1 Basic idea

1.1 Motivation

The tensor product provides a systematic way to combine two vector spaces (or more general algebraic objects) so that expressions built from elements behave bilinearly. A typical motivation arises when one wants to turn a bilinear operation \(B: V\times W \to U\) into a linear operation on a larger space built from \(V\) and \(W\). The tensor product achieves this by introducing new “combined” elements while enforcing the relations needed for bilinearity.

1.2 Universal property

Given vector spaces \(V\) and \(W\) over a field \(\Bbb{k}\), a tensor product \(V\otimes W\) is characterized (up to unique isomorphism) by the following property: there exists a bilinear map \[ \otimes: V\times W \to V\otimes W,\quad (v,w)\mapsto v\otimes w, \] such that for any vector space \(U\) and any bilinear map \(B: V\times W\to U\), there is a unique linear map \(\tilde B: V\otimes W \to U\) with \[ \tilde B(v\otimes w)=B(v,w). \] This property both defines the construction conceptually and guarantees that tensor products are the “most general” objects supporting bilinear-to-linear conversion.

1.3 Bilinear maps

Bilinearity means linearity in each argument separately: \[ B(\alpha v_1+\beta v_2,w)=\alpha B(v_1,w)+\beta B(v_2,w),\qquad B(v,\alpha w_1+\beta w_2)=\alpha B(v,w_1)+\beta B(v,w_2). \] The tensor product is designed precisely so that any such \(B\) depends only on the induced linear map from \(V\otimes W\).

1.3.1 Factorization through the tensor product

The universal property can be read as a factorization statement: bilinear maps from \(V\times W\) to \(U\) are equivalent to linear maps from \(V\otimes W\) to \(U\). In practice, to build \(\tilde B\), one specifies its values on pure tensors \(v\otimes w\); linearity then extends \(\tilde B\) to all of \(V\otimes W\).

2 Tensor products of vector spaces

2.1 Definition

For vector spaces \(V\) and \(W\) over \(\Bbb{k}\), the tensor product \(V\otimes W\) is the vector space equipped with a bilinear map \(V\times W\to V\otimes W\) that satisfies the universal property described above. Elements of \(V\otimes W\) are finite linear combinations of pure tensors \(v\otimes w\).

2.2 Construction

2.2.1 Free vector space approach

One common explicit construction starts from the free vector space generated by symbols \((v,w)\) with \(v\in V\), \(w\in W\). This space is then quotiented by relations enforcing bilinearity:

  • \((v_1+v_2,w)=(v_1,w)+(v_2,w)\),
  • \((\alpha v,w)=\alpha (v,w)\),
  • \((v,w_1+w_2)=(v,w_1)+(v,w_2)\),
  • \((v,\alpha w)=\alpha (v,w)\).

The equivalence classes of \((v,w)\) become the pure tensors \(v\otimes w\).

2.2.2 Quotient space approach

The quotient formulation can be described succinctly as \[ V\otimes W \;=\; \Bbb{k}[V\times W] / R, \] where \(\Bbb{k}[V\times W]\) denotes the free vector space on the set \(V\times W\) and \(R\) is the subspace generated by the bilinearity relations listed above. This presentation makes it explicit that the tensor product is obtained by imposing exactly the linearity constraints needed—and no more.

2.3 Basis and dimension

2.3.1 Tensor products of finite-dimensional spaces

If \(\{e_i\}_{i=1}^m\) is a basis of \(V\) and \(\{f_j\}_{j=1}^n\) is a basis of \(W\), then \(\{e_i\otimes f_j\}_{i,j}\) is a basis of \(V\otimes W\). Consequently, \[ \dim(V\otimes W)=mn. \] This fact is often used for calculations and for understanding the size of spaces appearing in applications.

2.4 Pure tensors and general tensors

A pure tensor is a single element of the form \(v\otimes w\). In contrast, a general tensor is a finite linear combination of pure tensors: \[ T=\sum_{k=1}^r v_k\otimes w_k. \] Not every tensor is pure; deciding whether a given tensor is decomposable into one such product depends on additional structure. Nonetheless, the span of pure tensors is the whole tensor product, so computations reduce to manipulating linear combinations of these building blocks.

3 Tensor products of modules

3.1 Definition over a ring

For modules over a commutative ring \(R\), the tensor product \(M\otimes_R N\) is defined so that it is universal with respect to \(R\)-bilinear maps \(M\times N\to A\) for \(R\)-modules \(A\). The construction parallels the vector space case but uses \(R\)-module linearity rather than field linearity.

3.2 Balancing relations

The tensor product over \(R\) incorporates the “balancing” condition that scalar multiplication can be transferred between factors: \[ (rm)\otimes n = m\otimes (rn), \] for \(r\in R\), \(m\in M\), \(n\in N\). These relations are imposed in the quotient construction and are essential for making \(M\otimes_R N\) compatible with module structures.

