6.2.1 Tensor Order Classification Area
Tensor Order Classification Area explores how tensors are categorized by their order, defining their structure and mathematical properties in algebraic contexts.
Tensor Order Classification Area is the branch of tensor classification that names and organizes tensors according to the total number of indices required to write one of their components, treating this count, denoted k, as the single defining criterion and assigning conventional names — scalar, vector, matrix-like, and higher-order — to successive values of k. This area answers only the question "how many indices does this tensor have," leaving the further questions of variance, arrangement, and concrete representation to the other classification areas.
The Defining Criterion
Order as an Index Count
For a tensor T whose components are written T_{i₁...i_p}^{j₁...j_q}, the order is the total number of indices appearing, upper and lower combined:
No other property of the tensor — not the dimension of the underlying vector space, not the numerical values of the components, not whether indices are upper or lower — enters into the determination of k.
Order as the Number of Multilinear Arguments
Equivalently, viewing T as a multilinear map T : V₁ × ... × V_k → F, the order is the number of vector-space arguments the map accepts, a characterization that agrees with the index-counting definition once each argument is paired with its corresponding index.
The Named Orders
Order Zero: Scalars
A tensor of order 0 is a scalar, an element of the field F with no indices at all; it requires no basis to be specified and is invariant, by definition, under every change of basis.
Order One: Vectors and Covectors
A tensor of order 1 is a vector (if its single index is contravariant) or a covector (if its single index is covariant); both are addressed by exactly one index and organized as a one-dimensional list of components once a basis is fixed.
Order Two: Matrix-Like Tensors
A tensor of order 2 has two indices and can be organized, once a basis is chosen, as a two-dimensional array or matrix; this is the order at which the array representation first takes the familiar rows-and-columns form.
Order Three and Above: Higher-Order Tensors
is classified simply as a higher-order tensor, referred to by stating k explicitly (an order-3 tensor, an order-4 tensor, and so forth), since no further special names are conventionally attached beyond this point.
Diagram of the Named Orders
How Order Behaves Under Tensor Operations
Addition Under the Tensor Product
follows directly from the index-counting definition, since forming a tensor product concatenates the index lists of the two factors.
Reduction Under Contraction
follows because contraction removes exactly one upper and one lower index, lowering the total count by two.
Exponential Growth Under Repeated Tensoring
Because order adds under the tensor product, the order of an n-fold tensor power V^{⊗n} grows only linearly in n, even though the dimension of the resulting space grows exponentially, a contrast that highlights order as a count of indices rather than a measure of the size of the space.
Distinguishing Order From Related Notions
Order Versus Dimension of the Underlying Space
A 3 × 3 matrix and a 100 × 100 matrix are both order-2 tensors; order says nothing about the dimension n of the vector space supplying each index, only about how many indices there are.
Order Versus Type
Order gives only the total k = p + q; it does not reveal how many of the k indices are upper versus lower, a finer distinction reserved for type classification, so a single order such as k = 2 covers the distinct types (2,0), (1,1), and (0,2).
Order Versus Rank of a Matrix
The word "rank," when used for the linear-algebraic invariant equal to the dimension of a matrix's column space, is unrelated to order: a 4 × 4 matrix of numerical rank 1 and one of numerical rank 4 are both, without distinction, order-2 tensors.
Why Order Classification Is Useful
A Minimal, Basis-Independent First Descriptor
Order requires no choice of basis and no knowledge of variance to compute, making it the fastest and most basic descriptor available for any tensor, and the natural first fact stated about a tensor before its type or valence is specified.
A Coarse Filter for Compatible Operations
Many tensor operations, such as elementwise addition, require operands of matching order, so checking order compatibility is typically the first and cheapest test performed before attempting a tensor operation, ahead of any check on type or dimension.