Motivation
A bilinear map takes two inputs but is linear separately in each. It cannot usually be treated as a linear map on the Cartesian product: multiplying both inputs by a scalar would introduce two copies of that scalar. The tensor product replaces the pair of inputs by a new module in which bilinear expressions become ordinary linear maps. Its definition is a mapping property, not a promise that every element is one elementary tensor.
This construction also tests how exact sequences interact with familiar operations. Mapping out of a module preserves kernels but may fail to lift maps across a surjection. Tensoring preserves quotients but may destroy an injection. Proving those statements precisely explains why projective modules and, later, flat modules arise.
For this chapter is a commutative ring with identity, and all modules are unital -modules. Earlier chapters allowed noncommutative rings; here commutativity permits both inputs and the Hom space to carry compatible left actions. A noncommutative tensor product requires a right module in one slot and separate side conventions. No such extension is implicit below.
Definitions
Definition 6.1
(Bilinear maps and the universal tensor product)
A map is -bilinear when each map obtained by fixing one argument is -linear. A tensor product consists of a module and a bilinear map such that every bilinear factors through a unique linear map with .
The factorization quantifies over every target module . To establish that a concrete module is a tensor product, one must produce the bilinear map and prove existence and uniqueness of every factorization. Dimension counting or suggestive notation alone is insufficient. An elementary tensor is one ; general elements are finite sums of elementary tensors, which need not themselves be elementary.
Definition 6.2
(The Hom module)
For modules , the set is an -module under pointwise addition and . Commutativity ensures the new function remains linear: . For a map , postcomposition and precomposition on the appropriate Hom spaces will be written explicitly so that the direction is visible.
Theorems
Theorem 6.3
(Construction, uniqueness, and tensor–Hom correspondence)
Tensor products exist and are unique up to a unique isomorphism preserving elementary tensors. For all modules over the commutative ring , there is a natural linear isomorphism
sending to the map .
Theorem 6.4
(Hom is left exact; tensor is right exact)
For a short exact sequence and modules , the sequences
and
are exact at all positions asserted by their displayed ends. Neither a final surjection in the first sequence nor an initial injection in the second is guaranteed in general.
Proof sketch or proof idea
Constructing the object from formal generators
Proof
(Construction and uniqueness of the tensor product)
Take the free -module with a basis symbol for every pair in . Let be the submodule generated by the relations
Define , and let be the class of . The relations make the proposed map bilinear. Given , the free-module property gives a unique linear map taking to . Bilinearity makes it kill every displayed relation, hence all of , so it descends to the quotient. Elementary tensors generate that quotient, so the descended map is unique. This proves existence with the claimed universal property.
If two objects have that property, apply each property to the other's bilinear map. This gives maps in both directions preserving elementary tensors. Their composites are identity because the identity map and the composite have the same values on all elementary tensors, and uniqueness forces equality. The isomorphism is canonical in this specified sense, rather than dependent on a chosen basis.
Bilinearity also gives the balancing identity , since both equal . For linear maps and , the bilinear rule induces a linear map . Identities and composition can be checked on elementary tensors because those generate. This justifies the maps in the exactness theorem.
Currying proves the Hom correspondence
Proof
(Currying and naturality)
A linear map gives the displayed map into Hom: its inner function is linear in , and the outer map is linear in . Conversely, a linear map gives . Evaluation is bilinear by the two linearity properties, so universality produces a unique . The constructions reverse each other on every elementary tensor, hence on the entire tensor product. They preserve addition and scalars pointwise.
Naturality here means that composing any input or output map before or after these constructions gives the same result. This follows by evaluating the two routes at ; both produce the same composed value. Thus the correspondence respects maps, not merely the sizes of two abstract Hom sets. It converts a question about linear maps out of a tensor product into a question about linear maps with values in another module of maps.
Left exactness is factorization through a kernel
Proof
(Left exactness of Hom)
If and is injective, then , proving the first Hom map injective. A map lies in the next kernel precisely when . Its image is therefore contained in . For each , injectivity gives a unique with . Uniqueness and linearity of prove that is linear. Thus , proving exactness at the middle Hom space. Every such composite is in the kernel because , so both containments are established.
