Asymptotic Rank Research Articles

Universal points in the asymptotic spectrum of tensors

Motivated by the problem of constructing fast matrix multiplication algorithms, Strassen (FOCS 1986, Crelle 1987–1991) introduced and developed the theory of asymptotic spectra of tensors. For any sub-semiringX\mathcal {X}of tensors (under direct sum and tensor product), the duality theorem that is at the core of this theory characterizes basic asymptotic properties of the elements ofX\mathcal {X}in terms ofthe asymptotic spectrum ofX\mathcal {X}, which is defined as the collection of semiring homomorphisms fromX\mathcal {X}to the non-negative reals with a natural monotonicity property. The asymptotic properties characterized by this duality encompass fundamental problems in complexity theory, combinatorics and quantum information.Universal spectral pointsare elements in the asymptotic spectrum of the semiring ofalltensors. Finding all universal spectral points suffices to find the asymptotic spectrum ofanysub-semiring. The construction of non-trivial universal spectral points has been an open problem for more than thirty years. We construct, for the first time, a family of non-trivial universal spectral points over the complex numbers, calledquantum functionals. We moreover prove that the quantum functionals precisely characterise the asymptotic slice rank of complex tensors. Our construction, which relies on techniques from quantum information theory and representation theory, connects the asymptotic spectrum of tensors to the quantum marginal problem and entanglement polytopes.

Border Rank Nonadditivity for Higher Order Tensors

Whereas matrix rank is additive under direct sum, in 1981 Sch\"onhage showed that one of its generalizations to the tensor setting, tensor border rank, can be strictly subadditive for tensors of order three. Whether border rank is additive for higher order tensors has remained open. In this work, we settle this problem by providing analogs of Sch\"onhage's construction for tensors of order four and higher. Sch\"onhage's work was motivated by the study of the computational complexity of matrix multiplication; we discuss implications of our results for the asymptotic rank of higher order generalizations of the matrix multiplication tensor.

A gap in the slice rank of k-tensors

The slice-rank method, introduced by Tao as a symmetrized version of the polynomial method of Croot, Lev and Pach and Ellenberg and Gijswijt, has proved to be a useful tool in a variety of combinatorial problems. Explicit tensors have been introduced in different contexts but little is known about the limitations of the method.In this paper, building upon a method presented by Tao and Sawin, it is proved that the asymptotic slice rank of any k-tensor in any field is either 1 or at least k/(k−1)(k−1)/k. This provides evidence that straight-forward application of the method cannot give useful results in certain problems for which non-trivial exponential bounds are already known. An example, actually a motivation for starting this work, is the problem of bounding the size of sets of trifferent sequences, which constitutes a long-standing open problem in information theory and in theoretical computer science.

Rank and duality in representation theory

There is both theoretical and numerical evidence that the set of irreducible representations of a reductive group over local or finite fields is naturally partitioned into families according to analytic properties of representations. Examples of such properties are the rate of decay at infinity of “matrix coefficients” in the local field setting, and the order of magnitude of “character ratios” in the finite field situation. In these notes we describe known results, new results, and conjectures in the theory of “size” of representations of classical groups over finite fields (when correctly stated, most of them hold also in the local field setting), whose ultimate goal is to classify the above mentioned families of representations and accordingly to estimate the relevant analytic properties of each family. Specifically, we treat two main issues: the first is the introduction of a rigorous definition of a notion of size for representations of classical groups, and the second issue is a method to construct and obtain information on each family of representation of a given size. In particular, we propose several compatible notions of size that we call U-rank, tensor rank and asymptotic rank, and we develop a method called eta correspondence to construct the families of representation of each given rank. Rank suggests a new way to organize the representations of classical groups over finite and local fields—a way in which the building blocks are the “smallest” representations. This is in contrast to Harish-Chandra’s philosophy of cusp forms that is the main organizational principle since the 60s, and in it the building blocks are the cuspidal representations which are, in some sense, the “largest”. The philosophy of cusp forms is well adapted to establishing the Plancherel formula for reductive groups over local fields, and led to Lusztig’s classification of the irreducible representations of such groups over finite fields. However, the understanding of certain analytic properties, such as those mentioned above, seems to require a different approach.

