1.11.1 Tensor Algebra Convention

Tensor Algebra Convention is the practical body of field-specific and text-specific agreements that fix, for a given community or a given document, which of the several possible choices identified at the foundational level, index range, ordering, sign, basis type, are actually adopted, so that tensor expressions can be written, read, and combined without ambiguity within that context. Where convention foundations catalog the individual points of arbitrary choice, tensor algebra convention is concerned with how those points are resolved in practice, how differing resolutions cluster into recognizable, named systems, and how a reader moves between one such system and another.

Because no single resolution of every foundational choice is universally adopted, distinct subfields of mathematics, physics, and engineering have converged on their own internally consistent, but mutually distinct, packages of conventions. Recognizing which package is in force in a given text, and translating a result from one package into another, is a core practical skill required to use tensor algebra across sources.

Convention as a Package, Not a Single Rule

Consistency Across Many Simultaneous Choices

A workable convention is not one isolated decision but a coordinated set of decisions, an index range, an index-ordering rule, a sign convention for antisymmetrization, a choice of orthonormal versus general basis, that must all cohere with one another. Changing a single element of the package, such as the index range, without correspondingly adjusting related formulas, produces a system that is no longer internally consistent, which is why conventions are best understood, and best adopted, as complete packages rather than as separable options to be mixed freely.

Named Convention Packages

Certain convention packages are common enough to carry recognized names within a field. In relativistic physics, for instance, the "mostly plus" and "mostly minus" metric signature conventions each fix, in a single choice, the sign pattern of every diagonal metric component, and this single choice then propagates through the sign of every formula involving the metric, raising and lowering, invariant intervals, and related quantities. Adopting a named convention package signals, in one statement, an entire coordinated set of downstream sign and ordering choices.

Convention Differences Across Fields

Mathematics and Abstract Algebra

In purely mathematical treatments of multilinear algebra, conventions tend to favor abstraction and generality: index ranges are often left as an unspecified 1 through n, orthonormal bases are not assumed by default, and the tensor product is treated as a formal, associative operation defined independently of any concrete numbering scheme. This convention package prioritizes statements that hold for an arbitrary vector space over any field, at the cost of being less directly tied to specific numerical computation.

Continuum Mechanics and Engineering

In continuum mechanics, conventions often favor orthonormal Cartesian bases as the default setting, which collapses the upper-lower index distinction and allows all indices to be written as subscripts without loss of information, alongside a strong preference for explicit, low-rank notation, vectors and second-order tensors such as stress and strain, since most quantities of engineering interest do not exceed rank two.

Computational and Machine Learning Contexts

In computational contexts, including machine learning frameworks that manipulate multidimensional arrays under the name "tensor," the convention package typically adopts zero-based index ranges, matching the underlying array-indexing behavior of the programming languages involved, and frequently drops the upper-lower index distinction entirely, since these frameworks generally do not track covariant and contravariant transformation behavior at all, using "tensor" instead simply to mean a multidimensional array.

Translating Between Convention Packages

Recognizing Which Package Is in Force

Before reconciling a formula from one source with a formula from another, the specific convention package each source uses must first be identified: the index range, whether the metric signature is fixed, whether a basis is assumed orthonormal, and how antisymmetric quantities are signed. Many texts state these choices explicitly in an early section, and locating this statement is the first step in any cross-source comparison.

Systematic Translation

Once the conventions in force are known, translating a formula from one package to another proceeds by systematically substituting the differing choices: shifting an index range by adding or subtracting one from every index bound, flipping the sign of terms that depend on the metric signature when moving between signature conventions, or reintroducing the upper-lower distinction when moving from a computational, variance-free convention into one that tracks covariance and contravariance explicitly.

Risks of Silent Mismatches

Because the visual form of an equation often looks identical regardless of which convention package produced it, errors arising from an unstated or overlooked mismatch, comparing a mostly-plus formula against a mostly-minus formula without adjusting signs, for instance, are especially difficult to detect from inspection alone, and typically surface only once a computed numerical result disagrees with an expected value or with an independently verified case.

Establishing Convention Within a Single Work

Stating Conventions Explicitly

Careful treatments of tensor algebra state their adopted conventions explicitly at the outset, specifying the index range, the ordering of mixed indices, the sign convention for the Levi-Civita symbol, and the assumed basis type, precisely so that every subsequent formula in the work can be interpreted without ambiguity and compared reliably against formulas from other sources once the stated conventions are accounted for.

Internal Consistency as the Governing Requirement

Whatever specific convention package a given piece of work adopts, the overriding requirement is internal consistency: every formula, derivation, and numerical example within that work must adhere to the same package throughout, since tensor algebra's notation, while compact and powerful, provides few automatic safeguards against a convention that has silently shifted partway through a derivation.