4.17.3 Tensor Universal Property Unique Linear Map

The tensor universal property ensures a unique linear map from tensor products to any bilinear map, central to algebraic structures and multilinear algebra.

Tensor Universal Property Unique Linear Map is the specific mathematical object guaranteed by the universal mapping property of the tensor product: given vector spaces V₁, ..., Vₙ over a field F, a target vector space W, and any multilinear map f: V₁ × ... × Vₙ → W, there exists exactly one linear map f̃: V₁ ⊗ ... ⊗ Vₙ → W such that f̃ composed with the canonical multilinear map ⊗ recovers f. This unique linear map is the mechanism through which multilinear phenomena on the factor spaces are translated, without loss of information, into linear phenomena on the tensor product.

Formal Statement

The Setup

Let V₁, ..., Vₙ and W be vector spaces over a field F. Let

\otimes : V_{1} \times ⋮ \times V_{n} \to V_{1} \otimes ⋮ \otimes V_{n}

denote the canonical map sending (v₁, ..., vₙ) to the elementary tensor v₁ ⊗ ... ⊗ vₙ. This map is multilinear but not linear: it is linear in each slot separately, yet the map as a whole does not respect addition or scaling across the full tuple.

The Property

For every multilinear map f: V₁ × ... × Vₙ → W, there exists a unique linear map

\tilde{f} : V_{1} \otimes ⋮ \otimes V_{n} \to W

satisfying

\tilde{f} (v_{1} \otimes ⋮ \otimes v_{n}) = f (v_{1}, \dots, v_{n})

for all v₁ ∈ V₁, ..., vₙ ∈ Vₙ. Equivalently, f = f̃ ∘ ⊗.

The diagram commutes: traveling along the top edge (f) gives the same result as traveling down through the tensor product and then along f̃.

Existence of the Unique Linear Map

Construction on Elementary Tensors

Every element of V₁ ⊗ ... ⊗ Vₙ is a finite sum of elementary tensors v₁ ⊗ ... ⊗ vₙ. The candidate map f̃ is first defined on elementary tensors by the rule

\tilde{f} (v_{1} \otimes ⋮ \otimes v_{n}) := f (v_{1}, \dots, v_{n})

and then extended by linearity to all finite sums.

Why the Extension Is Well Defined

The subtlety of existence is not the extension by linearity itself — any assignment on a spanning set extends linearly to the whole space — but rather that elementary tensors satisfy nontrivial relations. The same tensor t ∈ V₁ ⊗ ... ⊗ Vₙ typically has many different representations as a sum of elementary tensors, because the tensor product is built as a quotient of a free vector space by the subspace generated by the bilinearity relations:

(v_{i} + v_{i}') \otimes ⋮ = (⋮ v_{i} \otimes ⋮) + (⋮ v_{i}' \otimes ⋮)

and similarly for scalar multiplication in each slot. For f̃ to be a genuine function, its value cannot depend on which representative sum of elementary tensors is used. This is guaranteed precisely because f is multilinear: multilinearity of f mirrors the defining relations of the tensor product exactly, so any two representations of the same tensor are sent by the "assign, then sum" procedure to the same element of W. Formally, one constructs f̃ by first defining a linear map out of the free vector space on V₁ × ... × Vₙ using f, and then observing that this map vanishes on the subspace of relations, so it descends to a well-defined linear map on the quotient, which is V₁ ⊗ ... ⊗ Vₙ.

Linearity of the Result

Because f̃ is defined by extending an assignment linearly, it automatically satisfies

\tilde{f} (α t + β s) = α \tilde{f} (t) + β \tilde{f} (s)

for all t, s ∈ V₁ ⊗ ... ⊗ Vₙ and scalars α, β ∈ F. This is exactly the linearity claimed by the universal property.

Uniqueness of the Linear Map

The Argument

Suppose g: V₁ ⊗ ... ⊗ Vₙ → W is any linear map satisfying g(v₁ ⊗ ... ⊗ vₙ) = f(v₁, ..., vₙ) for all vᵢ. Since the elementary tensors span V₁ ⊗ ... ⊗ Vₙ, and g is required to agree with f̃ on every elementary tensor, linearity forces g to agree with f̃ on every finite sum of elementary tensors as well:

g (\sum c_{k} v_{1,k} \otimes ⋮ \otimes v_{n,k}) = \sum c_{k} f (v_{1,k}, \dots, v_{n,k}) = \tilde{f} (\sum c_{k} v_{1,k} \otimes ⋮ \otimes v_{n,k})

Since every tensor is such a finite sum, g = f̃ everywhere. A linear map is completely determined by its values on a spanning set, and the elementary tensors span the entire tensor product, which is the crux of why the linear map cannot fail to be unique once it is required to agree with f through the canonical map.

