6.21.5 Tensor Matrix Tensor Role

Exploring how tensor matrices represent and manipulate multidimensional data in algebraic structures.

Tensor Matrix Tensor Role is the recognition that, within the classification of second-order tensors, a matrix is not itself a tensor but a coordinate representation of one, and that its right to be called "tensor-like" depends entirely on how its entries transform when the underlying basis is changed. This role clarifies a common source of confusion: not every rectangular array of numbers arranged in rows and columns is a tensor in the technical sense, and the tensor matrix relationship is precisely the added structure — a specified transformation law tied to a specified variance type — that promotes a bare array into a genuine second-order tensor.

Matrix as Container, Tensor as Content

The Array Alone Is Not Sufficient

A matrix, in the purely computational sense, is just a rectangular table of numbers with no attached rule for how those numbers should change if a different basis were used. Many useful matrices in applied mathematics — adjacency matrices of graphs, matrices of raw data, transition matrices of certain combinatorial processes — are not tensors, because there is no vector space and change-of-basis action for which their entries transform correctly.

The Transformation Law Supplies the Tensor Role

A matrix earns the tensor role only when its entries Tⱼⁱ are declared to be the coefficients of an object T in V* ⊗ W (or another appropriate tensor product space) relative to chosen bases, and are required to obey the induced transformation rule under a change of those bases:

T_{j}^{i} \to \sum_{k, l} A_{k}^{i} C_{j}^{l} T_{l}^{k}

where A is the change-of-basis matrix acting on the contravariant index and C is the inverse change-of-basis matrix acting on the covariant index.

Only once this rule is imposed does "the matrix" become "the coordinate representative of the tensor T," and the two terms — matrix and tensor — cease to be interchangeable labels for the same static array and instead describe different levels of the same object: the numerical table versus the basis-independent entity it represents.

Diagram of the Relationship

Consequences of Taking On the Tensor Role

Determined Behavior Under Basis Change

Once a matrix is recognized as playing the tensor role, its behavior under any change of basis is completely determined by its type (p, q): there is no freedom to choose how the entries update, since the transformation law is fixed by the requirement that the abstract object T remain unchanged while its coordinate description adapts to the new basis.

Restriction on Legitimate Operations

The tensor role also restricts which operations on the matrix are meaningful as tensor operations. Adding two matrices entrywise is a legitimate tensor operation only if both matrices represent tensors of the same type relative to the same pair of bases; multiplying matrices is a legitimate tensor operation (contraction) only across a contravariant-covariant index pair. A matrix that has not been assigned the tensor role has no such restrictions, since arbitrary matrix arithmetic is always well-defined for arrays of the right shape — it is only meaningful as tensor arithmetic once the tensor role has been assigned.

When the Roles Diverge

Matrices That Are Not Tensors

The classic example of a matrix without a tensor role is a matrix representing a bilinear form evaluated at fixed, non-transforming inputs, or a table of statistical correlations, where there is no vector space and change-of-basis convention under which the transformation law would hold. Treating such a matrix as though it obeyed tensor transformation rules produces incorrect results the moment a change of basis or coordinate system is introduced.

Tensors That Are Not Naturally Matrices

Conversely, tensors of order greater than two — order three or higher — cannot be displayed as an ordinary two-dimensional matrix at all, even though each of their two-index "slices" can be. The tensor matrix relationship is therefore specific to order two: it is the unique order at which the tensor's full content fits exactly into the row-and-column format of a matrix, which is why second-order tensors occupy a privileged position bridging tensor algebra and matrix algebra.