2.21 Tensor Vector Space Isomorphism Structure

Tensor Vector Space Isomorphism Structure is the framework that classifies and organizes the different kinds of isomorphisms that relate a vector space to other vector spaces built from it, distinguishing sharply between isomorphisms that exist canonically, independent of any choice, and isomorphisms that exist only after an arbitrary choice such as a basis or a metric is fixed, a distinction that governs which identifications are legitimate to use when defining and manipulating tensors. Because tensor algebra constantly produces new vector spaces from old ones, through duals, tensor products, and spaces of multilinear maps, keeping track of which resulting spaces are "the same" only up to an arbitrary choice, and which are identified naturally, is essential to avoiding basis-dependent errors in tensor reasoning.

Isomorphism as an Equivalence of Vector Spaces

Definition

An isomorphism between vector spaces V and W over the same field F is a bijective linear map φ : V → W. Two vector spaces are isomorphic exactly when they have the same dimension, since a linear bijection can always be constructed between spaces of equal finite dimension by matching bases, and no linear bijection can exist between spaces of different dimension.

Why Isomorphism Alone Is Too Coarse

Because dimension is the only invariant distinguishing finite-dimensional vector spaces up to isomorphism, mere isomorphism does not preserve any of the extra structure, such as which vectors correspond to which, that tensor algebra depends on. Two spaces being isomorphic says only that an identification exists; it says nothing about whether that identification is unique, natural, or safe to use inside a formula that must hold for every choice of basis.

Canonical Isomorphisms

The Double Dual Isomorphism

The single most important canonical isomorphism in tensor algebra identifies a finite-dimensional space V with its double dual V**, via the evaluation map ev : V → V**, ev(v)(f) = f(v). This map is linear, injective, and, in finite dimension, surjective, and crucially it is defined without reference to any basis:

Because ev requires no auxiliary choice, V and V** are said to be canonically or naturally isomorphic, and this isomorphism is used freely throughout tensor algebra, for instance to reinterpret (p, q)-tensors as multilinear maps of a particular argument type.

Naturality in the Categorical Sense

The double dual isomorphism has the further property of naturality: it commutes with every linear map, meaning that for any linear map T : V → W, the diagram built from ev_V, ev_W, T, and the double pullback T** commutes exactly. This compatibility with all morphisms, not merely its basis-independence, is what makes the isomorphism canonical in the technical, categorical sense used in tensor algebra.

Non-Canonical Isomorphisms

The Isomorphism V ≅ V*

Although dim(V) = dim(V*) guarantees V and V* are isomorphic, no natural choice of isomorphism between them exists in general: any explicit isomorphism V → V* requires additional data, most commonly a chosen basis, which induces the isomorphism sending each basis vector to the corresponding dual basis covector, or a chosen nondegenerate bilinear form such as a metric or symplectic form, which induces the isomorphism v ↦ ⟨v, ·⟩.

Why the Distinction Matters for Tensor Formulas

Formulas or constructions in tensor algebra that silently use an isomorphism V ≅ V* implicitly commit to a specific choice, such as a metric, and will produce different numerical results, or even different qualitative behavior, if that choice is changed. Formulas that use only the canonical isomorphism V ≅ V** remain valid regardless of any basis or metric choice. The isomorphism structure framework is precisely the discipline of tagging every identification used in a tensor construction as either canonical or choice-dependent, so that the basis-independence or basis-dependence of the final result is always traceable.

The Automorphism Group and Its Action on Tensor Spaces

GL(V) as the Group of Self-Isomorphisms

The set of isomorphisms from V to itself forms a group under composition, the general linear group GL(V), consisting of all invertible linear transformations of V. Every change of basis of V corresponds to an element of GL(V), and the transformation laws governing tensor components are precisely the description of how the action of GL(V) on V propagates to an action on T^p_q(V).

Induced Action on Tensor Product Spaces

An isomorphism φ in GL(V) induces, functorially, an isomorphism on every tensor space built from V, including V ⊗ V, V*, and general T^p_q(V), by acting on each factor: on p copies of V directly via φ, and on q copies of V* via the inverse-transpose (φ^{-1})^*. This functorial extension is what guarantees that an object satisfying the defining transformation law of a (p, q)-tensor under one change of basis automatically satisfies it under every composition of changes of basis, since the induced action respects composition, (φ ∘ ψ)_* = φ_* ∘ ψ_*.

Isomorphism Invariants as Tensor Invariants

Basis-Independent Quantities

Any quantity computed from a tensor that is unchanged by every isomorphism induced from a change of basis, such as the trace of a (1,1)-tensor, the determinant of a linear operator represented as a (1,1)-tensor, or the rank of a tensor in the sense of the minimal number of simple tensor summands needed to express it, is a genuine invariant of the tensor, independent of coordinates. The isomorphism structure framework is what makes precise the statement that a quantity is "basis-independent": it is a quantity fixed by the entire induced action of GL(V), not merely unchanged under some particular sample change of basis.

Distinguishing True Invariants From Coincidental Agreement

A quantity that happens to agree across two specific bases is not thereby shown to be an invariant; only agreement across the full orbit of the GL(V)-action, or equivalently, a proof using only canonical isomorphisms, establishes true basis independence. This is why the vector space isomorphism structure, with its careful separation of canonical from non-canonical identifications, underlies the rigorous justification of every claim in tensor algebra that a particular construction is coordinate-free.