Why Linearity Is Essential to Uniqueness

Uniqueness relies critically on the requirement that the extension be linear, not merely a set-theoretic function agreeing with f on elementary tensors. A non-linear function could agree with f on elementary tensors while disagreeing with f̃ on other elements, since a general tensor may be written as a sum of elementary tensors in more than one way, and a non-linear rule need not respect these coincidences. Restricting to linear maps removes this ambiguity entirely.

Interpretation

Translating Multilinear Data into Linear Data

The practical significance of the unique linear map is that it converts a multilinear map, an object with a genuinely different algebraic behavior in each argument, into a single linear map on a single vector space. Once the tensor product is built, the entire theory of linear maps, including rank, kernel, image, matrix representation, and duality, becomes available for analyzing what was originally multilinear data.

Bijective Correspondence

The universal property does more than produce one linear map for one multilinear map: it establishes a natural bijection between the set of multilinear maps out of V₁ × ... × Vₙ and the set of linear maps out of V₁ ⊗ ... ⊗ Vₙ:

Multilinear (V_{1}, \dots, V_{n}; W) ≅ Hom (V_{1} \otimes ⋮ \otimes V_{n}, W)

sending f to f̃ and, in the reverse direction, sending any linear map g to the multilinear map g ∘ ⊗. These two assignments are mutually inverse, which is exactly the content of the existence and uniqueness statements taken together.

Characterizing the Tensor Product

Because the pair (V₁ ⊗ ... ⊗ Vₙ, ⊗) is, up to a unique isomorphism, the only pair with this universal property, the property can be taken as the definition of the tensor product itself, independent of any particular construction such as the quotient-of-a-free-vector-space model. Any two objects satisfying the same universal property for the same multilinear maps are canonically isomorphic, and the unique linear map furnished by each side of that isomorphism is what identifies them.

Basic Examples

Bilinear Forms

Let V, W be finite-dimensional and let f: V × W → F be a bilinear form. The universal property produces a unique linear functional f̃: V ⊗ W → F, so that bilinear forms on V × W correspond exactly to linear functionals on V ⊗ W, that is, to elements of the dual space (V ⊗ W)*.

Matrix Multiplication as a Multilinear Map

The map sending (A, B) to the matrix product AB, viewed with one factor held in Hom(V, U) and the other in Hom(U, W), is bilinear in (A, B). The associated unique linear map on Hom(V, U) ⊗ Hom(U, W) is the composition map, showing composition of linear maps to be, after tensoring, a single linear operator rather than a two-argument operation.

The Determinant on Tensor Powers

For an n-dimensional space V, the alternating multilinear map given by the determinant of n vectors factors through V ⊗ ... ⊗ V (n copies) as a unique linear functional, and further factors through the top exterior power ⋀ⁿV, illustrating how the same universal property, applied iteratively, underlies the construction of exterior and symmetric algebras as well.

Consequences in Tensor Algebra

Associativity and Reassociation Maps

The canonical isomorphisms (V₁ ⊗ V₂) ⊗ V₃ ≅ V₁ ⊗ (V₂ ⊗ V₃) and the commutation isomorphism V ⊗ W ≅ W ⊗ V are themselves constructed as the unique linear maps guaranteed by the universal property, applied to the evidently multilinear rearrangement maps on elementary tensors. Their uniqueness is what guarantees that these structural isomorphisms are canonical rather than dependent on an arbitrary choice.

Functoriality of the Tensor Product

Given linear maps φᵢ: Vᵢ → Vᵢ', the map sending (v₁, ..., vₙ) to φ₁(v₁) ⊗ ... ⊗ φₙ(vₙ) is multilinear, and the unique linear map it induces is the tensor product of linear maps φ₁ ⊗ ... ⊗ φₙ. The functoriality of the tensor product construction — its compatibility with composition and identities — follows directly from the uniqueness clause applied to composites of such maps.

Basis-Free Definition of Tensors

Because the unique linear map exists for every multilinear map regardless of any chosen basis, the universal property furnishes a basis-free characterization of tensors and of operations built from them, in contrast to definitions of tensors via arrays of components that transform according to prescribed rules under a change of basis.