to the computation and structural properties of the asymptotic tensor rank. Specifically:
1. **Computation problem**: The paper proves that the asymptotic tensor rank can be computed from above, that is, for any real number \(r\), there exists an algorithm that can determine whether the asymptotic rank of a given tensor \(T\) is less than or equal to \(r\). This algorithm is implemented by evaluating a finite list of polynomials.
2. **Structural properties**: The paper proves that the sublevel sets of the asymptotic rank are Zariski - closed, which means that these sets can be accurately described by the zeros of polynomials. This result further indicates that the asymptotic rank is semi - continuous, similar to the property of the matrix rank.
3. **Discreteness**: The paper proves that the values of the asymptotic rank on all tensors form a well - ordered set, that is, any non - increasing sequence of asymptotic ranks will eventually stabilize. In particular, for the matrix multiplication exponent \(\omega\), this means that there exists a constant \(\epsilon> 0\) such that no tensor has an exponent between \(\omega\) and \(\omega+\epsilon\).
4. **Generalization**: The paper also extends these results to a broader class of tensor parameters, including all functions in Strassen's asymptotic spectrum. This includes the asymptotic slice rank and the asymptotic subrank.
### Specific contributions
- **Polynomial characteristics**: The paper proves that the upper bound of the asymptotic rank can be computed by polynomials. Specifically, for any "computable" field \(F\) and \(k\geq3\), for any \(r\in\mathbb{R}\), there exists an algorithm that, given any \(k\)-tensor \(T\), can determine whether its asymptotic rank is less than or equal to \(r\).
- **Zariski - closeness**: The paper proves that the sublevel sets of the asymptotic rank are Zariski - closed on any field \(F\), which means that each such set can be accurately described by the zeros of a finite number of polynomials.
- **Discreteness from above**: The paper proves that the values of the asymptotic rank on all tensors form a well - ordered set, that is, any non - increasing sequence of asymptotic ranks will eventually stabilize.
- **Asymptotic spectrum**: The paper generalizes the above results to a broader class of tensor parameters, including all functions in Strassen's asymptotic spectrum.
### Significance
These results not only deepen the theoretical understanding of the asymptotic rank but also provide new tools for practical computation. In particular, for the matrix multiplication exponent \(\omega\), these results mean that the upper bound of \(\omega\) will be "locked" at a specific value, which is of great significance for understanding the complexity of matrix multiplication. In addition, these results also provide a new perspective for the study of other asymptotic tensor parameters.