Higher rank hyperbolicity

The large-scale geometry of hyperbolic metric spaces exhibits many distinctive features, such as the stability of quasi-geodesics (the Morse Lemma), the visibility property, and the homeomorphism between visual boundaries induced by a quasi-isometry. We prove a number of closely analogous results for spaces of rank $$n \ge 2$$ in an asymptotic sense, under some weak assumptions reminiscent of nonpositive curvature. For this purpose we replace quasi-geodesic lines with quasi-minimizing (locally finite) n-cycles of $$r^n$$ volume growth; prime examples include n-cycles associated with n-quasiflats. Solving an asymptotic Plateau problem and producing unique tangent cones at infinity for such cycles, we show in particular that every quasi-isometry between two proper $${\text {CAT}}(0)$$ spaces of asymptotic rank n extends to a class of $$(n-1)$$ -cycles in the Tits boundaries.

Towards a geometric approach to Strassen’s asymptotic rank conjecture

We make a first geometric study of three varieties in $\mathbb{C}^m \otimes \mathbb{C}^m \otimes \mathbb{C}^m$ (for each $m$), including the Zariski closure of the set of tight tensors, the tensors with continuous regular symmetry. Our motivation is to develop a geometric framework for Strassen's Asymptotic Rank Conjecture that the asymptotic rank of any tight tensor is minimal. In particular, we determine the dimension of the set of tight tensors. We prove that this dimension equals the dimension of the set of oblique tensors, a less restrictive class introduced by Strassen.

Asymptotic rank theorems

Asymptotic rank theorems

Asymptotic tensor rank of graph tensors: beyond matrix multiplication

We present an upper bound on the exponent of the asymptotic behaviour of the tensor rank of a family of tensors defined by the complete graph on $k$ vertices. For $k\geq4$, we show that the exponent per edge is at most 0.77, outperforming the best known upper bound on the exponent per edge for matrix multiplication ($k=3$), which is approximately 0.79. We raise the question whether for some $k$ the exponent per edge can be below $2/3$, i.e. can outperform matrix multiplication even if the matrix multiplication exponent equals 2. In order to obtain our results, we generalise to higher order tensors a result by Strassen on the asymptotic subrank of tight tensors and a result by Coppersmith and Winograd on the asymptotic rank of matrix multiplication. Our results have applications in entanglement theory and communication complexity.

Tensor surgery and tensor rank

We introduce a method for transforming low-order tensors into higher-order tensors and apply it to tensors defined by graphs and hypergraphs. The transformation proceeds according to a surgery-like procedure that splits vertices, creates and absorbs virtual edges and inserts new vertices and edges. We show that tensor surgery is capable of preserving the low rank structure of an initial tensor decomposition and thus allows to prove nontrivial upper bounds on tensor rank, border rank and asymptotic rank of the final tensors. We illustrate our method with a number of examples. Tensor surgery on the triangle graph, which corresponds to the matrix multiplication tensor, leads to nontrivial rank upper bounds for all odd cycle graphs, which correspond to the tensors of iterated matrix multiplication. In the asymptotic setting we obtain upper bounds in terms of the matrix multiplication exponent $\omega$ and the rectangular matrix multiplication parameter $\alpha$. These bounds are optimal if $\omega$ equals two. We also give examples that illustrate that tensor surgery on general graphs might involve the absorption of virtual hyperedges and we provide an example of tensor surgery on a hypergraph. Besides its relevance in algebraic complexity theory, our work has applications in quantum information theory and communication complexity.