3.3 Exactness properties

3.3.1 Right exactness

Tensoring with a fixed module is typically right exact. Concretely, if \[ M'\to M\to M''\to 0 \] is exact, then tensoring on the left with \(N\) yields an exact sequence \[ M'\otimes_R N \to M\otimes_R N \to M''\otimes_R N \to 0. \] This reflects how tensor products preserve surjections but may fail to preserve injections.

3.3.2 Failure of left exactness

Tensor products generally do not preserve kernels. In other words, an exact sequence \[ 0\to M'\to M \] may become non-exact after tensoring, meaning the map \(M'\otimes_R N \to M\otimes_R N\) might not be injective. This phenomenon motivates derived functors such as \(\mathrm{Tor}\).

3.4 Tensor product with scalar extension

When rings are related by a ring homomorphism \(R\to S\), scalar extension can be expressed using tensor products. For instance, if \(M\) is an \(R\)-module, then \[ M\otimes_R S \] is an \(S\)-module that captures \(M\) after enlarging scalars. This approach is common in algebra and geometry, where one changes coefficient rings to simplify problems.

4 Tensor products of algebras

4.1 Algebra structure on tensor products

If \(A\) and \(B\) are algebras over a commutative ring \(\Bbb{k}\), their tensor product \(A\otimes_\Bbb{k} B\) can be given an algebra structure. Multiplication is defined on pure tensors by \[ (a\otimes b)(a'\otimes b')=(aa')\otimes (bb'), \] and extended bilinearly. The result is an algebra whose unit is \(1_A\otimes 1_B\) when units exist.

4.2 Tensor product of associative algebras

For associative \(\Bbb{k}\)-algebras \(A\) and \(B\), the above multiplication is associative as well, so \(A\otimes_\Bbb{k} B\) becomes an associative algebra. This construction is widely used, for example, to model composite systems where operations on one component and another component act independently.

4.3 Tensor products of modules over algebras

More generally, if \(A\) is an algebra and \(M,N\) are \(A\)-modules, one can form tensor products over \(A\) (or over the ground ring) depending on the desired compatibility. When modules are taken over an algebra, balancing relations reflect the module actions of that algebra, affecting how morphisms and decompositions behave.

5 Multilinear algebra

5.1 Higher-order tensor products

Higher-order tensor products are formed by iterating the two-factor construction. For vector spaces \(V_1,\dots,V_k\), one defines \[ V_1\otimes \cdots \otimes V_k \] as a space representing \(k\)-linear maps into a target via an analogous universal property. This framework organizes multilinear expressions such as those appearing in multilinear forms, differential geometry, and multilinear algebraic identities.

5.2 Tensor powers

For a single vector space \(V\), the tensor power \(V^{\otimes k}\) is shorthand for \(V\otimes \cdots \otimes V\) with \(k\) factors. Tensor powers serve as a base for constructing symmetrized and antisymmetrized objects.

5.2.1 Symmetric powers

The symmetric power \(S^k(V)\) is obtained by factoring \(V^{\otimes k}\) by the action that symmetrizes tensors. Equivalently, one imposes relations corresponding to invariance under permutation of tensor factors. Symmetric tensors represent polynomial-like data; for example, they appear in the study of homogeneous polynomials and in invariant theory.

5.2.2 Exterior powers

The exterior power \(\wedge^k V\) is formed by imposing antisymmetry under swapping factors. In the exterior algebra, tensors change sign upon permutation according to the parity of that permutation. Exterior powers are used to encode oriented volume elements and to express wedge products in differential forms.

5.3 Multilinear forms

A multilinear form is a map \[ T: V_1\times \cdots \times V_k \to \Bbb{k} \] that is linear in each argument. Such a form corresponds to a linear functional on the tensor product: \[ T \leftrightarrow \tilde T \in (V_1\otimes \cdots \otimes V_k)^*. \] This identification is central for converting coordinate-based multilinear expressions into tensor objects with intrinsic transformation behavior.

6 Coordinates and computations

6.1 Index notation

In computations, tensors are often represented with indices. If \(V\) has a basis \(\{e_i\}\) and \(W\) has a basis \(\{f_j\}\), then a tensor \(T\in V\otimes W\) can be written as \[ T=\sum_{i,j} T^{ij} e_i\otimes f_j, \] where the coefficients \(T^{ij}\) are the components of \(T\) relative to chosen bases. Changing bases changes these components according to transformation rules determined by the tensor product structure.

6.2 Kronecker product relation

When vector spaces are realized as coordinate spaces, the tensor product closely relates to the Kronecker product of matrices and vectors. In finite-dimensional settings, if \(u\) and \(v\) are column vectors, then \(u\otimes v\) corresponds to a Kronecker product array built from their entries. This relationship is often used in numerical linear algebra, though it is important to remember that the tensor product is a basis-independent concept.

