3.13.5 Tensor Canonical Embedding Basis Independence

Tensor Canonical Embedding Basis Independence is the property that the canonical map ev: V → V** is defined without reference to any chosen basis of V, V*, or V**, and produces the same result regardless of what basis one might use to compute or verify it. This basis independence is the defining feature that separates the canonical embedding from other maps relating a vector space to a dual-type space, and it is the reason the term "canonical," or equivalently "natural," is attached to ev.

What Basis Independence Means

The Contrast with Basis-Dependent Isomorphisms

A finite-dimensional vector space V is isomorphic to its dual V*, but every such isomorphism requires first fixing a basis e_1, ..., e_n of V and then declaring the dual basis e^1, ..., e^n to be the image of e_1, ..., e_n under the isomorphism. Choosing a different basis of V produces a different isomorphism to V*, with no distinguished choice among them being more correct than another. Basis independence is the absence of this phenomenon: the canonical embedding ev is the same map no matter which basis, if any, is used to describe or compute it.

The Definition Makes No Basis Reference

The defining formula of the embedding:

mentions only the vector v, the covector φ, and the operation of applying φ to v. No basis of V or V* appears anywhere in this formula. This textual absence of any basis is the most direct evidence of the map's basis independence: the formula could be written down and understood before any basis of V is ever chosen.

Verifying Independence Under Change of Basis

Setting Up Two Bases

Suppose e_1, ..., e_n and f_1, ..., f_n are two different bases of V, related by a change-of-basis matrix A, so that f_i = A^j_i e_j. Each basis induces its own dual basis, e^1, ..., e^n and f^1, ..., f^n, of V*.

The Assigned Functional Does Not Depend on the Basis Used

For a fixed vector v, the functional ev(v) is computed by evaluating an arbitrary covector φ at v. This computation, ev(v)(φ) = φ(v), never asks which basis φ happens to be expressed in. Whether φ is written in terms of e^1, ..., e^n or f^1, ..., f^n, the number φ(v) produced is the same value, since φ and v are fixed, basis-independent objects and evaluation is an operation defined between them directly, not through coordinates.

Contrast with a Coordinate Computation

If one instead computes ev(v) using coordinates, writing v = v^i e_i and evaluating against the dual basis e^1, ..., e^n, a numerical list of components is produced. Switching to the basis f_1, ..., f_n produces a different numerical list of components, related to the first by the transformation law for tensors. The two lists of numbers differ, yet both describe the same underlying basis-independent functional ev(v), illustrating that basis independence refers to the invariance of the abstract object, not to the invariance of its coordinate representation.

Why Basis Independence Matters for Tensor Algebra

Compatibility with the Tensor Transformation Law

Tensors are defined as objects whose components transform in a specific way under a change of basis, precisely so that the underlying geometric object remains well defined regardless of the basis used to describe it. If the identification of V with V** depended on a choice of basis, then treating vectors as double-dual functionals inside tensor expressions would risk introducing spurious basis-dependence into what should be a basis-independent quantity. Because ev is basis-independent, this risk does not arise: identifying v with ev(v) is always safe, no matter what basis is subsequently chosen to compute with.

Naturality as a Formal Concept

Basis independence is the concrete manifestation of a more general notion called naturality, used to describe maps that are defined uniformly across an entire category of spaces, in the sense that they commute with all the natural transformations between objects in that category, such as linear maps between vector spaces. The canonical embedding is a natural transformation from the identity functor to the double-dual functor, and its basis independence is precisely the naturality condition made explicit at the level of vector spaces and linear maps.

Robustness Across Applications

Because ev does not depend on a basis, results derived using it remain valid after any change of coordinates. This robustness is what allows the reflexive identification V ≅ V** to be invoked freely inside coordinate-independent arguments in tensor algebra, differential geometry, and functional analysis, without needing to track or justify a particular coordinate system at each step.

Basis-Dependent Isomorphisms as a Point of Comparison

The V to V* Isomorphism Requires a Choice

The isomorphism between V and V* obtained by matching a basis e_1, ..., e_n to its dual basis e^1, ..., e^n sends e_i to e^i. If a different basis f_1, ..., f_n is used instead, the resulting isomorphism sends f_i to f^i, which is generally a different linear map from V to V* than the one built from e_1, ..., e_n, even though both are valid isomorphisms. This dependence on the chosen basis is exactly what the canonical double-dual embedding avoids.

Only One Canonical Choice Exists Between V and V**

Between V and V**, by contrast, there is exactly one map that arises without any auxiliary choice, namely ev. Other isomorphisms between V and V** may exist as well in the finite-dimensional case, since any two isomorphic finite-dimensional spaces admit many isomorphisms between them, but only ev qualifies as canonical, because it alone is defined without reference to a basis.

Diagrammatic Summary

The diagram indicates that regardless of whether the basis labeled e or the basis labeled f is used to think about V, the single arrow representing ev reaching V** remains identical, illustrating that the canonical embedding is insensitive to the choice of basis used to analyze it.