4.22.5 Tensor Multilinear Representation Limit

Tensor Multilinear Representation Limit is the collection of boundaries beyond which the standard means of representing a multilinear map, component arrays, matrices, hypermatrices, cease to apply cleanly or cease to apply at all, arising from infinite dimensionality, from degenerate arity, from the sheer computational cost of high arity, and from the inherently non-constructive nature of the universal property itself.

The Infinite-Dimensional Limit

No Finite Component Array

When a factor space Vᵢ is infinite-dimensional, there is no finite basis against which to tabulate a component array, and the representation of a multilinear map as a finite array of numbers simply fails to exist; a multilinear map on infinite-dimensional spaces must instead be described by an explicit formula, an integral, a series, or some other analytic expression, rather than by a table of coefficients.

Continuity Becomes an Additional Requirement

In infinite dimensions, an algebraic tensor product fails to capture the topological structure needed for many applications, since a purely algebraic multilinear map need not be continuous with respect to any natural topology on the spaces involved; representing multilinear maps that are also required to be continuous or bounded (as in Banach or Hilbert space settings) requires passing to a topological or completed tensor product, a genuinely different and more restrictive construction than the purely algebraic one used for finite-dimensional spaces.

The Arity Limit

Arity Zero: Constants

At arity n = 0, a "multilinear map" with no argument slots is simply a chosen element of W, a constant; the entire apparatus of multilinearity, argument slots, exchange rules, component arrays indexed by input positions, degenerates trivially, since there is nothing left to vary. Arity zero marks the lower boundary at which the multilinear representation framework stops saying anything nontrivial.

Arity One: Linear Maps

At arity n = 1, a multilinear map is an ordinary linear map, and its "component array" is simply the familiar matrix (or, for scalar output, the coefficient row) of that linear map; questions specific to multilinear structure, symmetry under argument exchange, alternation, tensor rank in the genuinely multilinear sense, are vacuous or trivial at this arity, since there is only one argument to permute.

Where Genuine Multilinear Phenomena Begin

The representation framework's distinctive features, matrices for bilinear forms, hypermatrices for higher arity, tensor rank as a nontrivial invariant, symmetric and alternating decompositions, first become meaningful at arity two and grow richer with each further increase in arity, marking arity two as the practical lower limit at which "multilinear" representation issues, as opposed to purely linear ones, begin to appear.

The Computational Limit at High Arity

Exponential Growth in Storage

As detailed in the higher array case, the number of entries needed to represent a general multilinear map of arity n on spaces of dimension d grows as dⁿ, an exponential rate that renders direct, explicit component-array representation infeasible for even moderately large n and d; this is a genuine practical limit on what can be represented and manipulated directly, independent of any deeper mathematical obstruction.

Intractability of Rank for Higher-Order Arrays

Determining the tensor rank of a general higher-order array, the minimal number of elementary tensors needed to express it, is known to be computationally intractable in general once the arity reaches three or more, in contrast to the arity-two (matrix) case, where rank is computed efficiently via row reduction; this marks a genuine limit on what can be feasibly computed about a multilinear map's representation, not merely stored.

The Non-Constructive Limit of the Universal Property

Existence Without an Algorithm

The universal property of the tensor product guarantees that a unique linear map f̃ exists satisfying f = f̃ ∘ ⊗ for any multilinear f, but the proof of this existence, via extension by linearity and a well-definedness argument on the quotient construction, does not itself supply an explicit, computable formula for f̃ in terms of anything more primitive than f itself; in this sense the universal-property representation is guaranteed to exist but is not, by itself, a constructive recipe distinct from simply restating the original multilinear map's values.

Uniqueness Up to Isomorphism, Not Uniqueness of Presentation

Because any two constructions satisfying the tensor product's universal property are canonically isomorphic but not identical as sets or as explicit formulas, there is no single "the" representation of the tensor product or of a multilinear map upon it; different constructions, quotients of free vector spaces, spaces of multilinear functionals, explicit bases of elementary tensors, all serve equally well, and choosing among them is a matter of convenience rather than mathematical necessity, marking a limit on how far the notion of "the" representation of a multilinear map can be pushed.

The Basis-Dependence Trade-Off as a Structural Limit

Basis-Free Insight Versus Explicit Computability

A fundamental tension runs through every representation of a multilinear map: basis-free representations, the linear map on the tensor product, the element of a dual tensor product, are best suited to structural reasoning but do not, by themselves, provide a means of explicit numerical computation; component-array representations provide explicit computability but depend on an arbitrary choice of basis and obscure which features of the array are intrinsic to the map and which are artifacts of that choice. No single representation achieves both basis-independence and direct computability simultaneously without additional work translating between the two.

Symmetric and Alternating Cases as Partial Relief

Symmetric and alternating multilinear maps offer partial relief from the general representation limit, since their component arrays compress substantially, to non-decreasing or strictly increasing index tuples respectively, reducing but not eliminating the tension between basis-free structure and explicit storage; general multilinear maps lacking either symmetry receive no such relief and remain subject to the full weight of the representation limits described here.

Why Recognizing These Limits Matters

Guiding the Choice of Method

Recognizing which representation limit is relevant to a given problem, infinite-dimensionality, high arity, need for an explicit algorithm, guides the choice of technique: a problem in infinite dimensions calls for analytic or functional-analytic methods rather than a finite array; a problem of high arity calls for approximate or structured decompositions rather than an exact minimal-rank factorization; a problem requiring an explicit formula cannot be resolved by an existence argument via the universal property alone.

Distinguishing Mathematical Existence From Practical Feasibility

The representation limits catalogued here separate what is guaranteed to exist mathematically, a unique linear map on the tensor product, a component array relative to any basis, from what is practically obtainable, an explicit formula, a computable rank, a tractable storage scheme, a distinction essential to correctly interpreting claims about multilinear maps in both pure and applied contexts.