6.2 Tensor Classification Areas

Tensor Classification Areas is the collection of distinct, complementary schemes by which a tensor is described and distinguished from other tensors, spanning the count of its argument slots, the split of those slots into contravariant and covariant positions, the specific ordered arrangement of that split, the concrete layout of its numerical components once a basis is chosen, and the structural properties — such as symmetry — that a tensor may or may not possess independently of any of the preceding schemes. Each area answers a different question about a tensor, and a complete description of any given tensor draws on several of these areas at once rather than any single one alone.

Slot Counting as the Foundational Area

Counting Arguments Before Naming Them

The most basic classification area is the plain count of how many arguments a tensor, viewed as a multilinear map, takes. This count is pure arithmetic: it starts from 0 for a scalar and 1 for a vector or covector, and combines additively under the tensor product and by subtracting two at a time under contraction.

Why Counting Precedes Every Other Area

Every other classification area presupposes a slot count already exists to work with: order attaches a name to it, type splits it, valence arranges the split, and component organization lays out an array whose number of dimensions equals it.

Order Classification

Naming a Tensor by Its Total Slot Count

Order classification takes the raw slot count and assigns it a conventional name: 0 for scalar, 1 for vector or covector, 2 for a matrix-like tensor, and simply "order k" for any k ≥ 3, with no further naming refinement beyond stating the number directly.

The Role of Order in Everyday Usage

Order is the classification area most commonly invoked in informal descriptions of a tensor, since it requires no knowledge of variance or arrangement, only the total number of indices visible in a component expression such as T_{ijk}.

Type Classification

Splitting the Slot Count by Variance

Type classification refines order by recording, of the total slot count k, how many slots p are contravariant (upper) and how many q are covariant (lower), written as the pair (p, q) with p + q = k.

Transformation Behavior Attached to Type

Type classification carries operational weight because upper and lower slots transform differently under a change of basis, upper slots by the inverse Jacobian and lower slots by the Jacobian directly, so knowing a tensor's type is necessary to know how its components change when the basis changes.

Valence Classification

Arranging the Split Into an Explicit Sequence

Valence classification refines type further by recording the specific left-to-right order in which upper and lower slots appear, rather than only their totals; two tensors can share type (p, q) while differing in valence whenever their upper and lower slots are interleaved differently.

Diagram Contrasting the Three Refining Areas

Component Organization

From Abstract Classification to Concrete Arrays

Once order, type, and valence are fixed, component organization addresses the separate, representational question of how a tensor's numerical components are laid out in a multi-dimensional array after a basis has been chosen for each slot, including index-ordering conventions and the formula for addressing a specific component.

Why Organization Is Basis-Dependent While the Other Areas Are Not

Slot count, type, and valence are properties of a tensor that hold regardless of which basis is used to describe it, whereas component organization only exists relative to a chosen basis, marking it as a fundamentally different kind of classification area from the three that precede it.

Symmetry and Structural Properties

Properties Layered on Top of a Fixed Classification

Beyond order, type, valence, and organization, a tensor may additionally be classified by structural properties such as symmetry or antisymmetry under exchange of same-variance slots, properties that are only meaningful once a specific valence has already fixed which slots share a variance and are therefore eligible to be exchanged.

Independence From the Numerical Rank of a Representation

A separate structural notion, the linear-algebraic rank of a matrix obtained by representing a type (1,1) tensor in a basis, is independent of order, type, and valence: tensors sharing all three can still differ in this numerical rank, marking it as its own classification area rather than a refinement of the others.

How the Classification Areas Relate

A Layered, Non-Redundant Hierarchy

The classification areas form a layered hierarchy in which each later area adds information the earlier ones lack — order gives a count, type splits the count, valence arranges the split, organization lays out concrete components, and structural properties describe patterns among the values — with no area duplicating information already captured by another.

Combining Areas for a Complete Description

A full description of a specific tensor typically states its order, its type, and, when relevant, its valence and symmetry properties together, since no single classification area by itself is sufficient to pin down everything that distinguishes one tensor from another.