2.19.3 Tensor Abstract Element Handling

Tensor Abstract Element Handling is the set of rules governing how an individual element of an abstractly defined vector space, and in particular of a tensor product space, is manipulated, compared, and represented without presupposing any specific internal structure for that element beyond what the vector space axioms and the defining relations of the construction provide. It addresses a subtlety that does not arise for ordinary coordinate vectors: an element of a tensor product space is not, in general, a single "simple" object built from one vector out of each factor, but an equivalence class of formal expressions subject to bilinearity relations, and handling such an element correctly requires rules for addition, scalar multiplication, decomposition, and equality that respect this quotient structure rather than any one representative of it.

Elements as Equivalence Classes

Construction via a Free Vector Space and Relations

One standard way to construct the tensor product V ⊗ W of two abstract vector spaces is to start from the free vector space generated by all ordered pairs (v, w), and then take the quotient by the subspace generated by the bilinearity relations, such as (v_1 + v_2, w) - (v_1, w) - (v_2, w) and (α v, w) - α (v, w). An element of V ⊗ W handled abstractly is, precisely, an equivalence class of formal sums of such pairs under this quotient, denoted using the symbol ⊗ in place of the pair notation.

Well-Definedness of Operations on Classes

Because an element is an equivalence class rather than a single formal sum, any operation defined on tensor product elements, such as addition or a linear map out of V ⊗ W, must be checked to give the same result no matter which representative formal sum is used to compute it. This well-definedness check, verifying that a definition respects the bilinearity relations, is the central technical task in abstract element handling and is what justifies treating the equivalence class itself, rather than any particular sum representing it, as "the" tensor.

Elementary Tensors Versus General Elements

Elementary (Decomposable) Tensors

An element of the form v ⊗ w, built from a single vector v ∈ V and a single vector w ∈ W, is called an elementary or decomposable tensor. Elementary tensors are the images of the canonical bilinear map used in the universal property of the tensor product, and every element of V ⊗ W can be written as a finite sum of elementary tensors, though not, in general, as a single elementary tensor.

Non-Uniqueness of Decomposition Into Elementary Tensors

A given general element x of V ⊗ W typically admits more than one representation as a sum of elementary tensors, since v ⊗ w + v ⊗ w' = v ⊗ (w + w') and similar identities allow different-looking sums to describe the identical element. Abstract element handling therefore treats such alternative decompositions as interchangeable, and any well-defined operation on x must give the same answer regardless of which decomposition is used to compute it.

Manipulation Rules Inherited From the Vector Space Structure

Addition and Scalar Multiplication

Elements of V ⊗ W are added and scaled using the ordinary vector space operations applied to the equivalence classes, [x] + [y] = [x + y] and α [x] = [α x], where the bracket notation makes explicit that these operations must be, and are, independent of the chosen representative sums for x and y.

Multilinearity as a Handling Rule, Not an Extra Assumption

The bilinearity relations used to construct V ⊗ W are not optional properties checked after the fact; they are the defining relations that determine which formal sums represent the same abstract element. Correct element handling means always reducing an expression using these relations before comparing two elements for equality or before applying a linear map defined via the universal property.

Rank as an Element-Level Invariant

Definition of Tensor Rank

Given a general element x ∈ V ⊗ W, its rank is defined as the minimum number of elementary tensors needed in any sum representing x. This is a property of the abstract element itself, not of a particular decomposition, since it is defined by minimizing over every possible representation of x.

Rank Versus Ambient Dimension

For finite-dimensional V and W of dimensions m and n, the rank of any element of V ⊗ W is bounded above by min(m, n), and only elements of rank one are elementary tensors; elements of higher rank are genuinely new objects produced by the tensor product construction, existing in V ⊗ W but not decomposable into a single pair from V and W.

Diagrammatic Summary

The diagram shows two different formal sums of elementary tensors, related by the bilinearity relations, both collapsing under the quotient construction to the same single abstract element x, illustrating why abstract element handling must always operate on the equivalence class rather than on any one representative sum.