Surjectivity at the final Hom space would require every to lift through . This is exactly the projective lifting requirement when demanded for every surjection. Earlier notes proved that free modules and their direct summands are projective. The exactness statement therefore identifies what additional hypothesis repairs the missing final surjection.
Right exactness is preservation of a quotient
Proof
(Right exactness of tensor)
First prove a useful general quotient identity. If is a submodule, let be the submodule of generated by all with . The quotient map induces a surjection
because every elementary tensor lifts, and finite sums lift term by term. To identify its kernel, define . Changing by changes the result by , so this is well defined and bilinear. It induces a map from back to . The forward map kills and descends in the other direction. Their composites fix every elementary tensor class, hence are identities. Thus the kernel is exactly .
For a short exact sequence, identify with and with . The image of is precisely the submodule generated by , since elements of are finite sums of elementary tensors. The quotient identity therefore says that this image is the next kernel and that is onto. No step asserts that is injective; the proof obtains exactly the right-end exactness claimed.
Proof X-ray
Universal maps need two inverse constructions
To identify the tensor of a quotient, lifting elementary tensors proves surjectivity but does not identify the kernel. The reverse bilinear map into the proposed quotient supplies that missing inclusion. Checking both composites on elementary tensors then proves the entire isomorphism. The universal property organizes these checks; it does not make well-definedness automatic.
Worked examples
Worked example 6.5
(Tensoring a quotient makes its relations act on the other factor)
For an ideal ,
where consists of finite sums of with . The formula is well defined and bilinear because changing by an element of changes the result by an element of . The inverse sends to . It kills by balancing, and the two composites are identity, again using balancing to move across the tensor sign.
In particular, for positive and : the quotient is . Thus the tensor of and is , whereas that of and is zero. Tensoring two nonzero modules need not give a nonzero module.
Concept lensStructural
Right exactness read off from the Hom correspondence
The tensor–Hom correspondence gives a second proof of right exactness. Fix and put . Contravariant left exactness, proved in an exercise below, makes exact, and by naturality this is the sequence
Taking to be the cokernel of shows that this cokernel is zero; taking to be the cokernel of shows that the kernel of lies in the image of .
Worked example 6.6
(Tensoring can destroy an injection)
Start with the short exact sequence and tensor with . The natural identification sends to . Under it, the first map is multiplication by two on , hence the zero map on a nonzero module. It is not injective. The rest becomes , which is exact at the middle and final terms, as the theorem predicts. The vanished injection is a property of the tensor operation, not an error in the original exact sequence.
Worked example 6.7
(Hom can fail to lift across a surjection)
Use the same integer exact sequence and let . Every map is zero because its image is killed by two and the integers have no nonzero such element. But contains the identity. Thus the induced map from to is not onto. The identity has no lift, which also proves that is not projective. This is the same relation obstruction that prevented the original sequence from splitting.
What the operations remember
Even over a field, a tensor need not be elementary. In , consider . The bilinear map induces a linear map to two-by-two matrices, sending this tensor to the identity. An elementary tensor maps to an outer-product matrix of rank at most one, whereas the identity has rank two. Thus this tensor cannot be elementary. The universal property provides the distinguishing map, so the conclusion is proved without assuming a coordinate model of the whole tensor product.
The quotient calculation shows that relations in one factor become relations acting on the other. A module killed by two tensors to zero with a module on which multiplication by two is invertible, since every elementary tensor can then move that factor of two across and vanish. This is an algebraic mechanism, not a dimensional analogy with vector spaces.
Definition 6.8
(Flat module)
A module is called flat when tensoring with it preserves injections, and consequently short exact sequences by the right-exactness theorem.
The example proves that is not flat over . Free modules are flat: tensoring with a finite or infinite direct sum of copies of produces the corresponding direct sum of copies of the input, and direct sums preserve injections coordinatewise. The direct-sum tensor identity can be proved by the universal property, as an exercise below requests. No general classification of flat modules is asserted here.
Finite projectivity supplies a dual basis
Write for the dual module. For finitely generated projective modules the familiar dual-space formula survives.