6.3 Matrix representations

6.3.1 Transformation rules

Given a change of basis in \(V\) and \(W\), tensor components transform compatibly with the tensor product. For example, if \(e_i'=\sum_a S_{ai} e_a\) and \(f_j'=\sum_b T_{bj} f_b\), then the components \(T^{ij}\) relative to the primed bases are related to the original components by multiplication with \(S\) and \(T\) in a way dictated by how \(v\otimes w\) transforms. These rules ensure that tensor equalities remain coordinate-free even when expressed using indices.

7 Universal properties and categorical viewpoint

7.1 Tensor product as a representing object

From a categorical perspective, the tensor product can be described as a representing object for a certain functor. The functor \[ U \mapsto \{\text{bilinear maps } V\times W\to U\} \] is naturally isomorphic to \[ U \mapsto \mathrm{Hom}(V\otimes W, U), \] expressing that \(V\otimes W\) “represents” bilinear maps by linear maps. This viewpoint clarifies why the tensor product is canonical up to unique isomorphism.

7.2 Monoidal categories

In many settings, tensor products form the tensor (monoidal) structure of a category: there is a bifunctor \( \otimes \), associativity constraints, and a unit object. While additional coherence data is needed in general, the guiding idea is that objects can be combined and that the process is associative up to specified natural isomorphisms.

7.3 Functoriality

The tensor product is functorial in each variable. If \(f: V\to V'\) and \(g: W\to W'\) are linear maps, then there is an induced linear map \[ f\otimes g: V\otimes W \to V'\otimes W' \] sending \(v\otimes w\) to \(f(v)\otimes g(w)\). This allows one to propagate linear maps through tensor constructions systematically, which is essential in homological algebra and representation theory.

8 Applications in applied mathematics

8.1 Finite element methods

In finite element methods, tensor product spaces help build multidimensional approximation spaces from one-dimensional components. This is especially convenient on rectangular or box-shaped domains, where separable basis functions yield efficient assembly of stiffness matrices and enable structured discretizations.

8.2 Differential equations

Tensor products appear in weak formulations of partial differential equations. Spaces of test and trial functions often involve tensor products when combining variables (e.g., space-time separations, or product domains). Moreover, coefficients in PDEs can be handled using tensor structures to express anisotropy and coupling between components.

8.3 Signal processing

In signal processing, tensor decompositions and tensor products provide algebraic tools for multilinear data such as images, videos, and multi-sensor measurements. While the raw tensor product describes how to combine spaces, practical algorithms often use associated constructions (e.g., decompositions into sums of simpler tensors) to reduce storage and extract structure.

8.4 Mechanics and continuum models

Continuum mechanics uses tensor algebra to represent stress, strain, and constitutive laws. Quantities like stress are modeled as second-order tensors, and constitutive relations use tensor product operations to map between different tensor spaces. This framework supports coordinate transformations and distinguishes physical invariants from components.

8.5 Quantum mechanics and state spaces

In quantum theory, the tensor product models composite systems: the state space of a composite of two subsystems is constructed as the tensor product of their individual state spaces. Observables and operations on composite systems are then expressed through tensor products of operators, enabling the formal description of interactions and measurement on combined systems.

9.1 Tensor fields

A tensor field assigns a tensor to every point of a manifold, typically in a way compatible with changes of coordinates. Such fields generalize the notion of vector-valued functions and are central in differential geometry and physics, where geometric objects vary across space.

9.2 Tensor bundles

Tensor fields are formalized as sections of vector bundles whose fibers are tensor products of tangent and cotangent spaces (or related spaces). A tensor bundle packages how tensors transform under coordinate changes by encoding them in the bundle structure.

9.3 Tensor contraction

Contraction is an operation that reduces tensor rank by pairing a covariant index with a contravariant index (or, more generally, using a bilinear pairing). For instance, contracting adjacent indices corresponds to applying evaluation maps between a space and its dual. Contractions are essential for defining traces, divergences, and inner products.

9.4 Inner and outer products

Given an inner product or a pairing between spaces, one can define derived tensor operations. The outer product of vectors produces a tensor whose components encode pairwise combinations. Inner-product-related constructions often correspond to contracting indices, yielding scalars or lower-rank tensors depending on the pairing structure.

10 Common pitfalls and notation

10.1 Distinguishing tensor product from Kronecker product

Although matrices and coordinate arrays often use Kronecker products to represent tensor products numerically, the two notions are not identical conceptually. The Kronecker product is a specific coordinate-based operation, whereas the tensor product is an intrinsic construction defined by universal properties and module relations. Confusing them can lead to incorrect assumptions about invariance under basis changes.

10.2 Order of factors

In general, \(V\otimes W\) is not literally the same object as \(W\otimes V\), though there is a natural isomorphism \(V\otimes W \cong W\otimes V\). When tracking signs or symmetries—especially in exterior powers—treating the order casually can produce errors.

10.3 Basis dependence and coordinate-free notation

Components of a tensor depend on chosen bases, but the tensor itself does not. A common mistake is to treat component identities as if they are basis-independent statements. Coordinate-free formulations using universal properties, naturality, and functoriality help prevent confusion when changing coordinates.