5.10.5 Tensor Product Component Count Relation

Tensor Product Component Count Relation is the correspondence between the dimension of a tensor product space and the number of independent scalar components — the individual numerical entries — required to specify a general element of that space in coordinates, together with the reduced component counts that arise once symmetry, antisymmetry, or a decomposable (rank-one) restriction is imposed on the tensor.

Baseline Component Count

For a tensor product V1 ⊗ V2 ⊗ ⋯ ⊗ Vn of finite-dimensional spaces with dimensions d1, …, dn, a general element requires exactly

independent scalar components to specify, relative to a chosen basis: one component per basis tensor in the induced basis, with no fewer than this number sufficing, since the induced basis tensors are linearly independent and each contributes an independent degree of freedom.

Reduced Component Count for Decomposable Tensors

Requiring a tensor to be decomposable (a single product of factors) collapses this component count dramatically, since a decomposable tensor's coordinates are entirely determined by far fewer independent numbers than the full array contains.

Additive Component Count for Decomposable Tensors

A decomposable tensor v1 ⊗ v2 ⊗ ⋯ ⊗ vn is fully specified by the components of its individual factors, requiring only

independent numbers (up to the scalar-redistribution ambiguity discussed elsewhere), even though the resulting coordinate array still has ∏ di entries. The remaining entries are not independent; they are all determined by the entrywise multiplicative pattern once the factor components are fixed.

Component Count as a Compression Ratio

The ratio between the full component count ∏ di and the decomposable component count ∑ di quantifies the compression achieved by a decomposable representation, and this ratio grows without bound as either the number of factors n or their dimensions grow, which is the quantitative basis for treating decomposable and low-rank tensor representations as a compression technique in applied settings.

Reduced Component Count Under Symmetry

When a tensor is constrained to be symmetric or antisymmetric — built from a single vector space V of dimension d rather than from distinct factor spaces — the component count is reduced according to standard combinatorial formulas rather than by the multiplicative or additive patterns above.

Symmetric Tensors

A symmetric tensor of order p on a d-dimensional space, invariant under every permutation of its p slots, has independent component count equal to the number of multisets of size p from d elements,

which is substantially smaller than d^p once p exceeds one, reflecting that permuting the indices of a symmetric tensor does not produce a new independent component.

Antisymmetric Tensors

An antisymmetric (alternating) tensor of order p on a d-dimensional space instead has independent component count equal to the binomial coefficient

which vanishes entirely once p exceeds d, reflecting that any antisymmetric combination involving a repeated index must be zero, and that only genuinely distinct, unordered selections of p indices contribute independent components.

Component Count in Rank Decompositions

A tensor expressed as a sum of r decomposable terms has a component count intermediate between the fully general count and the single-decomposable-term count, scaling linearly in r.

Parameter Count of a Rank-r Decomposition

A rank-r canonical polyadic decomposition of a tensor in V1 ⊗ ⋯ ⊗ Vn requires r · (d1 + d2 + ⋯ + dn) independent numbers (again up to the standard scalar redistribution ambiguities across the r terms), growing linearly in r rather than staying fixed or growing with the full product ∏ di. This linear-in-r growth is what makes low-rank decompositions attractive whenever the target tensor can be well approximated using a value of r much smaller than would be required to reach the full component count of the ambient space.

Component Count Versus Practical Compressibility

Achieving the theoretical component count of a rank-r decomposition assumes the target tensor is genuinely well approximated at that rank; the component count relation describes only the size of the representation, not whether a tensor of interest actually admits a good low-rank approximation, which is a separate question addressed by tensor rank and approximation theory.

Illustrative Diagram

The shrinking bars illustrate how the number of independent components drops sharply moving from a full general array, to a bounded-rank decomposition, to a single decomposable term.