Tensor rank is not multiplicative under the tensor product

The tensor rank of a tensor t is the smallest number r such that t can be decomposed as a sum of r simple tensors. Let s be a k-tensor and let t be an ℓ-tensor. The tensor product of s and t is a (k+ℓ)-tensor. Tensor rank is sub-multiplicative under the tensor product. We revisit the connection between restrictions and degenerations. A result of our study is that tensor rank is not in general multiplicative under the tensor product. This answers a question of Draisma and Saptharishi. Specifically, if a tensor t has border rank strictly smaller than its rank, then the tensor rank of t is not multiplicative under taking a sufficiently hight tensor product power. The “tensor Kronecker product” from algebraic complexity theory is related to our tensor product but different, namely it multiplies two k-tensors to get a k-tensor. Nonmultiplicativity of the tensor Kronecker product has been known since the work of Strassen.It remains an open question whether border rank and asymptotic rank are multiplicative under the tensor product. Interestingly, lower bounds on border rank obtained from generalized flattenings (including Young flattenings) multiply under the tensor product.

Asymptotic rank of spaces with bicombings

The question, under what geometric assumptions on a space X an n-quasiflat in X implies the existence of an n-flat therein, has been investigated for a long time. It was settled in the affirmative for Busemann spaces by Kleiner, and for manifolds of non-positive curvature it dates back to Anderson and Schroeder. We generalize the theorem of Kleiner to spaces with bicombings. This structure is a weak notion of non-positive curvature, not requiring the space to be uniquely geodesic. Beside a metric differentiation argument, we employ an elegant barycenter construction due to Es-Sahib and Heinich by means of which we define a Riemannian integral serving us in a sort of convolution operation.

The Flexible Fourier Form and Local Generalised Least Squares De‐trended Unit Root Tests*

AbstractIn two recent papers Enders and Lee (2009) and Becker, Enders and Lee (2006) provide Lagrange multiplier and ordinary least squares de‐trended unit root tests, and stationarity tests, respectively, which incorporate a Fourier approximation element in the deterministic component. Such an approach can prove useful in providing robustness against a variety of breaks in the deterministic trend function of unknown form and number. In this article, we generalize the unit root testing procedure based on local generalized least squares (GLS) de‐trending proposed by Elliott, Rothenberg and Stock (1996) to allow for a Fourier approximation to the unknown deterministic component in the same way. We show that the resulting unit root tests possess good finite sample size and power properties and the test statistics have stable non‐standard distributions, despite the curious result that their limiting null distributions exhibit asymptotic rank deficiency.

The asymptotic rank of metric spaces

In this article we define and study a notion of asymptotic rank for metric spaces and show in our main theorem that for a large class of spaces, the asymptotic rank is characterized by the growth of the higher isoperimetric filling functions. For a proper, cocompact, simply connected geodesic metric space of non-positive curvature in the sense of Alexandrov the asymptotic rank equals its Euclidean rank.

Rank test for heteroscedastic functional data

In this paper, we consider (mid-)rank based inferences for testing hypotheses in a fully nonparametric marginal model for heteroscedastic functional data that contain a large number of within subject measurements from possibly only a limited number of subjects. The effects of several crossed factors and their interactions with time are considered. The results are obtained by establishing asymptotic equivalence between the rank statistics and their asymptotic rank transforms. The inference holds under the assumption of α -mixing without moment assumptions. As a result, the proposed tests are applicable to data from heavy-tailed or skewed distributions, including both continuous and ordered categorical responses. Simulation results and a real application confirm that the (mid-)rank procedures provide both robustness and increased power over the methods based on original observations for non-normally distributed data.

Rank tests in heteroscedastic multi-way HANOVA

This article develops rank tests for the nonparametric main factor effects and interactions in multi-way high-dimensional analysis of variance when the cell distributions are completely unspecified. The design can be balanced or unbalanced with the cell sample sizes fixed or tending to infinity. An arbitrary number of factors and all types of ordinal data are allowed. This extends the use of rank methods to the Neymann–Scott and triangular array problems. The asymptotic distribution of the rank statistics is obtained by showing their asymptotic equivalence to corresponding expressions based on the asymptotic rank transform. Compared with test procedures based on the original observations, the proposed rank procedures are free of moment conditions, converge to their limiting distribution faster, and have better power when the underlying distributions are heavy tailed or skewed. These advantages are demonstrated by simulations and an application to a real data set.

