3.17.5 Tensor Row Vector Representation Limit

Tensor Row Vector Representation Limit is the boundary of applicability of the row-array convention for covectors, marking the circumstances under which representing a dual-space element as a finite one-row list of numbers ceases to be adequate, well-defined, or canonical. While the row form is a complete and faithful description of a covector on a finite-dimensional vector space once a basis is fixed, this representation reaches structural limits in infinite-dimensional settings, in the absence of a distinguished basis, under certain non-invertible transformations, and when covectors are combined into higher-rank tensors that a single row cannot encode.

Limits Arising from Dimension

Finite Versus Infinite Dimension

The row vector representation relies on expanding a covector ω as a finite sum ω = ωᵢeⁱ over a finite dual basis. When V is infinite-dimensional, an element of the algebraic dual V* need not be expressible as any such finite or even countably summable combination of dual basis functionals, since the algebraic dual of an infinite-dimensional space is strictly larger than the span of any dual basis constructed from a Hamel basis of V. In this setting, a row array, understood as a finite or sequential list of scalars, cannot capture every linear functional, and the representation is only meaningful for those covectors, such as continuous functionals on a normed space with a Schauder basis, for which a well-defined infinite but convergent expansion exists.

Truncation and Approximation

Even when an infinite expansion exists, representing a covector by a finite row necessarily truncates the true functional, introducing an approximation. The size of the omitted tail depends on the decay of the coefficients ωᵢ and the topology of V*, so the row vector representation limit in this context is a genuine numerical or analytic error bound, not merely a formal inconvenience.

Limits Arising from Basis Choice

No Canonical Row Without a Basis

The row form always presupposes a specific basis of V and its induced dual basis. A covector considered purely as an abstract element of V*, with no basis fixed, has no canonical row representation at all; any claim about "the" row form of ω implicitly smuggles in a choice that must be made explicit. This is a conceptual limit rather than a numerical one: the representation exists only relative to auxiliary data not contained in ω itself.

Degeneration Under Singular Change of Basis

If an attempted change of coordinates is described by a non-invertible matrix, it does not define a genuine change of basis, and the transformation law ω̃ = ωA for row components breaks down as a bijective correspondence. Applying such a singular A can map distinct covectors to the same row array or fail to produce a consistent dual basis at all, so the row representation is limited to invertible changes of coordinates by its very construction.

Limits Under Pullback by Non-Injective or Non-Surjective Maps

Loss of Distinguishing Information

When a covector ω on W is pulled back along a linear map T: V → W via Tω = ωT, the resulting row vector on V may fail to distinguish covectors that differed only on directions in W outside the image of T. If T is not surjective, several different covectors on W can pull back to the identical row array on V, so the row representation of Tω carries strictly less information than ω did, and this information loss is an intrinsic limit of representing pullback purely at the level of row arrays without also recording T's image.

Failure to Represent Covectors Outside the Row Space

Conversely, if T is not surjective, not every covector on V arises as some Tω; the achievable pulled-back row vectors are confined to the row space of T, a subspace of V of dimension equal to the rank of T. The row vector representation of pulled-back covectors is thus limited to this proper subspace whenever rank(T) < dim(V), and no choice of ω on W can produce a pullback row vector outside it.

Limits in Representing Higher-Rank Tensors

Beyond Rank-One Covariant Objects

A single row array naturally represents only a rank-one covariant tensor, that is, one covector. Symmetric bilinear forms, antisymmetric two-forms, and higher-rank covariant tensors require an array with more than one row, such as an n×n matrix for rank two, or a genuinely multi-indexed array for higher rank. The row vector representation, taken strictly as a single 1×n list, has no direct extension to these objects without generalizing to matrices or higher-order arrays, marking a structural limit on what a bare row vector can encode.

Partial Contraction and the Limit of the Row Metaphor

Even for rank-two covariant tensors, operations such as contracting only one of the two indices with a vector produce an intermediate object, a covector depending on the uncontracted slot, that is naturally represented as a row only after the remaining index has been fixed or otherwise handled; the row vector picture ceases to be self-contained once more than one covariant slot participates in a computation simultaneously.

Numerical and Computational Limits

Conditioning Under Near-Singular Transformations

In practical numerical work, even when a change-of-basis matrix A is technically invertible, if A is ill-conditioned, meaning close to singular, the transformation ω̃ = ωA can amplify small errors in ω into large errors in ω̃. The row vector representation remains mathematically valid in this regime but becomes practically unreliable, illustrating a further limit that is numerical rather than structural: the representation degrades in stability as the basis approaches a degenerate configuration, even though the underlying linear algebra remains well-posed.