Proposition 6.9
(Hom out of a finitely generated projective module)
If is finitely generated and projective, then for every module the rule defines an isomorphism .
Proof
(A finite dual basis)
Choose a finite free surjection and split it by projectivity. Write its projection as and section as . Let and let be the th coordinate of . Then
This finite dual-basis identity gives inverse maps between and . The forward map sends to ; the reverse sends to . Their first composite fixes by the displayed identity. For the other, applying any functional to that identity gives ; balancing then shows that is fixed. These elementary tensors generate, proving the isomorphism. Finite generation makes the sum finite, while projectivity supplied the split free presentation. Neither hypothesis was required for the earlier tensor–Hom correspondence, which has a different source and target.
Counterexample mode
A dual-vector-space formula needs an extra hypothesis
For finite-dimensional vector spaces, can be represented by . Extending that assertion to arbitrary modules is false. Take over . Then , so , but . Finite projectivity repairs this formula, while the tensor–Hom correspondence proved above remains valid without that assumption. These are different isomorphisms with different hypotheses.
Common mistakes
Common mistake
(Balancing moves scalars but does not divide by them)
The equality permits a scalar to move between factors; it does not permit division by a nonunit. A zero tensor does not force one input to be zero: in , the tensor vanishes although neither factor does. And a nonzero tensor product need not have every element elementary.
Common mistake
(A formula on elementary tensors is not yet a map)
Constructing a map on elementary tensors requires checking the bilinear relations before declaring the formula well defined. For example, the rule defines no map when : it would send to , because is not linear in .
Common mistake
(Exactness must be read with the displayed arrows)
Exactness must be read with the arrows and zero terms actually displayed. Hom reverses arrow direction when its first argument varies and preserves it when its second varies. Tensoring the stated short exact sequence preserves its surjective end, while a projective first Hom argument repairs the lifting end. These statements cannot be interchanged by dropping the argument labels.
Summary
The free-module quotient construction represents bilinear maps by linear ones. Currying gives the tensor–Hom correspondence. Hom's kernel factorization proves left exactness, while a two-sided quotient construction proves tensor's right exactness. Explicit torsion modules show the missing injection or surjection, and explain the need for flatness and projectivity.
Exercises
Exercise 6.10
(Prove directly by inverse maps, and prove .)
Foundation
Solution(3)
- 1
Define the two maps
The rule is bilinear, so it induces with . The rule is linear.
- 2
Check both composites
Key step
, and . Elementary tensors generate, so and are inverse.
Because: Balancing gives .
- 3
Compute the integer example
By the quotient example, . Bézout's identity gives , so the tensor product is .
Exercise 6.11
(Calculate for positive , including a proof of its cyclic order.)
Core
Solution(3)
- 1
Reduce to one value
A homomorphism is determined by , and a given defines one exactly when in .
Because: is generated by subject to .
- 2
Solve the condition
Key step
Let , , with . Then is equivalent to , hence to .
- 3
Identify the group
The admissible classes are , a cyclic subgroup of order . Evaluation at is an additive isomorphism onto it, so .
Exercise 6.12
(Prove the contravariant left-exact sequence , with maps given by precomposition with and .)
CoreProof
Solution(3)
- 1
The first map is injective
If , then vanishes on , so .
Because: is surjective.
- 2
The kernel of restriction factors through
Key step
A map satisfies exactly when it vanishes on . Then is well defined on , linear and unique, and .
- 3
The image lies in the kernel
∎Conversely because . This proves exactness at the middle Hom space; no final surjection is asserted.
Exercise 6.13
(Prove , including the finite-support check for arbitrary index sets.)
SynthesisProof
Solution(4)
- 1
Construct the forward map
The rule is bilinear, and the output has finite support because does. Universality gives a linear map into the right-hand direct sum.
- 2
Construct the backward map
Each inclusion induces . Their direct sum is a finite sum for each finitely supported tuple.
- 3
Check both composites on generators
Key step
fixes every elementary tensor placed in summand . Also .
- 4
Conclude
∎These tensors generate the two modules, so and are inverse isomorphisms.