The Determination of Sample Sizes in the Comparison of Two Multinomial Proportions from Ordered Categories

Summary We consider sample size determination for ordered categorical data when the alternative assumption is the proportional odds model. In this paper the sample size formula proposed by Whitehead (Statistics in Medicine, 12, 2257–2271, 1993) is compared with the methods based on exact and asymptotic linear rank tests with Wilcoxon and trend scores. We show that Whitehead’s formula, which is based on a normal approximation, works well when the sample size is moderate to large but recommend the exact method with Wilcoxon scores for small sample sizes. The consequences of misspecification in models are also investigated.

Rates of convergence for linear rank statistics with dependent data

Asymptotic distributions and rates of convergence are obtained for linear rank statistics with dependent random variables. It is assumed that the dependent random variables can be written as a set of N independent random vectors. The lengths of the independent random vectors are allowed to be unequal, but bounded; and the distribution functions of the vectors are arbitrary. Some results are given for certain unbounded score functions with approximate scores; most results assume that the score function has a bounded second derivative. The results are especially suitable for deriving asymptotic rank tests for models with dependent observations and unequal cell sizes, such as stratified random effects models and repeated measures designs with missing observations.

A Unified Approach to Rank Tests for Multivariate and Repeated Measures Designs

Abstract A unified approach to asymptotic rank tests is presented for a wide class of univariate and multivariate models, including repeated measures designs without compound symmetry. The models are assumed to be balanced and complete with more than one replication per cell. The proposed rank tests are constructed by ranking all of the observations together regardless of row, column, or component membership. This method of ranking offers increased power over the method of n-rankings. The resulting test statistic is a quadratic form in linear rank statistics. Asymptotic distributions are determined under Pitman alternatives that allow for both scale and location alternatives. The resulting statistics include tests for factor effects (both scale and location differences) in one-, two-, and higher-way layouts with repeated measures on one or several factors without assuming equicorrelation. Also included are tests for the multivariate two- and k-sample problem, as well as multivariate versions of the tests for multiway layouts and repeated measures designs. For many of the repeated measures designs without equicorrelation, no other rank based statistics have been previously studied. The results also include the asymptotic distributions for many possible rank transform tests for univariate and multivariate models, as well as a rich class of aligned rank tests.

A Unified Approach to Rank Tests for Multivariate and Repeated Measures Designs

Abstract A unified approach to asymptotic rank tests is presented for a wide class of univariate and multivariate models, including repeated measures designs without compound symmetry. The models are assumed to be balanced and complete with more than one replication per cell. The proposed rank tests are constructed by ranking all of the observations together regardless of row, column, or component membership. This method of ranking offers increased power over the method of n-rankings. The resulting test statistic is a quadratic form in linear rank statistics. Asymptotic distributions are determined under Pitman alternatives that allow for both scale and location alternatives. The resulting statistics include tests for factor effects (both scale and location differences) in one-, two-, and higher-way layouts with repeated measures on one or several factors without assuming equicorrelation. Also included are tests for the multivariate two- and k-sample problem, as well as multivariate versions of the tests for multiway layouts and repeated measures designs. For many of the repeated measures designs without equicorrelation, no other rank based statistics have been previously studied. The results also include the asymptotic distributions for many possible rank transform tests for univariate and multivariate models, as well as a rich class of aligned rank tests.

Robustness of One-and Two-Sample Rank Tests Against Gross Errors

The classical nonparametric hypotheses of symmetry and equality of distribution functions are extended to hypotheses of approximate symmetry and approximate equality, by allowing for gross errors, which is indispensable for practical applications. It is shown that for these hypotheses one- and two-sample rank statistics maintain their distribution freeness, which now refers to their stochastically extreme laws. These laws are evaluated asymptotically, also under similarly extended alternatives, in the author's previous local framework, which has not yet covered rank statistics due to a subtle asymptotic fine structure of infinitesimal neighborhoods. Consequences on asymptotic maximum size, minimum power and relative efficiency of rank tests are drawn. In particular, it is shown that if the scores are unbounded, then rank tests fail completely; and by suitable truncation of the classically optimal scores, an asymptotic maximin rank test is obtained.