2.20.2 Tensor Covector Space Context

Tensor Covector Space Context is the treatment of the dual space V* as a geometric object in its own right, focusing on how covectors, viewed collectively as the elements of a vector space, combine, compare, and admit an intrinsic geometric picture, distinct both from the abstract pairing studied in the dual vector space context and from the single-functional analysis of an individual covector. Where the linear functional context studies one covector's kernel and representation, and the dual basis context studies how to build coordinates for V*, the covector space context treats V* as a full vector space with its own subspaces, its own geometric intuition, and its own role as the model for objects such as gradients and physical quantities that are naturally covariant rather than contravariant.

V* as a Vector Space in Its Own Right

Closure Under Addition and Scalar Multiplication

The set of all linear functionals on V is closed under pointwise addition and scalar multiplication: for covectors f, g in V* and scalars α, β, the combination αf + βg, defined by (αf + βg)(v) = αf(v) + βg(v), is again a linear functional. This closure is what makes V* a vector space rather than merely a set of maps, and it means every construction available for an abstract vector space, subspaces, spanning sets, linear independence, direct sums, applies equally to collections of covectors.

Dimension and Isomorphism Type

For finite-dimensional V, dim(V*) = dim(V), so V and V* are isomorphic as abstract vector spaces, though not canonically so; any isomorphism between them depends on an arbitrary choice, such as a specific basis pairing or a metric. The covector space context treats this isomorphism as a fact about dimension alone, deliberately withholding any preferred identification, so that covectors retain their distinct transformation behavior.

Geometric Picture of the Covector Space

Covectors as Families of Hyperplanes

Rather than picturing a covector as an arrow, the covector space context favors the picture of a covector as a stack of parallel, evenly spaced hyperplanes, with the spacing inversely related to the covector's magnitude and the orientation of the planes determined by the covector's kernel. Addition of two covectors corresponds to superposing their hyperplane stacks according to a parallelogram-like rule, and scalar multiplication corresponds to compressing or expanding the spacing between planes.

Contrast With the Vector Arrow Picture

The arrow picture used for vectors in V emphasizes magnitude and direction as a displacement, while the hyperplane picture used for covectors in V* emphasizes rate of change or density: applying a covector to a vector counts how many hyperplanes the vector's arrow crosses, an integer-like count that generalizes the dot-product intuition to spaces without a metric. This distinction reflects why elements of V and V* are kept notationally and conceptually separate in the covector space context even when a metric later allows them to be numerically identified.

Subspace Structure Special to Covector Spaces

Annihilators as the Native Subspaces of V*

The natural subspaces of V* arising from subspaces of V are annihilators: for a subspace W of V, its annihilator W° = {f ∈ V* : f(w) = 0 for all w in W} is a subspace of V* of dimension dim(V) - dim(W). Annihilators give V* a subspace lattice that mirrors, in a dimension-reversing way, the subspace lattice of V, and this correspondence is order-reversing: larger subspaces of V correspond to smaller annihilators in V*.

Double Annihilator

Applying the annihilator construction twice returns the original subspace under the canonical identification V ≅ V**: the annihilator in V** of W° is exactly W. This double-annihilator identity is the covector-space analogue of double duality and confirms that the subspace structure of V* fully encodes the subspace structure of V, with no information lost in passing to the dual.

Covector Spaces as Models for Physical and Differential Quantities

Gradients as Native Elements of a Covector Space

The differential of a scalar-valued function at a point, informally its gradient, is naturally a covector rather than a vector: it acts on a displacement vector to produce the rate of change of the function in that direction, which is exactly the defining action of a linear functional. Treating gradients as elements of a covector space, rather than as vectors identified via a metric, keeps the transformation behavior of a gradient consistent under general changes of coordinates, including nonorthogonal ones.

Momentum and Other Covariant Physical Quantities

In physical contexts, quantities such as momentum, which naturally pairs with a displacement to produce an action or work, and frequency-wavevector pairs, which pair with spacetime displacement to produce a phase, are modeled as covectors rather than vectors. The covector space context is what supplies the correct transformation rule for such quantities under a change of units or coordinate system, distinguishing them from ordinary vector quantities such as velocity.

Interaction With the Tensor Product Construction

Covector Space as a Tensor Factor

V*, considered as a whole vector space, is the factor that supplies each covariant slot when a general (p, q) tensor space T^p_q(V) is built as a tensor product of p copies of V and q copies of V*. The covector space context is what justifies treating an entire copy of V* as a well-defined tensor factor, with its own basis, dimension, and subspace structure, rather than treating covariant slots merely as placeholders for individual functionals.

Symmetric and Antisymmetric Substructures

Within multiple copies of V*, the covector space context also supports the further structures of symmetric covector spaces, such as spaces of symmetric bilinear forms, and antisymmetric covector spaces, such as spaces of alternating forms used to define exterior powers. Both constructions are built by taking quotients or subspaces of tensor powers of V*, and both depend on V* having a well-defined vector space structure to begin with, which is exactly what the covector space context provides.